book of proceedings - ntuaeeee2014.epu.ntua.gr/portals/16/book of proceedings.pdf · book of...
TRANSCRIPT
Hellenic Operational Research Society University of Thessaly
3RD INTERNATIONAL SYMPOSIUM & 25TH
NATIONAL CONFERENCE ON OPERATIONAL
RESEARCH ISBN: 978-618-80361-3-0
Book of Proceedings
Volos, 26-28 June 2014 http://eeee2014.epu.ntua.gr
Book of Proceedings
1
ISBN: 978-618-80361-3-0
Book of Proceedings
2
Table of Contents
Multicriteria decision support for financing Greek agricultural units ............................................ 4
Supply Chain Modeling geared to Customer Satisfaction using the simulation s/w package
SIMUL8: A linear and a non-linear model ................................................................................... 11
Supply chain Linear and Non Linear optimization with regard to customer satisfaction: Solving
with GAMS ................................................................................................................................... 19
Internet and agro-tourism sector for regional development in Crete: ........................................... 28
A multicriteria ranking .................................................................................................................. 28
On the way to a reference model for supply chains in the construction industry ......................... 33
The problem of robustness in the MUSA method: Theoretical developments and applications.. 39
A combined MCDA approach for facilitating maritime transportation policies evaluation ........ 48
Optimal Strategic Design of Flexible Supply Chain Networks .................................................... 56
An Integrated Multi-Regional Long-Term Energy Planning Model Incorporating Autonomous
Power Systems .............................................................................................................................. 63
Comparison of GA-ANN and Traditional Box-Jenkins Methods for Railway Passenger Flow
Forecasting .................................................................................................................................... 70
Inspection of power grid by periodic vehicle routing formulation ............................................... 79
Environmental performance evaluation using a fuzzy aggregation-disaggregation approach ..... 83
Rationalizing electricity production investments from renewable energy sources in Greece using
a synergy of multicriteria methods ............................................................................................... 91
Orisma(c): Optimizing long term fleet wide crew assignment ................................................... 103
Simulation analysis of a pilot handling system for the rail transport of conventional semitrailers
..................................................................................................................................................... 109
Research on internet sufficiency of websites concerning women agricultural co-operatives in
Greece: A multicriteria approach ................................................................................................ 116
Adaptation of ITA for project portfolio selection within a group of decision makers ............... 122
F.W. Lanchester’s combat model application in a supply chain in a duopoly ........................... 127
An optimization modeling approach for the establishment of a bike-sharing network using Monte
Carlo Simulation and stochastic demand: a case-study of the city of Athens ............................ 131
Evaluating new service development effectiveness in tourism: An ordinal regression analysis
approach ...................................................................................................................................... 138
New Technologies & Labor Market ........................................................................................... 146
Regression modeling for spectral data sets: A multi-objective genetic approach ...................... 153
Optimal use of non-collaborative servers in two-stage tandem queueing systems .................... 161
Book of Proceedings
3
GreenYourRoute platform .......................................................................................................... 168
Reducing Waiting Time at Intermediate Nodes for Intercity Bus Transportation ...................... 171
Innovation management strategies for organizational performance ........................................... 177
Country risk evaluation methodology to support bilateral cooperation in the field of electricity
generation from renewable sources ............................................................................................ 183
Sustainable Food Security: A System Dynamics Decision-Making Methodology .................... 193
Towards the implementation of optimal train loading plan in the Athens – Thessaloniki freight
services ........................................................................................................................................ 201
An exact method for the inventory routing problem .................................................................. 209
Open Governmental data sources in Europe: A comparative evaluation of semantic and technical
characteristics .............................................................................................................................. 217
A branch and price solution algorithm for the tail assignment problem ..................................... 224
A multi-stage column generation solution approach for the bidline aircrew scheduling problem
..................................................................................................................................................... 231
A Calibration Tool for Macroscopic Traffic Flow Models ........................................................ 236
Air Traffic Management: The free flight concept....................................................................... 247
Comparison of pricing mechanisms in markets with non-convexities ....................................... 254
Development of Optimization Models for Addressing Various Decision and Information Related
Issues in Supply Chain Planning................................................................................................. 260
Measuring employee satisfaction in a Greek academic environment ......................................... 268
Identifying factors of bank service quality during economic crisis in Greece ........................... 275
Touristic Guide: A prototype software for touristic journey planning ....................................... 282
FindMyWay: A prototype web-based platform for journey planning in Athens city, Volos city
and Crete island........................................................................................................................... 286
Book of Proceedings
4
Multicriteria decision support for financing Greek agricultural units
Athanasios Valiakos
University of Piraeus, 80, Karaoli & Dimitriou Street, GR-18534 Piraeus, Greece
Yannis Siskos
University of Piraeus, 80, Karaoli & Dimitriou Street, GR-18534 Piraeus, Greece
Abstract
The European Agricultural Guarantee Fund (EAGF) finances direct payments to farmers in
Member States of European Union with specific implementation rules. This is done based on the
entitlements, which derived from the total production during the historical reference years.
European Union rendered this decoupled financial aid as a Single Direct Payment (SDP) scheme,
and the total production in Greece is significantly decreased. In view of Common Agricultural
Policy's reform, the evaluation of agricultural units is proposed using robust ordinal regression
(ROR) approach. In this paper, a case study of farmers in the industry of the juicing citrus is
conducted. A method is proposed as an evaluation tool for financially aid to the farmers, towards
the new policy, granting the production based approach more effective and more objectively
allocating the direct payments. An additive evaluation model is proposed based on a consistent
family of criteria composed be DEA’s input and output criteria. The phenomenon of “Sofa
Farmers” could be eliminated, since farmers would be financially aided after been evaluated. The
preference information used in UTASTAR method is given in the form of a partial pre-order on a
subset of farmers (reference set). In order to obtain robust conclusions, post-optimality analyses
are applied by computing complementary robustness measures as well as a goal programming type
regression model.
Keywords: Multicriteria decision analysis; Common Agricultural Policy; Ordinal regression;
Robustness.
Book of Proceedings
5
1. Introduction
The Common Agricultural Policy (CAP) is the agricultural policy of the European Union (EU),
which implements a system of agricultural subsidies and other programs (European Union, 2009).
In this paper, an evaluation tool is proposed for rural farms to calculate total production subsidy
taking into account other criteria, and not only the total area (European, Commission, 2011). This
way the «sofa farmers' phenomenon» (European, Commission, 2010) can be eliminated, in light
of the CAP's reform (European, Commission, 2013), since farmers would be financially aided after
been evaluated.
An evaluation methodology is therefore proposed for estimating the financial aid, as form of direct
payment. Furthermore, a multi-criteria additive value model, namely UTAStar, is constructed with
a consistent family of criteria composed be DEA’s input and output criteria. In addition, a linear
regression model is proposed to estimate the direct payment, based on the global utility value.
The rest of the paper unfolds as follows. In section 2, UTAStar is reformulated based on DEA’s
input and output criteria. Post-optimality analyses are applied by computing complementary
robustness measures and the estimation of direct payment is calculated as a linear regression based
on the global utility values. In section 3, a case study of agricultural units in the industry of the
juicing citrus is presented using the proposed methodology. The last section concludes this
approach.
2. Methodology UTAStar and Linear Regression
The methodology described in this paper is a robust ordinal regression approach (Greco S.,
Słowiński R., Figueira J., Mousseau V., 2010) and consists of two models to calculate financial
aid; the additive value function UTAStar (Siskos Y., Yannacopoulos D., 1985) and the least
squared method. The synergy of these models leads to the evaluated direct payment of each
agricultural unit. The model of this methodology is the multi-criteria additive value model
(Jacquet-Lagrèze E., Siskos J., 1982), which is based on a consistent family of criteria composed
by DEA’s input and output criteria (Charnes A., Cooper W.W., Rhodes E., 1978). Consider a finite
set of agricultural units 𝐴 = {𝑎1, 𝑎2, … , 𝑎𝑧} evaluated by criteria from a consistent family. The
criteria are divided as input and output oriented, following (Valiakos A., Siskos Y., 2013) work.
Therefore, the family of criteria is the vector 𝐺 = {𝑔𝑖𝐼 , 𝑔𝑟
𝑂}, 𝑖 = 1, . . . , 𝑚 ; 𝑟 = 1, … , 𝑠.
We select 𝐴𝑅: {𝑎1, 𝑎2, … , 𝑎𝑛} ⊂ 𝐴 a reference set of n agricultural units. UTAStar is applied on the
selected reference set (Jacquet-Lagrèze, E., Siskos, J., 2001). A two-stepped post-optimality
analysis is applied. In order to reduce in size the polyhedron, we can maximize the distance δ, of
two consecutive agricultural units, by solving the following LP,
max𝐷 = 𝛿 (1 )
under the constraints,
𝐴𝑙𝑙 𝑈𝑇𝐴𝑆𝑡𝑎𝑟 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑠 (2 )
Book of Proceedings
6
The second step is to check whether the model is significantly robust. The sensitivity analysis is
achieved by investigating the extreme maximum and minimum value functions of each criterion.
Post-optimality analysis is applied in order to evaluate the robustness of the additive model and to
calculate the most representative additive value function. The final solution of the model is the
average (barycentral) of the extreme maximum values forming the representative utility value. The
solution is extrapolated to the complete set of A, acquiring the global utility values 𝑢[𝑔(𝑎𝑖)], 𝑖 =
1, … , 𝑧. This utility value is used to assess the amount of subsidy of the agricultural units a
regression type method is applied, using the least squares method. Let 𝑌 = {𝑦1, … , 𝑦𝑧} be the
values of traditional direct payments. The estimation of linear regression, using least squared
method for (𝑢[𝑔(𝑎1)], 𝑦1),… , (𝑢[𝑔(𝑎𝑧)], 𝑦𝑧) is expressed by the following equation,
𝑦�̂� = �̂� + �̂�𝑢[𝑔(𝑎𝑖)], 𝑖 = 1,… , 𝑛 ( 3 )
where �̂�, �̂� are the estimators of the constants α, β. The estimator 𝑦�̂� is the estimation of the direct
payment.
3. Case Study - Juicing citrus
Although decoupled there could still be a 'connection' of economic aid under the current regulation
for the common agricultural policy (CAP) from 2015 to 2020. For that reason, farmers in the
industry of the juicing citrus, and more specifically oranges as selected in this approach. From
2003 until 2008 the direct payment was based on the total production. During the period 2008-
2009 the new scheme was implemented for the coupled financial aid of citrus. In Greece a 60% of
the single area payment scheme in citrus was connected with the production processing. Pursuit of
the European and Greek juicing industry was to maintain the link with the aid of citrus processing.
However, European Union rendered this decoupled in the year 2010, and since then the total
production is significantly decreased. A case study of 1,789 farmers, who were active in Argolida
region, is conducted from the total number of farmers in Greece, which are approximately 8,000
farmers. The case study involves five criteria, three input and two output, which are presented in
Table 1,
Criteria Name Measurement Min
Valu
e
Max
Value Description
Input Criteria
𝑔1𝐼 Labor Cost euro (€) 100 600,000
Number of hours of operator, family, and
hired farm labor – e.g. tilth, carving.
𝑔2𝐼
Production
Cost euro (€) 20 900,000
Expenses on fertilizers, pesticides, sprayers
other chemicals, plant values and others.
𝑔3𝐼 Capital euro (€) 50 300,000
Agricultural equipment, machinery and
buildings. Output Criteria
Book of Proceedings
7
𝑔1𝑂 Trees units (#) 10 3,500 Total number of trees operated.
𝑔2𝑂
Net
Production kilos (kg) 1,000 150,000 Weight of the final product, once processed.
Table 1: Evaluation criteria of agricultural units.
For the selected case scenario 10 farmers are chosen to form the reference set. Much effort was
spent to construct this set in order to be representative of the whole entity of farming categories
and to avoid being very complicated for the DM to rank. The constructed reference set, its
evaluation by the DM and the confirmation of this evaluation by the UTASTAR method are
presented in Table 2,
DM's
Ranking 𝒈𝟏𝑰 𝒈𝟐
𝑰 𝒈𝟑𝑰 𝒈𝟏
𝜪 𝒈𝟐𝜪
Global Value
u(g)
1 2,225.10 3,434.33 2,422.45 886 20,606.00 0.565692
2 5,912.80 1,106.07 29,737.13 460 33,182.00 0.543425
3 2,599.50 4,160.53 2,487.45 650 15,602.00 0.484212
4 774.50 7,115.00 33,984.86 340 21,345.00 0.448251
5 1,646.00 1,265.33 15,832.92 310 18,280.00 0.439279
6 3,402.00 3,832.00 13,132.32 290 19,160.00 0.431380
7 8,315.50 2,456.33 2,859.39 600 7,369.00 0.427204
8 1,867.50 22,715.67 16,565.55 350 21,470.00 0.391379
9 224.80 182.70 4,871.40 35 1,609.00 0.323105
10 21,723.00 1,681.07 8,367.06 150 2,554.00 0.296651
Table 2: Evaluation Criteria of agricultural units.
The model, which is finally adequately stable and sufficiently consistent with the preferences of
DM, is expressed in Equation 4,
u(g) = 0.07742u1(g1I ) + 0.13778u2(g2
I ) + 0.07739u3(g3I ) + 0.31151u4(g1
O) + 0.39596u5(g2O) ( 4 )
Least squared method is applied. The 𝑌𝑖 values are the traditional direct payment the agricultural
units received with the entitlements. The linear regression line (Eq. 3) and the estimators α, β, for
the set of agricultural units – extrapolation to the set A, is displayed in Figure 1,
Book of Proceedings
8
Figure 1: Estimation - Linear Regression line for Juicing Citrus.
In Figure 1, we can observe that most of units are assembled in high utility values. In Figure 2, a
comparison is made between the traditional direct payment and the final direct payment from the
methodology. For presentation purposes, only the reference set is included. In this case, some of
the units are over-financed while others are under-financed.
Figure 2: Comparison Traditional/Final Direct Payment of Reference set of Units.
It is worth mentioning that from the estimation, using the model of regression UTAStar the total
financial aid is 5,684,425.47 €, while with the EU funded this scheme with the amount of
5,684,425.32 €. Through the evaluation of the agricultural units not only we can achieve
rationalizing of the direct payment, but also the overall total direct payment remains the same.
Book of Proceedings
9
4. Conclusions
In this paper, a framework is proposed to calculate the direct payment, based on evaluation of
agricultural units using robust ordinal regression (ROR) approach. The synergy of two methods,
the additive evaluation model and the goal programming regression model is proposed to measure
the final financial aid. The additive evaluation model is proposed based on a consistent family of
criteria composed be DEA’s input and output criteria. The phenomenon of “Sofa Farmers” could
be eliminated, since farmers would be financially aided after been evaluated. The robustness of
the two phase methodology is controlled by post-optimality analyses by computing
complementary measures. The financial aid, as form of direct payment would be provided after
evaluation. The decoupled direct payment is still coupled, for evaluation purposes and the EU
budget can be controlled. For the direct payment of the scheme "juicing of citrus" the total EU
budget remains the same, by financing with the proposed methodology.
Acknowledgment
This study is funded and supported by the Institute of National Funds of Greece, since the first
author is under financial scholarship.
Data obtained from N. Samaras, Greek Payment and Control Agency for Community Aid
Guidance and Guarantee Fund, December 01, 2010.
References
European, Commission. “The CAP towards 2020: Meeting the food, natural resources and
territorial challenges of the future”, Official Journal of the European Union, 2010, Brussels.
European, Commission. “Proposal for a regulation of the European Parliament and of the Council:
Establishing rules for direct payments to farmers under support schemes within the framework of
the common agricultural policy”, Official Journal of the European Union, 2011, Brussels.
European, Commission. “Overview of the CAP reform 2014-2020”, Official Journal of the
European Union, 2013, Brussels.
European Union. “Council Regulation (EC) No 1782/2003 - Establishing common rules for direct
support schemes under the common agricultural policy and establishing certain support schemes
for farmers and amending Regulations”, Official Journal of the European Union, 2009, Brussels.
Greco S., Słowiński R., Figueira J., and Mousseau V. “Robust ordinal regression”, Trends in
multiple criteria decision analysis, Springer, 2010, pp. 241-283.
Jacquet-Lagrèze E., and Siskos J. “Assessing a set of additive utility functions for multicriteria
decision making”. European Journal of Operational Research, Vol. 10, Issue 2, 1982, pp. 151-
164.
Jacquet-Lagrèze, E., and Siskos, J. “Preference disaggregation: 20 years of MCDA experience”.
European Journal of Operational Research, Vol. 130, Issue 2, 2001, pp. 233-245.
Siskos Y., and Yannacopoulos D. “UTASTAR: An ordinal regression method for building additive
value functions”. Investigação Operacional, Vol. 5, Issue 1, 1985, pp. 39–53.
Book of Proceedings
10
Valiakos A., and Siskos Y. “From Data Envelopment Analysis to Multi-criteria Decision Support:
Application to Agricultural Units Evaluation in Greece”. 2nd International Symposium and 24th
National Conference on Operational Research, ISBN: 978-618-80361-1, Athens: Hellenic
Operational Research Society (HELORS), 2013, pp. 224-230.
Book of Proceedings
11
Supply Chain Modeling geared to Customer Satisfaction using the simulation
s/w package SIMUL8: A linear and a non-linear model
Papadopoulos T. Chrissoleon*
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Gavriel Eleni,
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Chrysochoidis-Trantas Panagiotis
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Kalotychos Thomas
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Bibos Aggelos
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
E-mail address of the corresponding author: [email protected]
Abstract
The aim of this paper is the simulation of a general multi-criteria supply chain model. The main
objective is to optimize the supply chain geared to customers’ satisfaction and then the
optimization of profit and costs. Specifically, the model is set up for a four echelon supply chain
with two objectives which are: to guarantee customers’ satisfaction and the optimization of the
holding cost, the transportation cost and the 3PL cost at central warehouses and therefore the
maximization of profit. For the purpose of this research, a linear and a non-linear simulation model
were developed. The customer’s demand is served immediately from the retailers or in a short time
from the warehouses. We present two scenarios in order to achieve this target. In the first one, the
60% of the demand is satisfied immediately from the retailers and the 40% from the warehouses.
In the second one, the 30% of the demand is satisfied immediately from the retailers and the other
70% from the warehouses.
Book of Proceedings
12
Keywords: Customer satisfaction, simulation of supply chain, simul8.
1. Introduction and literature review
Supply chain management and distribution networks have been designed by many researchers
during recent years. Most of the papers were typically oriented at cost minimization and profit
maximization. The satisfaction of customers’ demand is the main reason to increase the service
level of the supply chain. As a result, lately, many research studies try to optimize the supply chain
operations with regard to demand satisfaction.
The literature in the area of supply chain simulation is large. Here, due to space limitations, only
a few studies are mentioned. The interested reader is addressed to the master’s thesis of Gavriel
(2014) for more references and a systematic classification of the relevant material. Ganeshan
(1999) tried to find near-optimal stocking policies (reorder point, order quantity) on both the
retailer and the distribution center level, and lead time conditions, in order to satisfy the demand.
Lin et al. (2000) developed an extended-enterprise supply chain analysis tool due to the willingness
of IBM to reengineer its global supply chain in order to achieve quick responsiveness to customers
with minimal inventory.
Huq et al. (2006) used a mathematical and a simulation model to demonstrate that under specific
circumstances, an inventory replenishment system with two warehouses and n-retailers provide
better customer service without significant changes in the cost than one warehouse.
Lim et al. (2006) studied a production-distribution plan taking into account a multi-facility, multi-
product, and multi-period problem in order to determine the optimal production-distribution plan
in a network with a bill of material (BOM).
The main purpose of this paper is to create a linear and a non-linear simulation model of a general
multi-criteria supply chain which serve the demand and satisfy customers. The difference in our
study is that we attempt to guarantee that all customers will get the quantity they ordered in the
time promised and then we examine the optimization of the financial metrics. To solve the model
we use the simulation software simul8, the educational version. The rest of the paper is structured
as follows: In section 2, the general structure of the models and their main characteristics are
presented, in section 3 the final results of the two scenarios for the linear model are given, and the
final results of the two scenarios for the non-linear model are presented in section 4. In section 5,
a comparison between a system with one product and a system with two products is given. Finally,
the last section concludes the paper and gives a few areas for further research.
2. The Structure of the models and their main characteristics
We studied a supply chain consisting of 4 echelons: One supplier, one manufacturer, two
warehouses and four retailers. The supplier of raw material is focused only on the execution and
the delivery of the orders that receives from the manufacturer and not on the procedure of
production of raw materials. We assume that we have only two final products. The first one
consists of three items of raw material. The second one needs four items of raw material to be
produced. The manufacturer is responsible for the inventory replenishment of the warehouses and
of the retailers, through the warehouses. Moreover, he has a production department in order to
produce the products that are necessary for his inventory replenishment. Each warehouse serves
two retailers. The lead time between echelons is known and constant. We set two goals to achieve:
Book of Proceedings
13
the customers’ satisfaction and the optimization of profit and cost. In this study, we examined the
inventory and the replenishment policies (reorder point, order quantity) in each member of the
supply chain in order to minimize the holding cost, the transportation cost and the 3PL1cost at
warehouses and therefore to maximize the profit. Also, in the non-linear model the objective is to
calculate the cost of the defective products in the supply chain and how the demand fill rate is
affected.
We designed our model in the simulation software simul8, which runs for 90 days2. We suppose
that the daily demand for the first product follows the normal distribution and the demand of the
second product follows the poisson distribution. The four parameters for the poisson distribution
are (the number represents the average): 200, 300, 400, and 150. The four pairs of parameters for
the normal distribution are (the first number is the average and the second is the standard
deviation): (500,50), (600,100), (800,150), (1000,200). The values of distributions were randomly
selected based on historical data. In the real business world the demand is satisfied 100% from the
retailers or 100% from the distributors. So, we decided to divide this percentage between the
retailers and the distributors in order to see how the supply chain is affected. The two scenarios
are the following: Scenario1: 60%-40%- the 60% of the demand is satisfied immediately from the
retailers and the 40% from the warehouses in the next period. Scenario 2: 30%-70% - the 30% of
the demand is satisfied immediately from the retailers and the rest 70% from the warehouses in
the next period. Some businesses, not be able to satisfy immediately customers' demand, sell their
products in a discount price. Thus, we assumed that the quantity of products that are delivered
from the warehouses directly to customers have 10% discount on the final price. For each scenario
we start with a large amount of inventory and big values in the replenishment process (Q, R) for
all the echelons in order to be sure that the demand will be satisfied and then we end up to the
minimum quantity of inventory and the smaller values for the quantity of order (Q) and the reorder
point (R) in the replenishment process for each echelon which is necessary in satisfying the
customers’ demand.
In the non-linear model the additional element which describes the non-linearity is that while
transporting orders from warehouses directly to customers there are some defective products,
which are replenished by retailers in the next period of time. The objective consists of one more
value to minimize: the cost of the defective products in the supply chain and then it is examined
how the demand fill rate is affected. The percentage of the destroyed items is determined by a
uniform distribution with a range of values between (0.1, 0.2) every time that an order is delivered.
The mathematical function which is used for the calculation of the cost of the defective products
is equal to the percentage of the destroyed products in the power of two multiplied by a penalty
cost (constant value).
13PL cost is the price that is charged from the distributor (who acts as a 3PL provider) for every inbound and outbound unit of product 2 The simulation model of this paper has the same structure with the GAMS implementation, in the paper “Supply chain Linear and Non Linear optimization with regard to customer satisfaction: Solving with GAMS” (No.: QP-78-06), 3rd International Symposium and 25th National Conference on Operational Research, 2014. Simulation model runs for 90 days such as the GAMS' implementation. Thus, the simulation' results can be compared with the ones of the GAMS implementation.
Book of Proceedings
14
∑ (Destroyed_Itemspwct)2pwct
× σ
P,W,C,T
p,w,c.t=1
Assumptions:
There isn't any constraint about vehicles' capacity.
There isn't any constraint about the maximum limit in the inventory that any echelon can
store.
Lead-Time from the manufacturer to the warehouses is three working days, from the
warehouses to the retailers and from the warehouses to the customers is one working day.
3. Linear model
Comparison between the final results of scenario1: 60%-40% and scenario 2: 30%-70%. Through
the comparison we can see that in the first scenario the holding cost and the 3PL cost at warehouses
is smaller than in the second one. Also, the transportation cost is bigger from the warehouses to
the retailers but smaller from the warehouses directly to the customers.
Figure1: Comparison of the holding cost Figure 2: Comparison of the 3PL cost between the
two scenarios between the two scenarios
-
50,000.00
100,000.00
150,000.00
R1
R2
R3
R4
W1
W2
F_P
RO
DU
…
F_R
AW
…
Scenario 1:60%-40%
Scenario 2:30%-70%
Holding Cost
-
5,000.00
10,000.00
15,000.00
20,000.00
25,000.00
30,000.00
W1 W2
Scenario 1:60%-40%
Scenario 2:30%-70%
3PL Cost
Book of Proceedings
15
Figure 3: Comparison of the transportation cost between the two scenarios
4. Non-linear model
Comparison between the final results of scenario1: 60%-40% and scenario 2: 30%-70%. Through
the comparison of the two different scenarios we can see that the non-linear model reacts like the
linear model. The difference is due to the cost of the defective items which is 29% of the total cost
of the supply chain in scenario 1 and 59% in scenario 2. This happened due to the shift of the
demand from the retailers to the warehouses. The average of customers’ satisfaction who received
a non-complete order is 83% in both scenarios.
Figure 4: Comparison of the holding Figure 5: Comparison of the 3PL cost
cost between the two scenarios between the two scenarios
-
50,000.00
100,000.00
150,000.00
200,000.00
250,000.00
300,000.00
Scenario 1:60%-40%
Scenario 2:30%-70%
Transportation Cost
3PL Cost
Book of Proceedings
16
Figure 6: Comparison of the transportation cost Figure 7: Fill Rate of product P1
between the scenarios
5. System with one product Vs. system with two products
In the comparison between the system with one product and the system with two products, the
customers’ satisfaction is served in both systems. Despite the fact that we had two products in the
first system we didn't face any problem during transportation in the system with two products. This
is due to the fact that vehicles transfer the executed orders independently from the kind and the
quantity of the products. There was only a difference between them in the financial metrics. In a
system with N products the operations will become more difficult and a lot of problems may occur
in different parts of the system. This is something that we investigate as a continuity of this
research.
6. Conclusions and further research
In conclusion, we provided a simulation model of a general and easily applicable supply chain in
order to ensure customers’ satisfaction and therefore the optimization of the holding cost, the
transportation cost and the 3PL cost at central warehouses and the maximization of profit. Two
types of supply chain models were simulated: a linear and a non-linear model. The customer’s
demand is served immediately from the retailers or in a short time from the warehouses. We
presented two scenarios in order to achieve this target. The additional element which describes the
non-linearity is that while transporting orders from warehouses directly to customers there are
some damages in products, which are replenished by retailers in the next period of time. The
objective is to calculate the cost of the defective products in the supply chain and how the demand
fill rate is affected. Through the linear and the non-linear model it is verified that choosing the best
replenishment process (order quantity, reorder point), the correct level of inventory for each
member of the supply chain and the best production rate in the manufacturer the total cost of the
supply chain is decreasing despite the fact which scenario will be applied. Through the comparison
between the two different scenarios, scenario 1: 60%-40%, retailers kept bigger inventory but the
transportation cost from warehouses directly to customers was decreasing. Also, in the non-linear
-
50,000.00
100,000.00
150,000.00
200,000.00
250,000.00
300,000.00
Scenario 1:60%-40%
Scenario 2:30%-70%
Transportation Cost
0
0.5
1
1 101928374655647382
Product P1
FILL RATE
Book of Proceedings
17
model the percentage of destroyed Items is smaller than in the second scenario. A supply chain
can use one of the two different scenarios that are mentioned above. This decision depends on the
way that the supply chain is willing to satisfy the demand (immediately or after a short time). Also,
it must take into consideration other factors as well, such as the nature of the products. In the
comparison between the system with one product and the system with two products the customers’
satisfaction is served in both systems. There was only a difference between them in the financial
metrics.
This study can be extended by improving the transportation part by adding a specific number of
vehicles that are used from each member of the supply chain and with certain capacity. Also, the
determination of distances between the different members will result to a better calculation of the
transportation cost that may occur. Moreover, including some limitation about the maximum
quantity of inventory that each member can keep will lead to better results. Another part that can
be improved is the production process by checking if some problems occur during the production,
how it can be fixed and what is the effect on customers’ satisfaction. Finally, adding more than
two products the supply chain will be more realistic. This part is currently under investigation. A
model with N products is under development. We are in contact with 3 companies in order to
cooperate and give us real data to evaluate our generic model through a real case study.
Acknowledgement: In this research, Ms. Gavriel Eleni, Mr. Chrysochoidis-Trantas Panagiotis, Mr.
Kalotychos Thomas, and Professor Chrissoleon T. Papadopoulos have received a grant from
THALES (Project: ASPASIA), a project co-financed by the European Union (European Social
Fund – ESF) and Greek national funds through the Operational Program "Education and Lifelong
Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program:
Thales. Investing in knowledge society through the European Social Fund.
References
Ganeshan, B. “Managing supply chain inventories: A multiple retailer, one warehouse, multiple
supplier model”. Int. J. Production Economics. Vol. 59, 1999, pp. 341-354.
Huq, F., Cutright, K., Jones, V., Hensler, A. D. “Simulation study of a two-level warehouse
inventory replenishment system”. International Journal of Physical Distribution & Logistics
Management. Vol. 36, 2006, pp. 51-65.
Lim, J. S., Suk J. J., Kim, S. K., Park W. M. “A simulation approach for production-distribution
planning with consideration given to replenishment policies”. Int. J. Adv. Manuf. Technol. Vol.
27, 2006, pp. 593-603.
Lin, G., Ettl, M., Buckley, S., Bagchi, S., Yao, D. D., Naccarato, L. B., Allan, R., Kim, K., Koenig,
L. “Extended-Enterprise Supply-Chain Management at IBM Personal Systems Group and Other
Divisions”. Interfaces. Vol. 30, 2000, pp. 7-25.
Book of Proceedings
18
Mirzapour Al-e-Hashem, S.M.J., Baboli, A., Sazvar, Z. “A stochastic aggregate production
planning model in a green supply chain: considering flexible lead-times, nonlinear purchase and
shortage cost functions”. European Journal of Operational Research. Vol. 134, 2011, pp. 1-31.
Seferlis, P., and Giannelos, F. G. “A two-layered optimization-based control strategy for multi-
echelon supply chain networks”. Computers and Chemical Engineering. Vol. 28, 2004, pp.799-
809.
Book of Proceedings
19
Supply chain Linear and Non Linear optimization with regard to customer
satisfaction: Solving with GAMS
Chrissoleon T. Papadopoulos*
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Angelos G. Bimpos
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Panagiotis Chrysochoidis-Trantas
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Eleni Gavriel
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
Thomas Klotychos
Department of Economic Sciences, Division of Business Economics, Aristotle University of
Thessaloniki, University Campus, 54124 Thessaloniki, Greece.
E-mail address of the corresponding author: [email protected]
Abstract
The purpose of this study is to develop a general multi-criteria supply chain model. The main target
is the optimization of a supply chain with regard to customer satisfaction. The optimization of the
financial figures such as profit, holding cost, transportation cost etc., are taken into account too.
Specifically, a supply chain of three echelons is examined and the model minimizes the holding,
the transportation as well as the 3PL costs in every echelon, increasing thus the profits while
guaranteeing customer satisfaction. Two mathematical models are developed: A linear and a non-
linear. There are two main alternative scenarios presented as a means of achieving these targets.
The first scenario proposes that the retailers satisfy 60% of the daily customer demand while the
rest 40% is satisfied by the distributors the next day. The second scenario examines 30% immediate
demand coverage from the retailer and 70% demand coverage from the distributor the next day.
There is a price discount in the demand that is served from the distributors. In order to solve the
developed mathematical models, the GAMS software is used.
Key-words: Optimization of Supply Chains, Customer Level-Of-Service and Profit Maximization,
Holding-Cost Minimization, GAMS.
Book of Proceedings
20
1. Introduction and Literature Review
During the past years much research has been devoted to the optimization of supply chains. As
soon as the manufacturing breakthroughs came to an end, the majority of the companies tried to
maximize their profits by optimizing their transportation processes and by minimizing their storage
limits. Nowadays, it is well recognized that the final customer is the most important part in the
supply chain. He is the main reason the manufacturing of a product occurs. As a result, lately,
many research studies try to optimize the supply chain operations with regard to demand
satisfaction.
Supply chain optimization has been extensively researched. Due to space limitation, only a few
articles will be presented herewith, whereas for a more extensive literature review and a systematic
classification of the relevant material, the reader may read the Master’s Theses of Kalotychos
Thomas (2014) and Chrysochoidis-Trantas Panagiotis (2014). Paschalidis et al. (2004) in their
research tried to achieve the minimization of holding cost, under the constraint of the demand
coverage. They analyzed the interplay between the Perturbation Analysis and the Large Deviation
Analysis in their objective function and its appliance in a supply chain of up to two echelons.
Furthermore, Farahani and Elahipanah (2008) believe that the satisfaction of customer’s demand
will lead to the reduction of all the costs of the supply chain. They created and analyzed a three
echelon mathematical model with multiple products, capacity constraints and service times,
adopting the JIT distribution model.
Behin Elahi et al. (2011) optimized the supply chain, taking into consideration a percentage of
defective products, in each echelon of the supply chain. The supply chain consists of four echelons,
and each echelon has more than one elements. Their research focused on two main parts. First,
they tried to reduce the total cost of the supply chain and then they attempted to reduce the number
of defective products.
Mirzapour Al-e-Hashem et al. (2013) studied a stochastic production model in a green supply
chain, taking into consideration flexible lead times, nonlinear markets and holding cost.
The main purpose of this paper is to create an easily applicable, general multi-criteria model which
fully serves the demand and satisfies customers. The distinctive element of this study is that all
customers get the quantity they order at the promised time. Then the financial values are examined.
The aim is to achieve full customer satisfaction by minimizing the holding, the transportation and
the 3PL costs in the supply chain as a whole. In order to get the optimal solution GAMS software
was used (LINDOGLOBAL solver). The rest of the paper is structured as follows: In Section 2,
the two mathematical models are presented whereas in the third Section, a numerical example is
given. The fourth Section provides conclusive remarks as well as a set of suggestions for further
research.
2. The Mathematical Models
A supply chain of 3 echelons was studied: Manufacturer, Distribution Centers and Retailers. More
specifically, the supply chain consists of 1 manufacturer, 2 distribution centers and 4 retailers. The
Supplier echelon is considered just to define some variables. Each distributor is responsible for
serving exclusively two retailers. There are two different products. Three goals were set: Customer
satisfaction, profit optimization and cost minimization. The inventory management in order to
Book of Proceedings
21
minimize holding, transportation and 3PL costs was studied. Moreover, as a result of reducing the
above mentioned costs, the profit is optimized. The objective is the holding, transportation and
3PL costs minimization along the supply chain. The constraints are sub-divided into six categories:
The first one includes the constraints I to VII and these are the constraints which guarantee that
there will always be enough inventory at any echelon to fulfill the demand. The second category
includes constraints from VIII to XI and refers to the maximum limit in the inventory that any
echelon can store at any time period. Constraints XII to XV (third category) set the initial inventory
at every echelon at the beginning of the examination. The fourth category of constraints (from XVI
to XIX) are the equations which calculate at every time period the new available inventory. The
fifth category of constraints (from XX to XXV) reveal the inventory replenishment policy (the (Q,
R) inventory replenishment policy is used). Finally, the last constraint (XXVI) is the non-
negativity constraint of the variables.
Objective3
Minimization [RM HC at manufacturer + FP HC at manufacturer + TC at manufacturer + FP HC
at distributor + TC at distributor + LC at distributor + FP HC at retailer + TC at retailer]4
And in math terms
𝑀𝑖𝑛 ∑ ℎ𝑎𝑄𝑎𝑡 + ∑ ℎ𝑖𝑚𝑄𝑖𝑚𝑡
𝐼,𝑀,𝑇
𝑖,𝑚,𝑡=1
𝐴,𝑇
𝑎,𝑡=1
+ ∑ 𝑄𝑎𝑠𝑚𝑡𝑇𝑅𝐶𝑠𝑚
𝐴,𝑆,𝑀,𝑇
𝑎,𝑠,𝑚,𝑡=1
+ ∑ ℎ𝑖𝑤𝑄𝐴𝐿𝐿𝑖𝑤𝑡
𝐼,𝑊,𝑇
𝑖,𝑤,𝑡=1
+ ∑ (𝑄𝑖𝑚𝑤𝑡 +𝑄𝑅𝐸𝑇𝑖𝑚𝑤𝑡)𝑇𝑅𝐶𝑚𝑤
𝐼,𝑀,𝑊,𝑇
𝑖,𝑚,𝑤,𝑡=1
+ ∑ 𝑃𝑖𝑚𝑤(𝑄𝑖𝑚𝑤𝑡
𝐼,𝑀,𝑊,𝑅,𝐶,𝑇
𝑖,𝑚,𝑤,𝑟,𝑐𝑡=1
+ 𝑄𝑅𝐸𝑇𝑖𝑚𝑤𝑡)
+ 𝑃𝑖𝑤𝑟𝑄𝑖𝑤𝑟𝑡 + 𝑃𝑖𝑤𝑟(1 − 𝑎)𝐷𝑖𝑟𝑐𝑡 + ∑ ℎ𝑖𝑟𝑄𝑖𝑟𝑡
𝐼,𝑅,𝑇
𝑖,𝑟,𝑡=1
+ ∑ 𝑄𝑖𝑤𝑟𝑡𝑇𝑅𝐶𝑤𝑟
𝐼,𝑊,𝑅,𝑇
𝑖,𝑤,𝑟,𝑡=1
I. Qimwt =Must be equal to what distributor wants based on the final customer demand
II. QRETimwt =Must be equal to what distributor wants based on the retailers′ demand
III. Qiwrt = Must be equal to what retailer wants based on the final customer demand
IV. Qirt ≥ aDirct V. Qat ≥ Pri,mQit
VI. Qimt ≥ ∑ Qimwt−3 + QRETimwt−3 I,M,W,Ti,m,w,t=1
VII. Qiwt ≥ ∑ QiwrtI,W,R,Ti,w,r,t=1 + (1-a) Dirct
VIII. Qimt ≤ MaxIim
IX. Qat ≤ MaxIam
X. Qirt ≤ MaxIir
3 The description of the symbols used are given in the Appendix. 4 RM=Raw Materials, HC= Holding Cost, FP= Finished Product, TC= Transportation Cost, LC= Logistics Cost
Book of Proceedings
22
XI. QALLiwt ≤ MaxIiw
XII. Qirt = INir (t=0)
XIII. Qiwt = INiw (t=0)
XIV. Qimt = INim (t=0)
XV. Qat = INam (t=0)
XVI. Qirt = Qiwrt−1 + Qirt−1 − aDirct XVII. Qat = Qat−1 + Qasmt−1 − Qit−1Prim
XVIII. Qimt = Qimt−1 + Qit−1 − ∑ Qimwt−3 + I,M,W,Ti,m,w,t=1 QRETimwt−3
XIX. QALLiwt = Qimwt−3 + Qiwt−1 −∑ Qiwrt−1I,W,R,Ti,w,r,t=1 +(1-a) Dirct−1
XX. Qiwrt = n ∗ QOirt XXI. Rir ≤ Qirt + ∑ QRETimwt−l
4l=0
XXII. Qiwmt = n ∗ QOiwt XXIII. Riw ≤ Qiwt + ∑ Qimwt−l
3l=0
XXIV. Qasmt = n ∗ QOat XXV. Ram ≤ Qat + Qasmt−1
XXVI. Qimwt, QRETimwt, Qiwt, QALLiwt, Qiwrt, Qimwt, Qimt, Qasmt, Qit, Qat, Qiwrt, Qirt, Qiwrt ≥ 0
In addition, an extra cost in the retailers’ echelon was added, named penalty cost. This cost refers
to the loss occurring in the quantity of products that the customers get from the distributors. Then
the retailers are obliged to replenish the defective products to the end customers. The nonlinear
differentiation-addition is presented below. Also, a new constraint regarding the amount of
defective products is presented:
𝑀𝑖𝑛…+ ∑ (𝑘𝑖𝑟𝑐𝑡)2 ∗ 𝜎
𝐼,𝑅,𝐶,𝑇
𝑖,𝑟,𝑐,𝑡=1
XXVII. kirct = firt ∗ (1 − a) ∗ Dirct
3. Numerical Example
Microsoft Excel (random number generator) was used to input the data regarding customer
demand. More specifically, we assumed that the first product follows the Poisson distribution and
the second product follows the Normal distribution. The eight pair numbers are presented in the
Table below:
Product Retailer 1 Retailer 2 Retailer 3 Retailer 4
1 POISSON
(200)
POISSON
(300)
POISSON
(400)
POISSON
(150)
2 NORMAL
(500,50)
NORMAL
(600,100)
NORMAL
(800,150)
NORMAL
(1000,200)
Distributions regarding generated data
Book of Proceedings
23
Two different scenarios were examined. The first one assumes that the retailers are responsible for
satisfying 60% of the daily demand and the rest is covered by the distributors the next day. In the
second scenario only 30% of the demand is satisfied from the retailers’ inventory and the rest is
covered from the distributors. In the non-linear model, a percentage of products sent from the
distributors to the end customers is defective. The amount of the defective products has to be
replenished the next day from the retailers. (There is a price discount in the demand that is served
from the distributors).
Results: It is not fair to qualify one scenario as the best because the two scenarios represent two
different supply chains and two different product types (e.g., not all the products can be managed
and sold using the 2nd scenario). In Figure 1, the total cost, the revenues and the profits in the two
scenarios are given. Each enterprise would consider its product characteristics, demand pattern and
some other clues before choosing between the two scenarios.
Total Cost,
Revenues
and Profit Comparison between the two Scenarios
4. Conclusion and Further Research
In conclusion, we tried to satisfy the customer demand and based on that to achieve all the other
targets which can be the maximization of profit, the minimization of holding and transportation
cost, etc. We managed to develop a multi-criteria, supply chain model. The supply chain consists
of three main echelons before the final customer: The manufacturer, the distributors and the
retailers. Each one of the three echelons has three objectives to satisfy: maximization of profit and
customer satisfaction as well as minimization of holding, transportation and 3PL costs. The
capacity limitation of the warehouses in each echelon was also taken into consideration in the
linear model. Through the minimization of the above costs, the profit was maximized while the
customer satisfaction was guaranteed. In the linear model of this paper, we tried to research the
consequences in each one of the three criteria mentioned earlier, when only a part of the demand
(60% or 30%t) is satisfied immediately while the rest of the demand is served afterwards in a
predefined time. By comparing the two policies we managed to determine the exact inventory each
echelon needs to hold in order to satisfy the customers completely. The nonlinear model of this
paper has the same, basic structure as the linear model. Our supply chain consists of three echelons,
having the same targets as the linear one. The main difference in this model is that we tried to
0 €
5,000,000 €
10,000,000 €
15,000,000 €
20,000,000 €
25,000,000 €
60% Linear 30% Linear 60% NonLinear
30% NonLinear
Axi
s Ti
tle
Supply Chain
Total cost
Revenues
Profit
Book of Proceedings
24
consider the probability that some of the products transported from the distributors to the end
customers are defective, in a way that the final customer is not able to use them. A non-linear
penalty cost has been added in the objective function of the retailers. Both models were solved
using the GAMS (edition 23.5.1) software package, which is used for solving linear, nonlinear and
mixed integer models. Because of the limitations of the GAMS software used (demo version),
some simplifications were necessary to be made.
This study can be extended by establishing a transportation model, aiming to get the real conditions
of the supply chain even closer. In addition, more products could be added, with another supply
chain to be combined in the echelon of distributors. A sensitivity analysis about how many
unfulfilled orders would not change our result, would be useful. Also it is worthy to extend the
models by analyzing the impact of unreliable suppliers to the customer satisfaction. Furthermore,
a more complex production model with variable production rates could be examined. The
existence of overlapping deliveries from the distributors to the retailers might be a useful addition.
Acknowledgement: In this research, Mr. Chrysochoidis-Trantas Panagiotis, Mr. Kalotychos
Thomas, Ms. Gavriel Eleni and Professor Chrissoleon T. Papadopoulos have received a grant from
THALES (project: ASPASIA), a project co-financed by the European Union (European Social
Fund – ESF) and Greek national funds through the Operational Program "Education and Lifelong
Learning" of the National Strategic Reference Framework (NSRF) - Research Funding Program:
Thales is investing in knowledge society through the European Social Fund.
Appendix
Symbol
Description
Symbol
Description
i Product
{1,…, I} ECi,m
Labor cost per
Employee per Working
Hour for product i at
manufacturer m
s Supplier
{1,…, S} Cim
Machine operation cost
per product i at
manufacturer m
m Manufacturer
{1,…, M} Qit
Quantity produced of
product i at time t
w Warehouse
{1,…, W} ha
Holding cost for raw
materials a
r Retailer
{1,…, R} Qat
Inventory of raw
materials a at time t at
manufacturer m
Book of Proceedings
25
c Customer
{1,…, C} him
Holding cost for
product i at
manufacturer m
t Time Period
{1,…, T} Qimt
Inventory of product i
at manufacturer m at
time t
a Raw Materials
{1,…, A} TRCsm
Transportation cost per
product per distance
value from supplier s to
manufacturer m
Pimw
Price charged for
inbound units of product
i at distributor w
(Logistics Cost)
QRETimwt
Quantity of product i
transported from
manufacturer m to
distributor w at time t
for retailers’ use
Qimwt
Quantity of product i
transported from
manufacturer m to
distributor w at time t for
customers’ use
Orm Order cost for
manufacturer m
Pa
Price charged for raw
materials from supplier s
to manufacturer m
MaxTVCmw
Maximum
transportation vehicle
capacity from
manufacturer m to
distributor w
Qasmt
Quantity of raw materials
a transported from
supplier s to
manufacturer m at time t
MaxTVCsm
Maximum
transportation vehicle
capacity from supplier
s to manufacturer m
WHi,m,t
Working Hours for
product i at manufacturer
m at time t
Prim
Production rate per
working hour per
employee for product i
at manufacturer m
QOirt Fixed order quantity of
product i at time t from
retailer r QOiwt
Fixed order quantity of
product i at time t from
distributor w
Ei,m Employee for product i at
manufacturer m MaxIim
Maximum inventory
capacity of product i at
manufacturer m
MaxIam
Maximum inventory
capacity of raw materials
a at manufacturer m
INiw
Initial inventory of
product i at distributor
w
INim
Initial inventory of
product i at manufacturer
m
Bt
The minimum value
between inventory of
product i at distributor
w and demand for
Book of Proceedings
26
product i from retailer r
to distributor w
INam
Initial inventory of raw
materials a at
manufacturer m
Pirc
Price charged for
product i from retailer r
to customer c
Lt
The minimum value
between inventory of
product i at manufacturer
m and demand for
product i from distributor
w to manufacturer m
Dirct
Demand for product i
from customer c at
retailer r at time t
Piwr
Price charged for
outbound units of
product i at distributor w
(Logistics Cost)
hir Holding cost for
product i at retailer r
Qiwrt
Quantity of product i
transported from
distributor w to retailer r
at time t
Qirt Inventory of product i
at retailer r at time t
hiw Holding cost for product
i at distributor w TRCwr
Transportation cost per
product per distance
value from distributor
w to retailer r
Qiwt
Inventory of product i at
distributor w at time t for
customer use firt
Percentage of defective
products i the retailer r
has to compensate for
at time t
TRCmw
Transportation cost per
product per distance
value from manufacturer
m to distributor w
ORr Order cost for retailer r
kirct
Quantity of product i
transported from retailer r
to customer cat time t to
replace default products
GCrt
Generalized cost at
retailer r per time
period t
ORw Order cost for distributor
w MaxIir
Maximum inventory
capacity of product i at
retailer r
MaxTVCwr
Maximum transportation
vehicle capacity from
distributor w to retailer r
INir Initial inventory of
product i at retailer r
MaxIiw
Maximum inventory
capacity of product i at
distributor w QALLiwt
Total inventory of
product i at distributor
w at time t
Book of Proceedings
27
n
Binary variable for the
decision whether an order
occurs or not QOat
Fixed order quantity of
raw material a at time t
from manufacturer
σ
Constant value used for
the determination of
penalty cost
References
Paschalidis I.Ch., Liu Y., Cassandras C. and Panayiotou C. “Inventory Control for Supply Chains
with Service Level Constraints: A Synergy between Large Deviations and Perturbation Analysis”.
Annals of Operations Research. Vol. 126, 2004, pp. 231-258.
Elahi B., Pakzad-Jafarabadi Y., Etaati L. and Seyed Hosseini S.M. “Optimization of Supply Chain
Planning with Considering Defective Rates of Products in Each Echelon”. Technology and
Investment. Vol. 2, 2011, pp. 211-221.
Farahani R.Z. and Elahipanah, M. “A genetic algorithm to optimize the total cost and service level
for just-in-time distribution in a supply chain”. International Journal of Production Economics.
Vol. 111, 2008, pp. 229-243.
Mirzapour A.M.J., Baboli A. and Sazvar, Z. “A stochastic aggregate production planning model
in a green supply chain: considering flexible lead times, nonlinear purchase and shortage cost
functions”. European Journal of Operational Research. Vol. 230, 2013, pp. 26-41.
Table 1 Notation
Book of Proceedings
28
Internet and agro-tourism sector for regional development in Crete:
A multicriteria ranking
Zacharoula Andreopoulou
Laboratory of Forest Informatics, School of Forestry and Natural Environment, Aristotle
University of Thessaloniki, Box 247, 54124, Greece
Christos Lemonakis
School of Management & Economics, Technological Educational Institute of Crete, Agios
Nikolaos Branch, Greece
Christiana Koliouska
Laboratory of Forest Informatics, School of Forestry and Natural Environment, Aristotle
University of Thessaloniki, Box 247, 54124, Greece
Konstantinos Zopounidis
Department of Production Engineering and Management, Technical University of Crete, Greece
Authors’ e-mail addresses: [email protected],[email protected],[email protected],
Abstract
Nowadays, effective use of Internet provides an opportunity to identify successful practices and
policies for innovative business models in order to promote regional development through agro-
tourism. Agro-tourism sector can exploit natural and rural resources in the context of employment,
growth and competitiveness. In Crete, agro-tourism entrepreneurs have developed commercial
activities in the Internet where customers and firms are linked up together in the exchange of agro-
tourism services. This paper aims to assess websites of commercial purpose within agro-tourism
sector in Crete and rank them according to multiple criteria using the multicriteria analysis method
of PROMETHEE II.
Keywords: agrotourism, regional development, internet, website assessment, total ranking,
multicriteria analysis, promethee II, business model, e-commerce
1. Introduction
Agrotourism constitutes an integrated and sustainable regional development approach.
Agrotourism, farm tourism or agricultural tourism is the process of attracting visitors and travellers
to agricultural areas (Ezung, 2011), primarily for agricultural purposes. Opportunities for
uniqueness and customization are limitless within the Greek context (Zopounidis et al., 2014).
Nowadays, the Internet, apart from a channel to collect information, offers enterprises the
Book of Proceedings
29
opportunity to market goods and services to more customers than ever before (Griffin, 2000). The
enterprises aim at their participation in the internet society since the benefits are high and electronic
systems are ready to serve customers all over the world 24 hours per day and 7 days a week
(Andreopoulou et al., 2011; Koliouska and Andreopoulou, 2013). Many agro-tourism firms have
already developed their commercial websites, because their Internet presence tends to be a
fundamental component, in order to sustain a successful enterprise.
This paper aims to assess websites of commercial purpose within agro-tourism sector in Crete
and rank them according to multiple criteria. Initially, qualitative and quantitative features were
identified in the collected websites. Further, the websites were ranked according to these features
to be used as criteria using the multicriteria analysis method of PROMETHEE II. Furthermore, the
optimum agrotourism firms’ websites are identified and described to be used as models with the
total internet adoption.
2. Materials and methodology
The websites that promote agrotourism activities in the region of Crete are retrieved from the
Internet through large-scale hypertextual search engines, such as “Google”, “Yahoo”, “Pathfinder”
and “MSN Search”, which bring satisfying results. Various keywords and combinations were used
such as “agro-tourism sector in Crete”, “agro-tourism firm in Crete”, “agro-tourism services in
Crete”, etc.
There were a variety of features introduced in these websites, aiming to promote the agro-
tourism sector in Crete. The criteria/features were used to describe variables x1, x2, …, xn. These
criteria are presented in Table 1. Initially, qualitative analysis was implemented in order to examine
the type of common criteria/ features, representing internet adoption, found in these commercial
websites. Then, a quantitative analysis through a 2-dimentional table was performed in order to
examine the presence or absence of these features. The value of 0 and the value 1 were attributed
to the variables x1, x2, …, xn, for the non-existence and the existence of each criterion respectively.
Variable Features Variable Features
X1 Information about products,
services or activities
X8 Online reservation
(enabled with online payment)
X2 Current prices X9 Online communities
(forums, chat rooms, guestbooks)
X3 Contact information X10 Additional topics with information on
different categories
X4 Local information X11 Code access: website areas where
access is allowed only for members
through codes or passwords
X5 Links to other companies X12 Third person advertisement
Book of Proceedings
30
X6 Related sources of information X13 Personalization of the page, trace,
safety
X7 Online reservation (enabled
with traditional ways of
payment)
Table 1. Variables attributed to features, representing internet adoption
Then, the total ranking of the websites was studied. The method that was used for the total
ranking was the multicriteria analysis named PROMETHEE II. That method applies a linear form
of service in the particular case and fits better to the targets of the project even if it is compared to
other well-established methods (Andreopoulou et al., 2014). The identified e-marketing services
of the websites are used as criteria in order to determine the superiority of one website over another
website and the net flow for each website is estimated, which is the final number in order the
websites to be ranked from the best to the worst (Zopounidis, 2001).
3. Results and discussion
The research on the Internet resulted in the retrieval of 29 websites that support and promote
agro-toursim services for regional development in the region of Crete. Based on the application of
the PROMETHEE II method, the first and the last 10 cases of the total ranking of the agro-tourism
firms’ websites in Crete are presented in Table 3. In the same Table it is also presented the total
net flow that is estimated for each website and it is used for the comparison between the websites
in order to obtain the total ranking, as each website with a higher net flow is considered superior
in ranking.
The findings of the research show that, the values estimated for total net flows φ present a
spectrum of values between +0,82 to -0,73 and that indicates a great difference concerning
“superiority” between the first and the last case in the ranking of the agrotourism firm website.
The websites with high ‘superiority” are the websites which provide a price list (x2) and useful
links to other relevant firms (x5). These websites also, allow the users to make on-line reservation
(x7) and to pay through the Internet (x8). Some other features that improve the total net flow of a
website are the provision of information on different topics (x10), the third person advertisement
(x12) and the ability for the users to personalize the website (x13).
Furthermore, the eleventh variable (the ability to create a user account) and the third variable
(contact information) are not so essential for the website efficiency. The web visitors of these
commercial websites are mainly interested in the agro-tourism activities of the firms (x1) and not
so much for the local area information (x4). Furthermore, the ninth variable (social media,
guestbook) seems not to be critical. The twelfth variable (third person advertisement) accompanied
by the thirteenth variable (personalization of the page, trace, safety) refer to more experienced
users.
Book of Proceedings
31
Total
Ranking
Website Agro-tourism Firm Name Total Net
Flow
1 http://www.vamossa.gr Vamos Traditional Village 0,8297
2 http://sarris-house.gr/ Sarris House 0,5934
3 http://www.vamos-xatheri.gr Xatheri Villas 0,4725
4 http://www.milia.gr Milia Mountain Retreat 0,3626
5 http://drys-villas.gr Drys Villas 0,3516
6 http://www.syiahotel.gr Syia Hotel 0,3242
7 http://www.stratosvillas.com/ Stratos Villas 0,2967
8 http://www.melidonixvillage.gr/ Melidoni x village 0,2912
9 http://www.aspalathos-villas.gr Aspalathos Villas 0,2418
10 http://www.iliopetra-milopetra.gr Iliopetra Studios 0,1264
…. …………… …………
20 http://www.akrosoreon-crete.gr/ Akros Οreon -0,1703
21 http://www.mohlos.com/ Mohlos Villas -0,1978
22 http://www.arodamoslivadi.gr Arodamos -0,2198
23 http://www.kouritonhouse.gr Kouriton House -0,2253
24 http://agroikies.gr/ Agrikies Stratakis Estate -0,2857
25 http://www.listarossa.gr Listaros S.A. "ξα σου" -0,3352
26 http://www.lasinthos.gr Lasinthos -0,4121
27 http://www.earino.gr Earino -0,6154
28 http://www.viglatoras.gr Viglatoras -0,6264
29
http://www.arolithos.com
Arolithos Traditional Cretan
Village -0,7308
Table 2. Ten greater and ten worst in total ranking of Agro-tourism Websites in Crete and Total
Net Flows-TNF
4. Conclusions
Findings confirm that internet adoption in agro-tourism sector in Crete is still in initial level.
The results indicate that the effective internet sites require: price list, useful links to other relevant
firms, online reservation and payment system, information on various topics, third person
advertisement and personalization of the website.
Book of Proceedings
32
The findings are useful in improving internet technologies adoption through the improved
design and implementation of an internet presence to accomplish certain characteristics and to
generally optimize the internet activities in agrotourism sector in Crete. Given the public
awareness for the environment within tourism, eco-agrotourism websites should evolve in further
e-service provision (Andreopoulou and Koutroumanidis, 2009).
References
Andreopoulou, Z., Koliouska, C., Lemonakis, C. and Zopounidis, C. (2014). National Forest Parks
development through Internet technologies for economic perspectives. Operational Research, 1-
27. DOI 10.1007/s12351-014-0147-8 (electronic version)
Andreopoulou, Z. and Koutroumanidis, Τ. (2009). Assessment of the ICT adoption stage in eco-
agrotourim websites in Greece. 6th International Conference of Management of Technological
Changes, 3rd-5th September, Alexandroupolis, Greece. Book I. Pp.441-444.
Andreopoulou, Z., Manos, B., Viaggi, D., Polman, N. (2011). Agricultural and Environmental
Informatics, Governance, and Management: Emerging Research Applications. IGI Global. USA.
Ezung, T. Z. (2011). Rural Tourism in Nagaland, India: Exploring the Potential. International
Journal of Rural Management, 7(1-2), 133-147.
Griffin, M. (2000) Emarketing Planning: Accountability and Emetrics. Embelix Software [online]
http://www.templatezone.com/pdfs/ems_whitepaper.pdf (Accessed 1 July 2013).
Koliouska C. and Andreopoulou Z. (2013). Assessment of ICT Adoption Stage for Promoting the
Greek National Parks. Procedia Technology, Vol. 8, pp. 97-103.
Zopounidis C. (2001). Analysis of financing decisions with multiple criteria. Anikoula
Publications, Thessaloniki.
Zopounidis, C., Lemonakis, C., Andreopoulou, Z. and Koliouska, C., 2014. Agrotourism Industry
Development through Internet Technologies: A Multicriteria Approach. 53rd Meeting of the
EURO Working Group for Commodities and Financial Modelling (EWGCFM), Chania, Crete,
May 22-24.
Book of Proceedings
33
On the way to a reference model for supply chains in the construction
industry
Ilias P. Tatsiopoulos
National Technical University of Athens, School of Mechanical Engineering, Sector of Industrial
Management and Operational Research, Heroon Polytechniou 9, 15780 Zografou, Greece
Dimitrios-Robert I. Stamatiou
National Technical University of Athens, School of Mechanical Engineering, Sector of Industrial
Management and Operational Research, Heroon Polytechniou 9, 15780 Zografou, Greece
Abstract
The construction industry is a project based industry with many particularities that differ in
regions, projects and/or cultures. The fact that there are many actors involved at different levels in
the construction process, with low quality information exchange caused by restricted use of
communication channels, makes it inefficient. In this paper we examine the literature on supply
chain reference models for the construction industry. The search demonstrates that there are no
universally accepted reference models for the construction supply chain. The lack of findings is
probably due to the small amount of research on supply chain management in construction when
compared to manufacturing. In order to cover this gap, we propose the adoption of the Supply
Chain REMEDY reference model for the project based construction industry. We believe that the
efforts construction companies make towards customer delight, one of their basic objectives in
every project, could be supported by the existence of a reference model that takes into
consideration the particularities of the sector. In the first stage, a generic reference model for the
construction industry will be developed, tested and informed. In the second stage, partial models
will be developed for groups of construction projects described in the paper. The article presents
a brief literature review, followed by the presentation of the current literature on reference models
in the literature and the construction industry literature. Conclusions make the final part of the
paper.
Keywords: Construction, Supply chain management, Reference model, Supply chain remedy,
Construction supply chains
Book of Proceedings
34
5. 1. Introduction
The construction industry is a project based industry. It traditionally involves many actors that
work at a local basis and are usually involved in a number of projects simultaneously. The final
products have a long life span and, in most cases, the final consumer is unknown during the project
execution. Construction markets are closed to the global competition due to government subsidies,
national and local regulations and culture (Segerstedt & Olofsson 2010) and competition doesn’t
work as effectively as it does in other industries. In this paper we propose that a reference model
regarding the supply chains of construction projects will help to improve the efficiency of these
supply chains. In the following chapters, first, the concept and some problems of construction
supply chains are described, second, the description of the reference model is provided and last,
conclusions are drawn as to the feasibility of a reference model and the steps to follow.
6. 2. Construction supply chain management
Supply chain management is a branch that has received plenty of attention for the past decades
and will continue to receive attention at the same scale, if not larger, in years to come. Although
there is so much research going on, not all industries with supply chains have been studied to the
same extent. The construction industry is one of the industries less studied. Eccles (1981) defined
construction as “the erection, maintenance, and repair of immobile structures, the demolition of
existing structures, and land development”. Each of these functions usually involves a tuple of
actors. These actors do not always have the same amount of information coming their way and,
most likely, they do not belong to the same tier of the supply chain. Construction is a sector with
many particularities. One of the main problems is that the industry faces high fragmentation with
many SMEs, as noted by Briscoe & Dainty (2005). Other problems that impede the adoption of
supply chain management in the construction industry include discontinuous demand, uniqueness
of each project in technical, financial and sociopolitical terms (Segerstedt & Olofsson 2010),
concentration of main contractors exclusively on the clients (Saad et al. 2002) and the fact that
coordination is mainly driven through project management techniques and alignment of ICT
systems (Briscoe & Dainty 2005).
Construction supply chain management offers new approaches to reduce the cost and increase the
reliability and speed of construction (O’Brien 1999). Persson et al. (2010) defined construction
supply chain management as “the task of integrating organizational units along a supply chain,
including the construction site and subcontractors, and coordinating materials, information and
financial flows with the project site plan in order to fulfill the (ultimate) customer demands”.
Construction supply chains have many particularities, especially when one compares them to
manufacturing supply chains. Many times, projects have contracts stating financial clauses in case
of delays, material mismatches and declinations from the specified designs that may not exist to
such an extent in manufacturing. On the other hand, Cox & Ireland (2002) found that dominant
thinking in the construction sector lacks an understanding of contextual factors highly regarded in
the manufacturing industry, like Porter’s five forces.
In the 90s, after the publication of the Latham (1994) and Egan (1998) reports, large construction
firms first adopted the partnering concept proposed by Latham and then moved on to formulating
Book of Proceedings
35
even tighter relationships with companies that they collaborated with as proposed by Egan. Ever
since, there have been more reports (Egan 2002; Wolstenholme 2009; Bourn 2001; Harris 2013)
that propagate the need for tighter relationships between construction firms and higher quality
information sharing. A reference model for the construction industry supply chains, just as in the
manufacturing industry, can provide a basis for improving processes, information sharing and
collaboration between constructions firms involved in a project.
7. 3. Construction supply chain reference model
In order to check the applicability of a reference model to the industry and propose a suitable
reference model, we performed a search in three stages (Figure 1). First, we searched the literature
on supply chain management for available reference models. Second, we searched the literature
on construction supply chain management for previous attempts to adopt or create reference
models. Third, we propose what we believe to be a suitable reference model for construction
supply chains.
Figure 1: Research Process
Following the first step of the methodology, a brief literature review was conducted. The results
of the review are based on previous works on available reference model presentation (Fettke et al.
2005) and research conducted through the academic databases Scopus and SpringerLink. They
include seven reference models which we divide into three groups; process models, IT based
models and conceptual academic models. The first group is comprised by the Supply Chain
Operations Reference Model – SCOR (http://supply-chain.org/) and the Global Supply Chain
Framework (http://www.theglobalsupplychainforum.com/), the second group is comprised by the
SAP R/3 reference model by SAP AG (www.sap.com) and the Collaborative Planning, Forecasting
and Replenishment – CPFR reference model and the third group is comprised by the Mentzer
reference model (Mentzer et al. 2001) and efforts by Verdouw et al. (2011) and Klingebiel (2008).
Although none of these models is industry specific, most of them carry a manufacturing industry
nature.
The literature review on construction supply chains, dictated by the second step of the
methodology, yielded a very small number of results, indicating a need for research directed to the
matter. On the one hand, there have been some efforts in the literature (Persson et al. 2010; Cheng
et al. 2010; Yeo & Ning 2002; London & Kenley 2000) to apply the most popular reference
models, SCOR and GSCF, to construction supply chains. On the other hand, the efforts retrieved
for industry specific reference models were limited to the Process Protocol (Kagioglou et al. 2000)
and to a conceptual model proposed by (Aloini et al. 2012). The results from the first two steps
Check Supply Chain Management literature
for reference models
Check Construction Supply Chain
Management literature for reference models
Proposal of a reference model for construction
supply chains
Book of Proceedings
36
show that there is a gap in the literature concerning reference models on construction supply chain
management. In order to fill in this gap, at the third step of our methodology, the Supply Chain
REMEDY model was selected as a basis.
The Supply Chain REMEDY model (Gayialis et al. 2013; Ponis et al. 2014) is being developed
under the research project “A Holistic Approach for Managing Variability in Contemporary Global
Supply Chain Networks”. It is a process-based generic supply chain reference model that is being
built to be adaptable to different production strategies, from Engineer-to-Order to Make-to-Stock.
The model offers multiple views to the user; process, knowledge, risk, decision, algorithmic and
IT views. It is comprised of nine main functions that are, respectably, analyzed in over ninety high
level processes that focus on demand variability management. The knowledge and risk enhanced
views offer a valuable and versatile tool for all industries and key decisions that highly effect the
entire supply chain are pinpointed. The other views offer support to IT processes, especially in
industries that heavily rely on technology for their demand-supply operations. We propose that
the model is adapted to the project-based construction industry in order to cover the gap of
construction supply chain reference models. This industry specific approach is going to be
developed and updated following the methodology “modeling, measuring and improving”
proposed by Jianyuan & Fan (2006). Firstly, the existing high level processes will be brought
closer to the construction industry reality and analyzed to an industry specific level, but still generic
enough to be adaptable to any construction project. We recognize that no two construction projects
are the same, but there are distinctions to be made and groups to be formed. We divide construction
projects in two major categories; public and private projects. Public projects are government
owned but, in most cases, subcontracted to private construction firms. They can be divided into
infrastructure projects (motorways, gas lines, etc.) and common-wealth projects (parks, hospitals,
etc.). Private projects are more heterogeneous but, can be grouped into housing projects (private
or apartment buildings) and financially profitable projects (shopping malls, office blocks,
factories, etc.). These four groups will respectively provide a basis for partial construction supply
chain reference models. Secondly, the model and the partial models described above will be tested
on case studies in the literature and, if possible, through real life cases. Its performance will be
measured through carefully defined metrics and, thirdly, processes will be aligned accordingly
based on the review of the performance measurements.
8. 4. Conclusions
The research conducted realizes that construction supply chains have many particularities. These
particularities have made the adoption of supply chain management principles difficult. We
identified a gap in the literature regarding reference models specific to the industry and we sought
to contribute. The attempt to create a reference model for this industry’s supply chains, that takes
into consideration its project based nature, promises to create a starting point and supporting tool
for professionals to understand basic and advanced concepts of supply chain management that
elude them. The methodology to be followed sees to make the reference model very adaptable to
all companies in the sector and present opportunities to create new and manage existing
knowledge.
Book of Proceedings
37
Acknowledgement
The research efforts described in this paper are part of the research project “A Holistic Approach
for Managing Variability in Contemporary Global Supply Chain Networks” in research action:
“Thales - Support of the interdisciplinary and/or inter-institutional research and innovation”, which
is implemented under the Operational Programme: Education and Lifelong Learning, NSRF 2007-
2013 and is co-funded by European Union (European Social Fund) and Greek Government.
References
Aloini, D., Dulmin, R., Minnino, V. and Ponticelli, S., “A conceptual model for construction
supply chain management implementation”. Proceedings of the 28th Annual ARCOM Conference.
Edinburgh, UK, 2012, pp. 675–685.
Bourn, J., “Modernising construction”. London, 2001
Briscoe, G. and Dainty, A., “Construction supply chain integration: an elusive goal?”. Supply
Chain Management: An International Journal, Vol. 10(4), 2005, pp.319–326.
Cheng, J.C.P., Law, K.H., Bjornsson, H., Jones, A. and Sriram, R.D., “Modeling and monitoring
of construction supply chains”. Advanced Engineering Informatics, Vol. 24(4), 2010, pp.435–455.
Cox, A. and Ireland, P., “Managing construction supply chains: the common sense approach”.
Engineering, Construction and Architectural Management, Vol. 9(5/6), 2002, pp.409–418.
Egan, Sir J., “Accelerating change”, London, 2002.
Fettke, P., Loos, P. and Zwicker, J., “Business process reference models”. In C. J. Bussler & A.
Haller, eds. Business Process Management Workshops. Springer Berlin Heidelberg, 2005, pp.
469–483.
Gayialis, S.P., Ponis, S.T., Tatsiopoulos, I.P., Panayiotou, N.A. and Stamatiou, D.R.I., , A
“Knowledge-based Reference Model to Support Demand Management in Contemporary Supply
Chains”. Proceedings of the 14th European Conference on Knowledge Management. Kaunas,
Lithuania, 2013, pp. 236–246.
Harris, E.C., “Supply Chain Analysis into the Construction Industry: A Report for the Construction
Industrial Strategy”, London, 2013.
Jianyuan, Y. and Fan, Q., “The System Model for Supply Chain Process Optimization”.
Proceedings of 2006 International Conference on Management of Logistics and Supply Chain.
Sydney, Australia, 2006, pp. 266–271.
Kagioglou, M., Cooper, R., Aouad, G. and Sexton, M., “Rethinking construction: the Generic
Design and Construction Process Protocol”. Engineering, Construction and Architectural
Management, Vol. 7(2), 2000, pp.141–153.
Book of Proceedings
38
Klingebiel, K., “A BTO Reference Model for High-Level Supply Chain Design”. In G. Parry &
A. Graves, eds. Build To Order: The Road to the 5-Day Car. Springer London, Limited, 2008, pp.
257–276.
London, K. & Kenley, R., “Mapping construction supply chains: widening the traditional
perspective of the industry”. Proceedings 7th Annual European Association of Research in
Industrial Economic EARIE conference. Lausanne, Switzerland: European Association of
Research in Industrial Economics, 2000.
Mentzer, J.T., DeWitt, W., Keebler, J.S., Min, S., Nix, N.W., Smith, C.D. and Zacharia, Z.G.,
“Defining supply chain management”. Journal of Business Logistics, 22(2), 2001, pp.1–25.
O’Brien, W., “Construction supply-chain management: a vision for advanced coordination,
costing, and control”. NSF Berkeley-Stanford Construction Research Workshop, 1999, pp.1–7.
Persson, F., Bengtsson, J. and Gustad, Ö., “Construction Logistics Improvements Using the SCOR
Model–Tornet Case”. Advances in Production Management Systems, 338, 2010, pp.211–218.
Ponis, S.T., Gayialis, S.P., Tatsiopoulos, I.P., Panayiotou, N.A, Stamatiou, D.R.I. and Ntalla, A.C.,
“Modeling Supply Chain Processes : A Review and Critical Evaluation of Available Reference
Models”. Proceedings of 2nd International Symposium and 24th National Conference on
Operational Research. Athens, Greece, 2014, pp. 270–276.
Saad, M., Jones, M. and James, P., “A review of the progress towards the adoption of supply chain
management (SCM) relationships in construction”. European Journal of Purchasing & Supply
Management, Vol. 8(3), 2002, pp.173–183.
Segerstedt, A. and Olofsson, T., “Supply chains in the construction industry”. Supply Chain
Management: An International Journal, Vol. 15(5), 2010, pp.347–353.
Verdouw, C.N., Beulens, A.J.M., Trienekens, J.H. and van der Vorst, J.G.a.J., “A framework for
modelling business processes in demand-driven supply chains”. Production Planning & Control,
Vol. 22(4), 2011, pp.365–388.
Wolstenholme, A., “Never Waste a Good Crisis: A Review of Progress since Rethinking
Construction and Thoughts for Our Future”, London, 2009.
Yeo, K. and Ning, J., “Integrating supply chain and critical chain concepts in engineer-procure-
construct (EPC) projects”. International Journal of Project Management, Vol. 20(4), 2002,
pp.253–262.
Book of Proceedings
39
The problem of robustness in the MUSA method: Theoretical developments
and applications
Yannis Politis
School of Science and Technology, Hellenic Open University, Parodos Aristotelous 18, GR26335
Patra, Greece, email: [email protected]
Evangelos Grigoroudis
School of Production Engineering and Management, Technical University of Crete, University
Campus, GR73100 Chania, Greece.
Ifigeneia Pologiorgi
School of Production Engineering and Management, Technical University of Crete, University
Campus, GR73100 Chania, Greece.
Abstract
The MUSA method is a collective preference disaggregation approach which has been developed
in order to measure and analyze customer satisfaction. It follows the main principles of ordinal
regression analysis under constraints using linear programming techniques and it is used for the
assessment of a set of marginal satisfaction functions in such a way that the global satisfaction
criterion becomes as consistent as possible with customer’s judgments. Considering that the
MUSA method is based on a linear programming modelling, the problem of multiple or near
optimal solutions appears in several cases. This has an impact on the stability level of the provided
results. The quality of collected data collected and the incapability to interact with customers
complicates the task of finding stable solutions. For this reason different ways to overcome this
problem may include asking customers to give additional information (e.g. information about the
importance of the criteria along with the usual satisfaction questions) or introducing additional
constraints in the basic LP of the method which will reduce the polyhedron of feasible solutions.
This study presents the implementation of an extension of the MUSA method in a real case study
concerning the evaluation of customer satisfaction from Greek mobile service providers. More
specifically, additional constraints regarding special properties of the assessed model variables and
additional customer preferences about the importance of the criteria have been incorporated in the
LP of the original MUSA method and have been modelled as a Multiobjective Linear
Programming (MOLP) problem. The main aim of the study is to show how the introduction of
these additional constraints and information can improve the stability level of the estimated results.
Different stability and fitting measures have been used in order to analyze and compare the
provided results. The application showed that the introduction of the additional constraints and
information in the original MUSA method has improved the method’s robustness, still without
affecting the conclusions drawn by the implementation of the basic MUSA model, enhancing thus
the proposed conclusions and improvement actions.
Keywords: MUSA Method, Robustness Analysis, Mobile Services, Satisfaction analysis.
Book of Proceedings
40
1. Introduction
The MUSA (MUlticriteria Satisfaction Analysis) method is a preference disaggregation
approach following the main principles of ordinal regression analysis. It measures and analyzes
customer satisfaction assuming that customer’s global satisfaction is based on a set of criteria
representing service characteristic dimensions. The data collected to measure customer satisfaction
can have such characteristics that make it impossible to find a feasible solution consistent with
customers’ preferences. Incapability to interact with customers further complicates the task of
finding stable solutions. For this reason actions are required in order to improve the stability of the
results. Such actions may include asking customers to give additional information (e.g. information
about the importance of the criteria along with the usual satisfaction questions) or introducing
additional constraints in the basic LP of the method which will reduce the polyhedron of feasible
solutions. Based on an extension of the MUSA method with the introduction of additional
constraints, a real-world application in Greek mobile service providers is presented.
2. The MUSA Method
2.1 Mathematical Development
According to the MUSA method, customer’s global satisfaction is based on a set of criteria
representing service characteristic dimensions. The main object of the MUSA method is the
aggregation of individual judgments into a collective value function. The method is an ordinal
regression-based approach used for the assessment of global and partial satisfaction functions *Y
and *
iX respectively, given customers’ judgments Y and iX (Grigoroudis and Siskos, 2002;
Grigoroudis and Siskos, 2010):
* *
1 1
with 1n n
i i i
i i
Y b X b
(1)
where ib is the weight of the i-th criterion and the value functions *Y and *
iX are normalized in
the interval [0, 100]. Introducing a double-error variable, the ordinal regression equation becomes
as follows:
1
n* *
i i
i
Y b X
(2)
where *Y is the estimation of overall value function *Y and and
are the overestimation
and underestimation error, respectively. The following transformations which represent the
successive steps of the value functions *Y and *
iX can be introduced in the model:
* 1 *
* 1 *
for 1 2 1
for 1 2 1 and 1 2
m m
m
k k
ik i i i i i
z y y m= , ,...,
w b x b x k = , ,..., i = , ,...,n
(3)
The final LP of the basic MUSA model has the following form:
Book of Proceedings
41
1
1 1
1 1 1
1
1
1
1 1
[max]
subject to
0 1, 2, ,
100
100
, , , 0 , , ,
ji j
i
M
j j
j
t tn
ik m j j
i k m
m
m
n
ik
i k
m ik j j
F
w z j M
z
w
z w i j k m
(4)
where jt and
jit are the judgments of the j-th customer globally and partially for each criterion
1,2, ,i n and M is the number of customers.
2.2 Stability Analysis
The stability analysis is considered as a post-optimality analysis problem, taking into account
that the MUSA method is based on a LP modeling. During the post-optimality analysis stage, n
LPs (equal to the number of criteria) are formulated and solved. Each LP maximizes the weight of
a criterion and has the following form:
1
1
*
max for 1,2, ,
subject to
all the constraints of LP (4)
i
ik
k
F w i n
F F
(5)
where *F is the optimal value of the objective function of LP (4) and is a small percentage of *F . The average of the optimal solutions given by the n LPs (5) may be considered as the final
solution of the problem. In case of instability, a large variation of the provided solutions appears
and the final average solution is less representative.
2.3 Basic Results
The method estimates a set of useful indices for benchmarking purposes such as the average
global and partial satisfaction indices S and iS , which can be assessed according to the following
equations:
*
1
*
1
1
100
1 for 1,2, ,
100
i
αm m
m
αk k
i i i
k
S p y
S p x i n
(6)
where mp and k
ip are the frequencies of customers belonging to the my and k
ix satisfaction
levels respectively. Similarly, the average global and partial demanding indices, D and iD ,
Book of Proceedings
42
represent the average deviation of the estimated value curves from a “normal” (linear) function
and reveal the demanding level of customers. They are normalised in the interval [–1, 1] and, are
assessed as follows:
1*
1
1
1
1*
1
1
1
100 1
1 for 2
1100
1
100 1
1 for 2 and 1,2, ,
1100
1
i
αm
m
m
αk
i
k i
i i
k i
my
αD α
m
α
kx
αD α i n
k
α
(7)
2.4 Fitting and Robustness Indicators
The fitting level of the MUSA method refers to the assessment of a preference collective value
system (value functions, weights, etc.) for the set of customers with the minimum possible errors.
Three fitting indices are proposed which are normalized in the interval [0, 1]. The Average Fitting
Index ( 1AFI ) depends on the optimum error level and the number of customers:
*
1 1100
FAFI
M
(8)
An alternative fitting indicator is based on the percentage of customers with zero error
variables, i.e., the percentage of customers for whom the estimated preference value systems fits
perfectly with their expressed satisfaction judgments. This average fitting index 2AFI is assessed
as follows:
0
2
MAFI
M (9)
where 0M is the number of customers for whom 0 . This is a rather strict indicator
considering that it examines only the existence of non-zero errors, without taking into account the
values of these error variables.
A third fitting indicator 3AFI examines separately every level of overall satisfaction and
calculates the maximum possible error value for each one of these levels:
*
3* *
1
1
max ,100m m m
m
FAFI
M p y y
(10)
The Average Stability Index ( ASI ) is an indicator of the MUSA’s method stability level. ASI
is nothing else than the mean value of the normalized standard deviation of the estimated weights
during the post-optimality stage and is calculated as follows:
2
2
1 1
1
11
100 1
n nj j
i in
j j
i
n b b
ASIn n
(11)
Book of Proceedings
43
where j
ib is the estimated weight of the i-th criterion in the j-th post-optimality analysis LP. ASI
is normalized in the interval [0, 1].
3. Modeling Additional Information and Properties
3.1 Preferences on Criteria Importance
Introducing additional constraints in the MUSA method can limit the space of feasible
solutions and therefore increase the stability of the method. These constraints may concern
preferences on criteria importance or desired properties of the provided results.
Particularly, a customer satisfaction survey may include, besides the usual performance
questions, preferences about the importance of the criteria. Using such questions, customers are
asked either to judge the importance of a satisfaction criterion using a predefined ordinal scale, or
rank the set of satisfaction criteria according to their importance.
The evaluation of preference importance classes iC is similar to the estimation of thresholds
iT . An ordinal regression approach may also be used in order to develop the weights estimation
model. The WORT (Weights evaluation using Ordinal Regression Techniques) model is presented
in LP (12) in which the goal is to minimize the sum of errors under a set of constraints according
to the importance class that each customer j considers that a criterion i belongs (Grigoroudis and
Spiridaki, 2003):
2
1
1 1
1
1
1
1
1
1
1
1
1
[min]
subject to
ˆ100 0,
100 0,
ˆ, , 2,..., 1
100 0
ˆ100 0,
i
i
i
i
ij ij
j i
a
it ij ij
t
a
it l ij
t
ij la
it l ij
t
a
it q ij ij q
t
F S S
w T S b C
w T Si
b C l q
w T S
w T S b C
1
1 1
1
2 1
1 2
100
, , 0 , , ,
ian
ik
i k
q
q q
ik ij ij
j
w
T
T T
T T
w S S i j k
(12)
Book of Proceedings
44
where ˆijb is the preference of customer j about the importance of criterion i , is a positive
number used to avoid cases where ij ib T i , and is a small number introduced to increase the
discrimination of the importance classes.
3.2 Desired Properties of the Results
A linkage between global and partial average satisfaction indices may be assumed, since these
indices are considered as the main performance indicators of the business organization. In
particular, the global average satisfaction index S is assessed as a weighted sum of the partial
satisfaction indices iS :
* *
1 1 1 1
in nm m k k
i i i i i
i m i k
S b S p y b p x
or 1 1
2 1 1 2 1
im n km k
t i it
m t i k t
p z p w
(13)
Similarly, a weighted sum formula may be assumed for the average demanding indices:
1
n
i i
i
D b D
or
1 11 1 1
1 1 1 1 1
1
100( 1) ( 1) ( 1) ( 1)
( 1) ( 1)
i im k
t it i itn
m t k t t
i i i
m z k w w
(14)
Equations (13) and (14) may be introduced as additional constraints in the LP (4). However,
these additional constraints should be used carefully, since they do not guarantee a feasible solution
of the LP, especially in case of inconsistencies between global and partial satisfaction judgments.
For this reason, the aforementioned equations may be written using a goal programming
formulation and used alternatively as post-optimality criteria. A double error variable es , es and
ed , ed
can be introduced in each one of the equations for this reason.
3.3 Extension of the MUSA Method
Using both customers’ performance and importance judgments and introducing, at the same
time, additional constraints about the average satisfaction and demanding indices, an extension of
the MUSA method may be modeled as a Multiobjective Linear Programming (MOLP) problem.
1
1 1
min
min
min
subject to
all the constraints of LPs (4) and (12)
constraints (13) and (14)
M
j j
j
n M
ij ij
i j
F
S S
es es ed ed
(15)
The above problem may be solved using any MOLP technique (e.g., compromise
programming, global criterion approach). Here, an alternative heuristic method, consisting of four
steps (lexicographic approach), is presented:
Step 1. Min F subject to all constraints of the examined problem.
Book of Proceedings
45
Step 2 (and 3). Min (or ) subject to all constraints of the examined problem and
*
1F F .
Final step (stability analysis). Max ib subject to all constraints of the examined problem and
*
1F F , *
2 , *
3 , where *F , * , and * are the optimal error values for the
basic MUSA model, the WORT model, and the desired properties of the produced results, and 1
2 , 3 are small percentages of *F , * , and * , respectively.
4. Real-World Application
4.1 Satisfaction Criteria and Survey Conduct
The above procedure has been implemented for the analysis of service quality in the Greek
mobile service industry. A sample of 80 questionnaires from the three mobile service providers of
Greece (Cosmote 57.5%, Vodafone 22.5%, Wind 20%) has been used. The collected data included
mostly customers’ judgments about the performance of the companies globally and partially (i.e.,
foe each criterion), as well as a ranking of these satisfaction criteria from the most to the least
important one. The selected satisfaction criteria included the following: offers, provided services,
provided devices, network, webpage, charges, branch network, and company image.
4.2 Results and Comparison
As it can been observed in Table 2, 1AFI and 3AFI are particularly high for the basic as well
as for the extension of the MUSA method. These indices are slightly worse for the extension of
the method which is quite expected as according to the heuristic method applied in the MOLP
problem, there is a small decrease of the optimal solution of the basic problem in order to achieve
a better consistency regarding the other two objective functions. 2AFI is particularly low in both
cases but as already mentioned this is a rather strict index. Regarding ASI , there is an significant
increase of 9.90% with the introduction of additional information and constraints in the basic
MUSA model.
1
AFI 2
AFI 3
AFI ASI
Original MUSA method 98.08% 17.50% 93.39% 79.11%
Proposed extension 94.58%
(‒0.53%)
7.50%
(‒57.14%)
91.92%
(‒2.25%)
86.94%
(+9.90%)
Table 2. Fitting and stability indices
The action diagram combines the relative importance and the performance of the criteria as
they have been estimated according to the MUSA and the extension of the MUSA methods.
Book of Proceedings
46
According to this diagram the company’s profile is regarded as their competitive advantage
considering that it has both the highest importance and satisfaction level. The devices provided by
the companies could potentially regarded as a competitive advantage while the rest of the criteria
have a relatively low satisfaction and importance level but could potentially be regarded as first
priority improvements if there is an increase of their importance. Generally, the conclusions
provided by both the basic MUSA and the extension of the MUSA method are similar proving that
the introduction of the additional constraints has improved the stability of the provided results
without however diversifying the conclusions and the strategic actions that the companies should
follow.
Figure 2. Action diagram
5. Concluding Remarks
The MUSA method is a rather flexible approach and thus several extensions may be developed
taking into account additional information or data. The implementation of the proposed extension
of the MUSA method in a real case study revealed that the stability of the provided results has
been improved. Furthermore, the introduction of additional constraints in the original MUSA
method has not diversified the strategic actions that the companies should follow, leading to more
stable conclusions. Finally, developing additional measures of robustness may facilitate the
investigation of various extensions of the MUSA model, while future research could include the
study of the impact of the model parameters or of different extensions of the MUSA method
through an extended simulation.
HighLow IMPORTANCE
High
Low
PERFORMANCE
Company's prof ile
Off res
Branch network
Provided devices
NetworkProvided services
Charges
Webpage
MUSAExtension
Book of Proceedings
47
Acknowledgement
This research has been co‐financed by the European Union (European Social Fund – ESF) and
Greek national funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework (NSRF) ‐ Research Funding Program: THALES.
Investing in knowledge society through the European Social Fund.
References
Grigoroudis, E. and O. Spiridaki. “Derived vs. stated importance in customer satisfaction surveys”.
Operational Research: An International Journal. Vol. 3 No. 3, 2003, pp. 229-247.
Grigoroudis, E. and Y. Siskos. “Preference disaggregation for measuring and analysing customer
satisfaction: The MUSA method”. European Journal of Operational Research, Vol. 143 No. 1,
2002, pp. 148-170.
Grigoroudis, E. and Y. Siskos (2010). Customer Satisfaction Evaluation: Methods for Measuring
and Implementing Service Quality. Springer, New York, 2010.
Siskos, J. “Analyses de régression et programmation linéaire”. Révue de Statistique Appliquée,
Vol. 23 No. 2, 1985, pp. 41-55.
Book of Proceedings
48
A combined MCDA approach for facilitating maritime transportation policies
evaluation
Eliza GAGATSI
Department of Civil Engineering, Aristotle University of Thessaloniki, University Campus 54124
Thessaloniki Greece, email: [email protected]
George GIANNOPOULOS
Hellenic Institute of Transport- Centre for Research and Technology Hellas, 6th km Charilaou
Thermi Rd, Thessaloniki, Greece
Christos PYRGIDIS
Department of Civil Engineering, Aristotle University of Thessaloniki, University Campus 54124
Thessaloniki Greece
Georgia AIFANDOPOULOU
Hellenic Institute of Transport- Centre for Research and Technology Hellas, 6th km Charilaou
Thermi Rd, Thessaloniki, Greece
Abstract
This paper presents a methodology developed at the frame of an on-going PhD research, for
supporting decision making in maritime transportation, based on a combination of two different
MCDA methods in a multi-actors environment. In the first part, it elaborates on the development
of the methodology with emphasis on the evaluation process and the operational synergy of the
two multi-criteria evaluation techniques exploited (PROMETHEE and AHP) and the key benefits
of this mixed approach on improving both methods applicability and limiting their deficiencies. A
stakeholders’ mechanism is exploited for the evaluation of the compared alternatives (policies) to
facilitate the consensus building among stakeholders with conflicting objectives through the
provision of a transparent policy selection process and to ensure the selection of realistic policy
measures evaluated under criteria that correspond to the actual needs of the relevant stakeholders.
In the second part the methodology is applied in a real case, namely to the evaluation of 4 existing
policy proposals aiming to support the Greek coastal sector. The key advantages of the combined
methodology along with the preliminary results and main messages from the real case application
are discussed in the concluding part of the paper.
Keywords: MCDA , Maritime Transport Policy, PROMETHEE, AHP, multi-actors evaluation
Book of Proceedings
49
1. Introduction
Policy making is a challenging and highly complex process. Decision making, as a process linking
policy formulation to the actual policy implementation, is characterized by a high level of
complexity and arduousness. Reaching the appropriate decision involves the optimization of a
multitude of parameters and a complex interplay of information, interests and opinions of a variety
of affected social groups (Hey et al, 1997). To support decision making, a variety of decision
supporting methods have been developed over the years suitable for various evaluation
environments. Examples of some well know methods are those of Cost-Benefit Analysis (CBA),
Cost-effectiveness analysis (CEA), Economic Effects Analysis (EAA), balance sheets,
Multicriteria analysis (MCA) etc.
In the particular sector of maritime transportation- that is examined at the frame of this paper-,
policy is made and implemented under conditions of multiple objectives (deriving from the variety
of involved stakeholders) and constraints (Frankel, 1992) in an environment characterized by
strong complexity in the relations between jurisdictions, administrators, politicians and the
industry (Roe, 2009).
Being a strong analytic tool that supports decision making in an under uncertainty environment
and at can facilitate the building of consensus among all involved actors in a well explicit way,
MCDA became over the years a popular evaluation method with numerous recorded applications
in complex problems. The lack of one single & central goal, common to all integrated policies–
such as those in maritime transport, combined with a great number of impacts that cannot always
be monetised, regards MCDA methods more suitable for supporting policy evaluation than other
widely used financial-economic evaluation methods (eg CEA,CBA) that fail to capture the holistic
view of a problem the necessary incorporation both tangible & intangible (or ‘fuzzier’) aspects.
The most popular among the various techniques to conduct a MCDA, applied in the field of
transport are multi-attribute theory variants (AHP, MAUT, MAVT, SMART, SMARTER),
outranking methods (PROMETHEE, ELECTRE) & regime analysis. The selection of the
appropriate evaluation methodology is crucial and needs to be carefully examined in relation to
the problem particularities, needs and constraints (eg in terms of time, available data etc) ; using
different methods can sometimes even lead to divergent results especially when, as Finco and
Nijkamp highlights (Finco&Nijkamp,1997), a complete ranking of the under examination
alternatives is needed. Recently, the combination of various MCDA methods starts gaining ground
facilitated by the advancing technologies that ease their use. Applying a multi-method approach
can facilitate policy making by reviewing preferences and judgments derived from more than one
MCA method (Mysiak, 2006).
Τhis paper proposes an operational synergy of two MCDA methods to facilitate policy making in
the area of maritime transportation. The proposed combination of PROMETHEE & AHP seeks to
improve both methods’ applicability, decreasing their deficiencies, while the application in a
multi-actors environment facilitates the building of consensus among actors through the provision
Book of Proceedings
50
of a transparent participatory policy selection mechanism. A real case application of the
methodology is discussed in the second part of the paper along with some preliminary results and
key messages on the methodology applicability.
2. The methodological Framework
The evaluation mechanism that is presented in the following is based on a multi-method approach
applied in a multi-actors environment and is structured around 3 building blocks, namely, the
stakeholders’ analysis, the combination of two MCDA methodologies (PROMETHEE & AHP)
and the exploitation of two independent mechanisms in the evaluation process, namely the experts’
and the stakeholders’ group. The methodological approach is depicted in the next figure while its’
main building blocks are described in the following paragraphs:
Figure 1: Methodological Framework
The Stakeholders’ Analysis (SA): The term ‘stakeholders’ refers to people or groups who have an
interest, financial or otherwise, in the consequences of any decision taken (Macharis, 2011) or
any policy selected in the specific case. In the proposed evaluation methodology, the SA is used
as “an aid to properly identify the range of stakeholders which needs to be consulted and views
should be taken into account in the evaluation process” (Macharis et al, 2012). More specifically
it facilitates:
the evaluation matrix structuring, through the preliminary identification of the evaluation
criteria (representing the stakeholders’ priorities) and of the evaluation alternatives which
refer to key policy proposals deriving from the various stakeholders
the definition of the stakeholders group which comprises the evaluating body as described
below
Combination of 2 MCDA methods: the methodology proposes a combination of 2 MCDA methods
namely the AHP and the PROMETHEE methods, an operational synergy that aims at reducing
both methods weaknesses and enhancing their strengths.
Book of Proceedings
51
AHP supports the design of the decision making hierarchy and the definition of the criteria weights,
eliminating the main disadvantage of PROMETHEE, the lack of a structured way of defining the
criteria weights. The evaluation methodology was initially structured under the AHP logic in a
multi-actors environment. The ease of applicability and the structure of AHP, which follows the
intuitive way in which managers solve problems (Ishizaka A& Labib A(2009) were among the
reasons for examining AHP applicability in the particular research. However the assumption of
the structural elements independence that is prerequisite in AHP, turned to be a critical limitation
not easy to be secured in a complex environment such as this of maritime transportation. The
Analytic Network Process, proposed by Saaty (Saaty, 2008) for overcoming AHP limitations by
handling interdependence among elements through its ‘supermatrix’ was examined as an
alternative but also considered not suitable for the examined case. Its application led to a pretty
complex ‘network’ of elements with increased data requirements (resulting from the large number
of pairwise comparisons required to capture the interdependencies and the among the network
elements relations) difficult to be collected in the particular case which involves many actors
coming from different environments that are usually not familiarized such methods.
PROMETHEE on the other hand was considered more advantageous compared to both methods
since it helped overcoming the interdependencies requirement (AHP) facilitating at the same time
the various stakeholders in the alternatives evaluation through a more simplified and easy to use
evaluation matrix (compared to the numerous pairwise comparisons required to ‘solve’ the ANP
network).
Both methods support the group level decision making process. AHP exploits the geographical
mean of the individual pairwise comparisons (Zahir 1999) while in the case of PROMETHEE, that
is used in this case, the final evaluation results from the calculation of the weighted sum of the
individual net flows (Figure 2).
Tools and supporting mechanisms: the multi-actors methodology relies on the exploitation of 2
independent mechanisms in the evaluation process. In particular, an experts’ group is consulted
for the selection of the evaluation criteria (through a DELPHI process) and a stakeholders’ group,
defined through the SA supports the evaluation by providing the final ratings of the alternatives vs
the evaluation criteria. Both groups are also providing weights to the criteria (by an AHP
application), supporting the examination (by means of the PROMETHEE method) of different
scenarios leading to useful conclusions and results.
3. Application of the proposed methodology
The combined MCDA methodology is applied for the evaluation of 4 policies aiming to support
the Hellenic coastal transportation sector. The evaluation case is described below focusing on the
problem structuring, the data collection and the data analysis procedures.
3.1 Structuring of the evaluation problem
The first step towards the evaluation problem structuring was the stakeholders’ analysis. Following
the SA, 6 evaluation criteria were identified, based on the analysis of the key stakeholders’ group
priorities as identified following a bibliographical review. The proposed evaluation criteria were
Book of Proceedings
52
examined by an 8-member experts’ group (coming from the academic and research community)
following an application of a- two round Delphi method. This process led to the development of a
4×4 evaluation matrix.
Following the PROMETHEE requirements, for each criterion a preference function (PF) has to
be defined. The PF is used to ‘translate’ the difference between the evaluations obtained by two
examined alternatives into a preference degree that ranges from zero to one. According to Vincke
and Brans (1985) there are 6 basic types of PF: (1) usual criterion, (2) U-shape criterion, (3) V-
shape criterion, (4) level criterion, (5) V-shape with indifference criterion and (6) Gaussian
criterion. In the examined case all 4 criteria are qualitative; also, in the evaluation matrix, a small
number levels on the criteria scale (5-point scale) is used for all 4 criteria. Based on the above
properties of the evaluation criteria, the Usual (type I)PF was selected (Deshmukh S.C. (2013)).
Alternatives/criteria C1:
Transport
Cost
C2: Trip
cost(cost
to the user)
C3:
Subsidy
Cost
C4:
Level of
Service
Preference function/
Min/max
Usual
min
Usual
min
Usual
min
Usual
max
Unit/measurement type 1-5 Likert scale
Weights S1(Scenario 1= equal weights senario) w1=w2=w3=w4
Weights S2(Scenario 2- Stakeholders group) w1(s1),w2(s1),w3(s1),w4(s1)
Weights S3(Scenario 3- Experts group) w1(s2),w2(s2),w3(s2),w4(s2)
A1: Re-design of the national ferry network under the logic of a
network composed by several Hub ports&many peripheral ports
around them (Hub & Spoke)
Score 11 Score 21 Score 31 Score 41
A2: Reduction of VAT and other non-remunerative taxes in
passenger and vehicle fares Score 12 Score 22 Score 32 Score 42
A3: Application of less strict coastal fleet manning regulations
through measures Score 13 Score 23 Score 33 Score 43
A4: Application of ‘Road Equivalent Tariff- RET’ methodology on
the un-profitable lines network. Score 14 Score 24 Score 34 Score 44
Table 2: Evaluation matrix
3.2 Data collection
The data collection was based on a survey to the stakeholders’ group comprising of 4 categories
namely: shipping lines, ports, users’ representatives and labor representatives. The survey sample
is presented in the next table.
Stakeholder category Sample No
Shipping lines 4 Associations ,27 individual companies 31
Ports 1 Association , 22 ports/port authorities 23
Users representatives 5 National/regional Associations, 18 local representatives 23
Labor representatives 1 National Association 1
TOTAL 78
Table 3: Survey sample synthesis (stakeholders group)
A dedicated questionnaire was developed based on the evaluation matrix presented in the previous
session. Further to the alternatives evaluation towards the 4 criteria, the stakeholders were asked
to allocate weights to the evaluation criteria through a set of 6 pairwise comparisons (AHP).
Further to the pair comparison, a direct (proportional) allocation of weights was requested as an
Book of Proceedings
53
internal inconsistency control mechanism. The personal (structured) interview method was used
for the data collection along with the on-line survey option.
3.2 Data analysis
For the data analysis, the Visual PROMETHEE software was exploited under a GDSS function.
The process for data entry of the group decision problem is depicted in the next figure. For each
stakeholder, an individual PROMETHEE based evaluation takes place based on his/hers
evaluation table and relevant weights allocated to each evaluation criterion. For each action, the
positive, negative and net flows are calculated. The positive and negative flows represent the
intensity with which one action is preferred ore overcome by the others respectively, while the net
flows (φi(ai)) represents the balance between the positive and negative flows. The net flows of
each evaluation are used to formulate the evaluation matrix of each Stakeholder group (ie SG1:
Shipping Lines, SG2: ports etc). The net flows of each group are again calculated and entered into
the final evaluation matrix of the group decision problem (figure 2). The Decision maker can
provide weights on each group. The examined case was performed for an equal based weights
scenario.
The above process was repeated for three scenarios; in the first scenario all criteria are given the
same weights while in the second and third scenario, the evaluation is based on the weights
allocated to the criteria by the experts’ group and the stakeholders group as described above
Evaluati
on
matrix of
Stakehol
der 1
φ11
φ12
φ13
φ14
Evaluati
on
matrix of
Stakehol
der 2
φ21
φ22
φ23
φ24
Evaluatio
n matrix
of
Stakehol
der n
φν1
φν2
φν3
φν4
φ11
φ12
φ13
φ14
φ21
φ22
φ23
φ24
φν1
φν2
φν3
φν4
φ..1
φ..2
φ..3
φ..4
SGI
evaluat
ion
matrix
φ(G1)1
φ(G1)2
φ(G1)3
φ(G1)4
Stakeholders Group I
φ(G1)1
φ(G1)2
φ(G1)3
φ(G1)4
φ(G2)1
φ(G2)2
φ(G2)3
φ(G2)4
φ(Gν)1
φ(Gν)2
φ(Gν)3
φ(Gν)4
φ(G..)1
φ(G..)2
φ(G..)3
φ(G..)4
wΟ1,wΟ2….wΟν
Evaluation matrix- Stakeholders Group
Weights
(of the various groups)
….
Evaluati
on
matrix of
Stakehol
der 1
φ11
φ12
φ13
φ14
Evaluati
on
matrix of
Stakehol
der 2
φ21
φ22
φ23
φ24
Evaluatio
n matrix
of
Stakehol
der n
φν1
φν2
φν3
φν4
φ11
φ12
φ13
φ14
φ21
φ22
φ23
φ24
φν1
φν2
φν3
φν4
φ..1
φ..2
φ..3
φ..4
SGII
evaluat
ion
matrix
φ(G2)1
φ(G2)2
φ(G2)3
φ(G2)4
Stakeholders Group II
Evaluati
on
matrix of
Stakehol
der 1
φ11
φ12
φ13
φ14
Evaluati
on
matrix of
Stakehol
der 2
φ21
φ22
φ23
φ24
Evaluatio
n matrix
of
Stakehol
der n
φν1
φν2
φν3
φν4
φ11
φ12
φ13
φ14
φ21
φ22
φ23
φ24
φν1
φν2
φν3
φν4
φ..1
φ..2
φ..3
φ..4
SGn
evaluat
ion
matrix
φ(Gn)1
φ(Gn)2
φ(Gn)3
φ(Gn)4
Stakeholders Group n
Figure 2: Application of the PROMETHEE GDSS
4. Conclusions
The policy making process is a highly complex process, requiring not only in-depth knowledge
of the sector, but also the employment of a methodology that can facilitate the identification of
alternative policy measures and the selection of the most appropriate ones, under conditions of
multiple objectives, through a consensus building mechanism. The present paper presents such a
Book of Proceedings
54
methodology based on a combination of two multicriteria methods namely AHP and
PROMETHEE under a multi-actors evaluation environment. The methodology is designed to
support policy making in maritime transportation, a particular complex and important for the
economy and employment sector that is characterised by strong power of the involved stakeholders
that directly or indirectly can affect policy making and implementation.
The methodology proposed is built upon three main elements, the Stakeholders Analysis, the
operational synergy of two MCDA method and the exploitation of 2 independent mechanisms in
the evaluation process. The combination of the above elements provides key advantages in the
policy evaluation process:
stakeholders’ & independent experts' involvement throughout the process, ensures the
identification of policy measures that are realistic and the employment of assessment
criteria that correspond to their actual needs.
a major part of the vagueness usually characterizing policy formulation is removed as the
method provides a structured, step-wise approach for identifying and selecting policy
measures based on transparent and easy to use (ie PROMETHEE) methods
it facilitates consensus building among stakeholders with conflicting objectives, as it
provides a transparent process for commonly reaching conclusions on the policy measures
to be employed
the stakeholders participation improves the ‘ownership’ of results leading to the
stakeholders engagement, necessary component for a successful policy implementation
The application of the methodology to the evaluation of 4 policies aiming to support the viability
of the Greek coastal transportation that is currently on-going , confirms the applicability of the
proposed mechanism in the selected case highlighting however some difficulties in the application
of AHP in the stakeholders environment.
ACKNOWLEDGEMENT
This research has been co-financed by the European Union (European Social Fund-NSF) & Greek
national funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework(NSRF)-Research Funding Program: Heracleitus II.
Investing in knowledge society through the European Social Fund.
References
Bakker P, Koopmans, C. &Nijkamp, P. (2009). Appraisal of integrated transport policies. VU
University, Serie Research Memoranda (No 0052).
Deshmukh S.C. (2013) Preference Ranking Organization Method Of Enrichment Evaluation
(Promethee), International Journal of Engineering Science Invention ISSN (Online): 2319 –
6734, Volume 2 Issue 11, November2013,PP.28-341,S.C.Deshmukh
Finco, A. & Nijkamp, P. (1997). Sustainable land use: methodology and application. Research
Memorandum 1997-64
Book of Proceedings
55
Frankel G E (1992). Hierarcical Logic In Shipping and Decision Making. Maritime Policy
Management, Vol 3, 211-221.
Hey C, Nijkamp P, Rienstra S& Rothenberger D, (1997),Assessing Scenarios on European
Transport Policies by Means of Multicriteria Analysis, ECON Papers, 97-086 III, Tinbergen
Institute Discussion Papers
Ishizaka A& Labib A(2009) Analytic Hierarchy Process and Expert Choice: Benefits and
limitations, OR Insight (2009) 22, 201–220. doi:10.1057/ori.2009.10
Macharis C& Nijkamp P (2011). Possible bias in multi-actor multi-criteria transportation
evaluation: Issues and solutions. Research Memorandum 2011-31.
Macharis C, Turcksin L (2012). Multi actor multi criteria analysis (MAMCA) as a tool to
support sustainable decisions:State of use. Decision Support Systems(54), pp. 610-620.
Mysiak J. (2006). Consistency of the results of different MCA methods: a critical review.
Environment and Planning C: Government and Policy 2006, Volume 24, pages 257-277.
Roe, M. (2009). Maritime governance and policy-making failure in the European Union. Int. J.
of Shipping and Transport Logistics, 2009 Vol.1, No.1, pp.1 – 19
Saaty,T.(2008). Decision making with the analytic hierarchy process. Int. J. of Services Sciences,
1, pp. 83-98
Book of Proceedings
56
Optimal Strategic Design of Flexible Supply Chain Networks
Magdalini A. Kalaitzidou
Aristotle University of Thessaloniki, University Campus Thessaloniki, 54124 Thessaloniki,
Greece, email: [email protected]
Pantelis Longinidis
University of Western Macedonia, Karamanli & Lygeris Street, 50100, Kozani, Greece.
Panagiotis Tsiakis
Wipro Consulting Services, 3 Sheldon Square, London W2 6PS, United Kingdom.
Michael C. Georgiadis
Aristotle University of Thessaloniki, University Campus Thessaloniki, 54124 Thessaloniki,
Greece.
Abstract
This paper presents a mathematical programming model for the optimal design of Generalized
Supply Chain Networks (GSCNs) that incorporates strategic flexibility in network’s configuration.
The model is formulated as a deterministic Mixed-Integer Linear Programming (MILP) problem
and solved to global optimality using standard branch-and-bound techniques. Optimality is
assessed in terms of SCN’s overall cost while its applicability, benefits, and robustness are
illustrated by using a real case study.
Keywords: Supply chain network design, Generalized nodes, MILP, Deterministic.
1. Introduction
In recent years, the problem of designing the SCN has gained much interest from business as its
contribution to sustainable competitive advantages is universally acknowledged. Facility location
is the core decision within strategic design of SCNs. According to Drezner and Hamacher (2004)
facility location problems involve a set of spatially distributed customers whose location is known
and a set of facilities to satisfy their demands, whose locations are to be determined. Melo et al.
(2009) conduct a remarkable review on facility location models and demonstrate how their
characteristics affect strategic SCN management. Likewise, Melo et al. (2006) revealed how the
structure of the network is strongly affected by external supply of materials, inventory
opportunities, storage limitations, relocation, expansion or reduction of capacities. Thanh et al.
(2008) presented a dynamic model in the design of production-distribution system by making
strategic decisions: supplier’s selection; opening-closing facilities etc. over a planning period. In
a very recently work, Cardoso et al. (2013) proposed a model for the design and planning of SCNs
with reverse flows and demand uncertainty.
Book of Proceedings
57
The vast majority of the relevant works on the research stream of SCN design assumes a structure
of the network with distinct and consecutive echelons, consisting of nodes with predetermined
function, where product flow moves from an echelon’s nodes to subsequent echelon’s nodes. The
aim of this paper is to introduce a flexible composition to network’s structure, as the function of
the proposed generalized nodes are optimally defined rather than selected from a set of potential
alternatives. Moreover, intra-layer material flow connection is permitted among these generalized
nodes. To the best of our knowledge, this is the first work that provides such a flexibility option to
SCN.
2. Mathematical formulation
2.1 Problem description
This work addresses the design of a multi-product, multi-echelon SCN. The model proposes an
innovative configuration to network’s structure by entering a level consisted of generalized
production/warehousing nodes (P/W) whose function is not a priori assumed, as in mainstream
fixed echelon SCNs. These nodes can receive material from any potential supplier or any other
P/W node and deliver material to any customer zone or any other P/W node, as shown in Figure
1.
Figure 3 The proposed GSCN structure (a) against the typical fixed echelon SCN structure (b).
We denote the set of all nodes in the network as n∈N. This includes not only the generalized nodes
n∈N P/W but also suppliers nodes n∈N S and customer zones nodes n∈N C. Overall we have
N=S∪P/W∪C. The objective is to minimize the overall capital and operational cost and determine
the optimal structure of the network. The model defines: (i) suppliers; (ii) generalized node’s
location and role; (iii) material flow among SCN’s levels; and (iv) functional elements (capacity,
material flow, purchases etc.).
2.2 Mathematical model
A deterministic MILP model is formulated where each product can be produced at several
generalized P/W nodes in different location with known and time-invariant product demand
(Kalaitzidou et al., 2014). All transportation flows determined are considered to be time-averaged
Suppliers
(fixed
location)
Generalized P/W nodes
(location & function to be selected)
Customer zones
(fixed location)
??
?
?
Suppliers
(fixed
location)
Plants
(only location to
be selected &
no vertical flows
allowed)
Customer zones
(fixed location)
Warehouses
(only location to
be selected &
no vertical flows
allowed)
a b
Book of Proceedings
58
quantities whereas customer zones are single sourced. The objective is to minimize the overall
capital and operational cost of the network and is as follows: min∑ {Cn
PYnP + ∑ Eenγen
Pe + Cn
WYnW + γn
WWn +∑ CinWH
i (∑ Qin′nn′∈S∪P/W +n∈P/W
∑ Qinn′n′∈C∪P/W ) + ∑ δenP ∑ λekk∈Kn ξkne + ∑ (∑ Cin′n
T Qin′nn′∈S + ∑ Cinn′T Qinn′n′∈P/W +i
∑ Cinn′T Qinn′n′∈C )} + ∑ (Cn
SYnS + ∑ Cin
S Sin) in∈ S
Capital cost is consists of infrastructure cost whereas handling, production, transportation and
purchasing contribute to operational cost. Infrastructure cost is related to the establishment of a
warehouse or a production plant at a particular node n∈P/W. If a production capability is
established at a node n∈P/W then its infrastructure cost has a stable element (CnPYnP) and a variable
element (∑ EenγenP )e . The former is the product of the annualized fixed cost required to establish
a production capability (CnP) with the binary variable that expresses the establishment of this
capability (YnP). The latter element is the sum of the products of the continuous variable expressing
the total rate of availability of manufacturing resource e (Een). Similarly, if a warehousing
capability is established at node n∈P/W then its infrastructure cost has a stable element (CnWYn
W),
the product of the annualized fixed cost required to establish a warehousing capability (CnW) with
the binary variable that express the establishment of this capability (YnW), and a variable element
(γnWWn), the product of a coefficient expressing the unit cost associated with the warehousing
capacity (γnW) with the continuous variable expressing the warehousing capacity (Wn).
Regarding operational cost, handling cost is expressed as a linear function of the total throughput
at node n∈P/W. By multiplying the total throughputs with the unit handling cost for material i
(CinWH) and summarising the resulting products we gain the handling cost. Production cost is related
to the utilization of various resources e at node n∈P/W and is determined as the sum of the products
of the unit cost of consumption of resource e at node n∈P/W (δenP ) with the total utilization of each
resource e (∑ λekk∈Kn ξkn). Utilization is the product of the amount of manufacturing resource e
required to perform unit amount of task k (λek) and the continuous variable expressing the rate of
operation of task k at node n∈P/W (ξkn). Transportation cost is decomposed into three terms each
of which sums the products of unit transportation cost of material i from a node n to another node
n' (Cin′nT ), and vise versa (Cinn′
T ), and the corresponding continuous variables expressing the rate
of flow of material i that arrives at a node n from another node n' (Qin′n) and vise versa (Qinn′).
The first term expresess transportation cost of material i transferred from node n'∈S to node n∈P/W
(∑ Cin′nT Qin′nn′∈S ), the second term expresses the transportation cost of material i transferred from
node n∈P/W to other node n'∈P/W (∑ Cinn′T Qinn′n′∈P/W ), and the third term expresses the
transportation cost of material i transferred from node n∈P/W to node n'∈C (∑ Cinn′T Qinn′n′∈C ).
Finally, purchasing cost has a stable element (CnSYnS), the product of the annualized fixed cost of
establishing a relationship with node n∈S (CnS), and the binary variable that expresses the selection
of node n∈S as a material provider in the network (YnS), and a variable element (Cin
S Sin), the
product of the unit purchase price of material i from node n∈S (CinS ) and the continuous variable
expressing the purchased amounts of material i from the selected node n∈S (Sin). By summarising
this variable element for all materials we gain the purchasing cost from each node n∈S and then
Book of Proceedings
59
by adding the resulting variable element with the stable element and summarizing for all suppliers
we reach purchasing cost.
The MILP optimization model has six sets of constraints that formulate the structure of the
network, the flow of materials within the network, the core operations in the network (purchasing,
production, and warehousing), and customer satisfaction. Constraints (1) and (2) demonstrate the
conditions for the establishment of a node n∈P/W. In specific, constraint (1) states that if a
production capability is established at a node n∈P/W (YnP = 1) then the corresponding node
n∈P/W should be established as the binary variable that expresses its establishment is forced to
take the value of one (Yn = 1). In the same fashion, constraint (2) states the conditions for the
establishment of a warehousing capability at node n∈P/W (YnW = 1).
Yn ≥ YnP, ∀ n ∈ P/W (1)
Yn ≥ YnW, ∀ n ∈ P/W (2)
If a node n∈P/W is established (Yn = 1) it should receive material from at least one other node
n'∈S∪P/W and should provide material to at least one other node n'∈P/W∪C. As shown in
constraint (3), if a node n∈P/W is established (Yn = 1) the binary variable that expresses the
establishment of a material transportation link (Xn′n) is forced to take the value one for at least
one pair of n'∈S∪P/W with n∈P/W and provided that n≠n'. In the same manner, constraint (4)
shows that if a node n∈P/W is established (Yn = 1) the binary variable that expresses the
establishment of a material transportation link (Xnn′) is forced to take the value one for at least
one pair of n∈P/W with n'∈P/W∪C and provided that n≠n'.
Yn ≤ ∑ Xn′nn′∈S∪P/W\{n}
, ∀ n ∈ P/W (3)
Yn ≤ ∑ Xnn′
n′∈C∪P/W\{n}
, ∀ n ∈ P/W (4)
A connection between a node n'∈S and a node n∈P/W can exist only if both the supplier is
contracted and the generalized node is established. Constraint (5) forces the binary variable
expressing the contracting of node n'∈S (Yn′S ) to be unity when the material transportation link,
between a node n'∈S and a node n∈P/W, is established (Xn′n = 1). On the other hand, constraint
(6) forces the binary variable expressing the establishment of node n∈P/W to be unity when the
material transportation link, between a node n'∈S and a node n∈P/W, is established (Xn′n = 1).
Xn′n ≤ Yn′S , n′ ∈ S, n ∈ P/W,n ≠ n′ (5)
Xn′n ≤ Yn, n′ ∈ S, n ∈ P/W,n ≠ n′ (6)
Similarly, a connection between two nodes n∈P/W and n'∈P/W can exist only if both nodes are
established. Constraints (7) and (8) stress this condition while constraint (9) requires the
establishment of node n∈P/W if it is going to transfer material to node n'∈C.
Xnn′ ≤ Yn, ∀ n ∈ P/W,∀ n′ ∈ P/W , n ≠ n′ (7)
Xnn′ ≤ Yn′ , ∀ n ∈ P/W,∀ n′ ∈ P/W , n ≠ n′ (8)
Xnn′ ≤ Yn , ∀ n ∈ P/W,∀ n′ ∈ C , n ≠ n′ (9)
Book of Proceedings
60
As the model does not allow reverse flows, intra-layer flows between suppliers and customer
zones, and direct flows from suppliers to customer zones appropriate fixing to zero takes place for
the binary variables (Xnn′) expressing the establishment of the above prohibited transportation
links. Moreover, the flow of materials (Qin′n) lies between upper and lower bounds provided that
the corresponding transportation connection has been established. In nodes n∈P/W where
production capability is established the overall balance for the production of material i is the inflow
(∑ Qin′nn′∈S∪P/W\{n} ) minus the outflow (∑ Qinn′n′∈C∪P/W\{n} ) of material i plus the rate of
production of material i at that node, as shown in constraint (10).
∑ Qin′nn′∈S∪P/W\{n}
+ ∑ vikk∈Kn
ξkn = ∑ Qinn′
n′∈C∪P/W\{n}
, ∀ i, n ∈ P/W (10)
The term (vik) expresses the amount of material i produced by unit amount of task k and multiplied
with the continuous variable expressing the rate of operation of task k at node n∈P/W (ξkn) we
have the rate of production. The total utilization of each resource e (∑ λekk∈Kn ξkn) is limited to
the total rate of availability of resource e at node n∈P/W (Een) as shown in constraint (11).
∑ λekk∈Kn
ξkn ≤ Een , ∀ e, n ∈ P/W (11)
Upper and lower bound are imposed for both the total rate of availability of resource e at node
n∈P/W (Een) and purchased amounts of material i from the selected node n∈S (Sin). Additionally,
appropriate constraints force the model to transfer all purchased material to generalized nodes and
also to satisfy all demand. Finally, warehousing capacity (Wn) lies between higher and lower
limits, provided that warehousing capability is established while it is approached as linear function
of handled material flow as shown in constraint (12) with (ainin)/(ain
out) expressing the relationship
between capacity of warehouse at node n∈P/W, to material i handled that enters/leaves the node.
Wn ≥ ∑ Qin′ni,n′∈S∪P/W
ainin + ∑ Qinn′
i,n′∈C∪P/W
ainout, ∀ n ∈ P/W , n ≠ n′ (12)
3. Case study
The applicability of the GSCN design and operation model is illustrated by using a real case study
in the European area (Tsiakis et al., 2001). This study is being held for the interests of a European
company. This network is comprised by total thirty-eight nodes, whose locations are sited among
the European area. More specific there are, five potential suppliers, fifteen potential production
plants/warehouses and eighteen customer zones. The number of materials/products provided by
the suppliers is fourteen.
4. Results
The proposed model (GSCN) is compared with a counterpart model with fixed-echelons (FSCN),
both of which were implemented in GAMS 24.1.3 software, using CPLEX 12 solver. Identical
data were used for both models and production process is approached in the same way. Figure 2,
present the optimal network for GSCN and FSCN, respectively. The former establishes 3 P/W
nodes with both capabilities in location countries: ES, IT, BE and 1 P/W node with only
Book of Proceedings
61
warehousing capability (CH), all of which are provisioned from 2 suppliers (BG, RO), while the
latter establishes 3 plants (PT, BE, CH) and 5 warehouses (IT, DK, PT, BE, CH) all of which are
provisioned from the same suppliers.
The GSCN model shows its superiority firstly from the objective function, and secondly from the
network’s flexibility. Both models employ the same objective function that was counted 742,016
and 1,051,214 relative money units (rmu) for GSCN and FSCN, respectively. This cost gap, is due
to the fact that FSCN model is forced by the a priori structure to build more facilities (sum of plants
& warehouses) in order to satisfy customer demand. Furthermore, FSCN’s model ends up in a
structure where the more sizeable material flow connections are among the plants and warehouses
that are located and build at the same country-area. This fact shows the necessity of a generalized
node with both warehousing/manufacturing capabilities. Additionally, a sensitivity analysis was
performed and the outcome revealed that the GSCN model reacts fairly enough in demand changes
and is insensitive to all other parameters.
Figure 2 Optimal GSCN configuration (a) against the optimal FSCN configuration (b).
5. Results
This paper introduces a mathematical model that provides flexibility options on designing and
operating SCNs. The model is capable of deciding the appropriate suppliers and material flow
connections including intra-layer flows but mainly the location and role/capability of the
generalized nodes. It is concluded that, giving the network the option to have nodes that act with
both manufacturing and warehousing capability (or choose among them) and simultaneously to
avoid having separated manufacturing and warehousing layers, minimizes the overall cost, but
mostly benefits in material handling cost.
6. Acknowledgement
This research has been co-financed by the European Union (European Social Fund – ESF) and
Greek national funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework (NSRF) - Research Funding Program: Thales. Investing
in knowledge society through the European Social Fund.
ES
GR
BE
FR
PL
NO
PT
TR
IT
UK
FI
SE
NL
DE
CH
AT
IE
DK
BG
RO
ES
ES
IT
IT
BE
BE
CH
4046
1317
1520
6051
10724
1892
2204
10186
11237
898
2133
1892
190
32653
32453
3807
6746
7095
61675550
32867
3987
3643
67834426
4230
5805 60746450
7421
Supplier
Warehousing cap.
Manufacturing cap.
Flow from supplier to P/W
Flow from P/W to P/W
Flow from P/W to customer zone
Customer zone
ES
GR
BE
FR
PL
NO
PT
TR
IT
UK
FI
SE
NL
DE
CH
AT
IE
DK
BG
RO
PTPT
CHCH
BEBE
20950
2559
379
1533
Supplier
Warehouse
Plant
Forward flow between echelons
Customer zone IT
DK
19659
616
34522
1701
10881
53755
28694
10862
5225
33942
3807
3643
32653
32453
7095
5550
67464426
6167
6450
4230
7421
328676783
4046
3985
5805
6074
a b
Book of Proceedings
62
References
Cardoso S.R., Barbosa-Póvoa A.P.F.D., and Relvas S. “Design and planning of supply chains with
integration of reverse logistics activities under demand uncertainty”. European Journal of
Operational Research, Vol. 226, 2013, pp.436-451.
Drezner Z., Hamacher H.W. “Facility location: Applications and Theory”, Springer, New
York. 2004.
Kalaitzidou, M. A., Longinidis, P., Tsiakis, P., and Georgiadis, M. C. “Optimal Design of
Multiechelon Supply Chain Networks with Generalized Production and Warehousing Nodes”.
Industrial & Engineering Chemistry Research, Vol. 53, 2014, pp. 13125-13138.
Melo M.T., Nickel S., and Saldanha-Da-Gama F. “Dynamic multi-commodity capacitated facility
location: A mathematical modeling framework for strategic supply chain planning”. Computers
and Operations Research, Vol. 33, 2006, pp. 181-208.
Melo M.T., Nickel S., and Saldanha-Da-Gama F. “Facility location and supply chain management
– A review”. European Journal of Operational Research, Vol. 196, 2009, pp. 401-412.
Thanh P.N., Bostel N., and Péton O. “A dynamic model for facility location in the design of
complex supply chains”. International Journal of Production Economics, Vol. 113, 2008, pp. 678-
693.
Tsiakis, P., Shah, N., and Pantelides C. C. “Design of multi-echelon supply chain networks under
demand uncertainty”. Industrial & Engineering Chemistry Research, Vol. 40, 2001, pp. 3585-
3604.
Book of Proceedings
63
An Integrated Multi-Regional Long-Term Energy Planning Model
Incorporating Autonomous Power Systems
Nikolaos E. Koltsaklis
Department of Chemical Engineering, Aristotle University of Thessaloniki, University Campus,
54124 Thessaloniki, Greece email: [email protected]
Pei Liu
Department of Thermal Engineering, Tsinghua University, State Key Laboratory of Power
Systems, 100084 Beijing, China.
Michael C. Georgiadis
Department of Chemical Engineering, Aristotle University of Thessaloniki, University Campus,
54124 Thessaloniki, Greece.
Abstract
This paper addresses the long-term generation expansion planning (GEP) problem of a large-scale,
central power system incorporating the possible interconnection with various autonomous power
systems. A multi-regional, multi-period, Mixed Integer Linear Programming (MILP) model was
developed to determine the optimal power capacity additions per time interval and region and the
power generation mix per technology and time period. The model is tested on the Greek power
system taking also into consideration the scheduled interconnection of the mainland power system
with those of some autonomous islands (Cyclades and Crete), and aims at providing full insight
into the composition of the long-term energy roadmap at a national level.
Keywords: MILP, GEP, Autonomous power systems’ interconnection, Renewable energy sources,
Power sector.
Introduction
For decades, strategic long-term planning of a power system has been focused on guaranteeing
security and quality of supply, reduction of dependence from imported fuels and power grid’s
stability and reliability. Nowadays, deregulation of electricity markets along with the introduction
of environmental issues have made this task more complicated, requiring a great deal of parameters
to be taken into consideration. Long-term GEP determines the optimal type of energy technologies
to be installed, the optimal electricity production mix per time period, as well as the location and
time construction of the newly built units.
There has been extensive research regarding the GEP problem of the Greek power system. Agoris
et al. (2004) presented an analysis in order for the Greek power system to achieve the Kyoto targets
using the R-MARKAL and WASP IV models. Dagoumas et al. (2007) examined the evolution of
Book of Proceedings
64
the Greek power sector up to 2020 by developing several scenarios concerning the implementation
of a post-Kyoto target. Rampidis et al. (2010) simulated the Greek power sector using the module
BALANCE of the Energy and Power Evaluation Program (ENPEP) in order to investigate the
feasibility of a variety of investment plants announced by power companies. Roinioti et al. (2012)
presents an analysis of the Greek energy system for the period between 2009 and 2030 using the
Long range Energy Alternatives Planning (LEAP) modeling tool.
Our work constitutes an extension of our previous approach (Koltsaklis et al., 2014) on the grounds
that: (i) incorporates possible interconnection with autonomous power systems, (ii) gives the
option of electricity (imports and exports) and CO2 emissions (purchases and sales) trade to the
model, (iii) provides a more detailed representation of units’ operation (by dividing total capacity
of each unit into blocks with specific technical characteristics), and (iv) includes an annual budget
constraint for possible limitations on the total amount of money to be allocated.
1. Problem statement and mathematical formulation
The problem under consideration is formally defined in terms of the following items: A given
planning horizon is divided into a set of uniform time periods 𝑡 ∈ 𝑇. Electricity demand can be
described by means of a load duration curve. In our work, eight load blocks 𝑏 ∈ 𝐵 are taken into
account, each of which with specific load duration 𝑑𝑢𝑟𝑏 . The overall power system is divided into
a number of sectors 𝑠 ∈ 𝑆, each of which is characterized by a specific electricity demand 𝐷𝑒𝑚𝑠,𝑏,𝑡,
existing initial capacity 𝑃𝑚,𝑡𝑚𝑎𝑥, as well as interconnection parameters (e.g., injection losses 𝐼𝑁𝐽𝑠,𝑏,𝑡
and Load Loss factor 𝐿𝐿𝐹𝑠,𝑏,𝑡). There is also the option in the model to interchange (imports and
exports) electricity with other neighboring grids/countries, offering more flexibility to the
operational planning of the network. A set of power generation technologies 𝑚 ∈ 𝑀 is available
to be installed in each sector 𝑠 ∈ 𝑆 based on different technical (e.g., availability 𝐴𝑣𝑓𝑚,𝑡),
economic (investment cost, 𝐼𝑛𝑣𝐶𝑜𝑠𝑡𝑚,𝑡, fixed operational and maintenance cost, 𝐹𝑂𝑀𝑚,𝑡, variable
operational and maintenance cost, 𝑉𝑂𝑀𝑚,𝑡, fuel cost, 𝐹𝐶𝑚,𝑡, CO2 emission cost, 𝐶𝑂2𝐶𝑜𝑠𝑡𝑡), and
environmental (CO2 emission rate, 𝐶𝑂2𝐸𝐹𝑚,𝑏𝑙) criteria. These technologies include both existing
capacity, 𝑚 ∈ 𝑀𝑒𝑥, new candidate power plants, 𝑚 ∈ 𝑀𝑛𝑒𝑤, and new units with firm
commissioning plans, 𝑚 ∈ 𝑀𝑓𝑥. Technical efficiency, 𝑒𝑓𝑚,𝑏𝑙, and CO2 emission rate 𝐶𝑂2𝐸𝐹𝑚,𝑏𝑙,
are divided into a number of blocks 𝑏𝑙 ∈ 𝐵𝐿 to represent more realistically the operation of power
generating units. Grid’s stability is also taken into account by incorporating minimum peak reserve
requirements, 𝑅𝑠𝑚𝑔𝑡, and maximum penetration rates of renewable energy technologies (RES),
𝑀𝑎𝑥𝑅𝑒𝑛𝑡. Furthermore, pumped storage is incorporated in the model to facilitate load balancing.
The energy policy tools to be utilized in order to promote the use of RES include a CO2 emission
cap 𝐶𝑂2𝐶𝑎𝑝𝑡, a mandatory production of a share of the total electricity generation form RES,
𝑇𝑟𝑔𝑡, as well as CO2 emission pricing, 𝐶𝑂2𝑝𝑟𝑖𝑐𝑒𝑡. Finally, CO2 emissions trade is taken into
consideration providing the opportunity to the administrator to determine the optimal strategy
concerning the total environmental performance of the system.
Book of Proceedings
65
The objective function to be optimized concerns the minimization of the total power system
expansion cost so as to satisfy the electricity demand of each sector 𝑠 ∈ 𝑆 in each time period 𝑡 ∈
𝑇. The objective function is given by (1), and the demand balance is expressed by Equation (2). A
detailed presentation of the initially developed mathematical model can be found in Koltsaklis et
al. (2014).
𝑀in ∑∑InvCostm,ttm
⏞ Investment cost
+∑∑FOMCostm,ttm
⏞ Fixed O&M cost
+∑∑VOMCostm,ttm
⏞ Variable O&M cost
+∑(CO2Expt − CO2Revt)
t
⏞ CO2 trade cost
+∑∑FCm,ttm
⏞ Fuel cost
+∑∑(ImpCostnc,t − ExpCostnc,t)
tnc
⏞ Electricity trade cost
− ∑∑PumCostpum,ttpum
⏞ Pumping cost
(1)
∑ pinjm,b,tm∈(MEXIS ∩ Ms)
⏞ existing power units′generation
+ ∑ pinjm,b,tm∈(MNEW ∩ Ms)
⏞ new candidate power units′ generation
+ ∑ pinjm,b,tm∈(MFIXED ∩ Ms)
⏞ fixed units′generation
∑efls′,s,b,ts′≠s
−∑ efls,s′,b,ts≠s′
⏞ net electricity flow rates
+ ( ∑ iminjnc,b,tnc∈(MNC∩MS)
− ∑ exwdrnc,b,tnc∈(MNC∩MS)
⏞ net electricity imports
) =
∑ pwdrpum,b,tpum∈(MPUM∩MS)
⏞ pumping load
+ Dems,b,t⏞ electricity demand
∀ s, b, t (2)
2. Case study
The applicability of our model has been tested on a case study of the Greek power system. The
country has been divided into two sectors, i.e., North and South sector, based on their geographical
locations. The model takes also into consideration the interconnection of Cyclades with mainland
in 2016, as well as Crete’s interconnection with the central power system in 2020. When it comes
to electricity requirements projection, Greece’s severe economic crisis has been taken into account,
since electricity demand starts from 54 TWh in 2014, rises with a moderate rate to almost 59 TWh
in 2020, and reaches 68.2 TWh in 2030.
Book of Proceedings
66
3. Results and discussion
3.1 Power capacity additions
Diagram 3: Power capacity mix per technology during each time period
The Reference scenario examines the case where the majority of old lignite power generation units
are to be decommissioned earlier than their expected lifetimes. The studied period is between 2014
and 2030. Between 2014 and 2028, an amount of 3,817 MW (3,306 MW by 2021) is to be
withdrawn from the Greek power system due to the fact that these units are characterized by very
low electrical efficiency and significant carbon footprint. The results of the model indicate the
construction of four new lignite plants (two units in 2021, one unit in 2022, and one in 2028) with
a total installed capacity of 2,400 MW along with a lignite power station having a firm
commissioning plan (615 MW in 2019). This investment plan and strategy can be explained by
the fact that lignite constitutes a domestic, low cost fuel and the candidate new lignite power
generating units have better technical performance and lower carbon emission factor when
compared to the existing conventional lignite units. In total, lignite power capacity is reduced by
715 MW between 2014 and 2030, as depicted in Diagram 1.
With the exception of a new, large natural gas unit with a firm commissioning plan (811 MW in
2015), no new natural gas plants are to be constructed since there was an overinvestment of gas
units during the previous years when there was no sign of the country’s deep economic recession.
The model does not determine the construction of coal power plants on the grounds that coal price
in not as competitive as domestic lignite’s one. Hydro units report a small increase in their
capacities since they start from 3,175 MW in 2014 and reach almost 3.5 GW in 2030. Oil capacity
is reduced by 431 MW due to the gradual decommissioning of some old heavy fuel oil units. Since
there is a national target stating that the share of RES in electricity generation must be at least
0
5000
10000
15000
20000
25000
30000
2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030
MW
Time period (year)
Lignite Natural Gas Oil Hydro Coal Photovoltaic Wind
Book of Proceedings
67
equal to 40% of the total from 2020 onwards, the results indicate significant capacity additions in
wind turbines and photovoltaic units. Wind turbines start from around 1.7 GW in 2014, rise to 5.3
GW in 2020 and reach 6.2 GW by 2030. 27% of that capacity in 2030 is to be installed in both
Crete and Cyclades. They comprise the largest power generation technology, in terms of installed
capacity, since 2021. Following a similar pattern, photovoltaic units report an increase of 2 GW in
their capacities between 2014 and 2030. The largest share of the newly installed renewable
technologies is observed in Cyclades and Crete, being interconnected to the mainland by 2016 and
2020 correspondingly.
3.2 Power generation
Diagram 2: Power generation mix per technology during each time period
As can be observed in Diagram 2, lignite units maintain a constant rate in their electricity
production, since it starts from around 23 TWh in 2014 and results in 25.6 TWh in 2030. Natural
gas units report an almost constant production of around 14 TWh between 2014 and 2018, while
their electricity generation is reduced during the next five years due to the large penetration of RES
units. Finally, they increasingly contribute to the rising electricity demand satisfaction during the
last years of the studied period since almost 12.5 TWh are generated by these units. Hydro units
play a balancing role in the electricity demand satisfaction with an average of almost 5 TWh during
the whole period. Oil power plants are characterized by a decreasing contribution to the demand
balance due to the fact that a significant amount of RES units are to be installed in the
interconnected islands (Cyclades and Crete) taking the place of the old, carbon intensive diesel
and heavy fuel oil plants which continue to operate up to levels that are necessary for the stability
of the local power grid.
-10
0
10
20
30
40
50
60
70
80
2014 2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030
TW
h
Time period (year)
Lignite Natural Gas Oil Hydro Wind Photovoltaic Imports Exports Pumping Electricity Demand
Book of Proceedings
68
The wide variation in the electricity production mix comes from renewable energy technologies,
and especially from wind turbines. Their generation begins from 3.6 TWh in 2014, rises to almost
12 TWh in 2020 and reaches 14.5 TWh in 2030. Similarly, photovoltaic units almost double their
production from 3.8 TWh in 2014 to 7.2 TWh in 2030. Almost 32% of that generation is produced
in Crete and Cyclades. Net electricity imports account for a significant share of the total electricity
demand satisfaction starting from 2.6 TWh in 2014 and approaching almost 5 TWh in 2030. Note
that the northern interconnections of the country (Albania, FYROM, Bulgaria and Turkey) are
mainly importing, while the southern interconnection, i.e., Italy, has an exporting profile.
3.3 Electricity flow rates
Diagram 3: Electricity flows among domestic zones during each time period
The combined effect of the binding target regarding mandatory electricity production from RES
and the fact that currently autonomous islands, which are to be interconnected to the mainland, are
characterized by higher availability in terms of wind potential and solar irradiation, converts them
into net electricity exporters to the mainland after their interconnection. A mitigation of electricity
flows, or even reverse flow during the last two years, is also observed from the North to the South
system. Diagram 3 illustrates this trend depicting the electricity flows among domestic zones
during each time period.
4. Conclusions
This work presents an integrated, multi-regional, long-term generation expansion planning model
incorporating possible interconnection with autonomous power systems. Our approach enables
decision makers to develop and design alternative pathways for a power system by providing
detailed calculations on the specific characteristics of each sector of the studied power grid. The
-2
0
2
4
6
8
10
12
14
16
2015 2016 2017 2018 2019 2020 2021 2022 2023 2024 2025 2026 2027 2028 2029 2030
TW
h
Time period (year)
FROM NORTH TO SOUTH FROM CYCLADES TO SOUTH FROM CRETE TO SOUTH
Book of Proceedings
69
results highlight that advanced lignite units will continue to play a strategic role in the Greek power
mix accompanied by natural gas units and rapidly penetrated RES units. Islands’ interconnection
leads to the mitigation of environmental impact due to the decommissioning of old oil units as well
as to significant cost savings by offering more flexibility to the total power system. Current
research is focused on developing a demand response mechanism along with a Monte Carlo
analysis to investigate the influence of several crucial and uncertain parameters into the long-term
development of a power system.
Acknowledgement
Financial support from the European Commission’s FP7 EFENIS project (Contract No:
ENER/FP7/296003) “Efficient Energy Integrated Solutions for Manufacturing Industries” and
Marie Curie “Energy Systems Engineering” (ESE) project is gratefully acknowledged
References
Agoris D., Tigas K., Giannakidis G., Siakkis F., Vassos S., Vassilakos N., Kilias V., and
Damassiotis M. “An analysis of the Greek energy system in view of the Kyoto commitments”.
Energy Policy. Vol. 32, 2004, pp. 2019-2033.
Dagoumas A.S., Kalaitzakis E., Papagiannis G.K., and Dokopoulos P.S. “A post-Kyoto analysis
of the Greek electric Sector”. Energy Policy. Vol. 35, 2007, pp. 1551-1563.
Koltsaklis N.E., Dagoumas A.S., Kopanos G.M., Pistikopoulos E.N., and Georgiadis M.C. “A
spatial multi-period long-term energy planning model: A case study of the Greek power system”.
Applied Energy. Vol. 115, 2014, pp. 456-482.
Rampidis I.M., Giannakopoulos D., and Bergeles G.C. “Insight into the Greek electric sector and
energy planning with mature technologies and fuel diversification”. Energy Policy. Vol. 38, 2010,
pp. 4076-4088.
Roinioti A., Koroneos C., and Wangensteen I. “Modeling the Greek energy system: Scenarios of
clean energy use and their implications”. Energy Policy. Vol. 50, 2012, pp. 711-722.
Book of Proceedings
70
Comparison of GA-ANN and Traditional Box-Jenkins Methods for Railway
Passenger Flow Forecasting
Nataša Glišović
Department for Mathematical Sciences, State University of Novi Pazar Vuka Karadzica bb, 36300
Novi Pazar, Serbia, email: [email protected]
Miloš Milenković
Division for Management in Railway, Rolling stock and Traction, The Faculty of Traffic and
Transport Engineering, University of Belgrade 11000 Belgrade, Serbia
Nebojša Bojović
Division for Management in Railway, Rolling stock and Traction, The Faculty of Traffic and
Transport Engineering, University of Belgrade 11000 Belgrade, Serbia
Rešad Nuhodžić
Railway Infrastructure of Montenegro Trg golootockih zrtava 13, 81110 Podgorica, Montenegro
Abstract
The exact prediction of the traffic conditions has become more and more significant due to the
vital role in the basic functions of the management of the traffic and railway processes of decision
making. This study presents an integrated genetic algorithm (GA) and artificial neural network
(ANN) for Railway Passenger Flow Forecasting using stochastic procedures. This paper aims at
showing the comparison of the hybrid model of the genetic-neural networks with traditional Box-
Jenkins model. The technique of the genetic algorithm is used for determining the design of the
neural networks. We compared the performances of proposed methods for multi step ahead
prediction of passenger flows on Serbian railways. The performance of this approach is explored
and results are presented. Compared performances shown that the proposed hybrid model gives
better results than the traditional Box-Jenkins model.
Keywords: SARIMA, Genetic algorithm, Artificial neural network, Railway passenger flow
forecasting.
Book of Proceedings
71
1. Introduction
In this paper we studied the railway passenger demand in Serbia. We used the data representing
monthly passenger flows on all lines of Serbian railway network provided by the Statistical Office
of the Republic of Serbia. At the time of our analysis a time series of monthly passenger flows
from January, 2006 to March 2014 was available. We proposed comparison of two methods,
traditional Box Jenkins method and Genetic algorithm-Artificial neural networks for forecasting
the total number of passengers on Serbian railways using stochastic procedure. We compared the
performances of proposed methods for multi step ahead prediction of passenger flows on Serbian
railways.
In the traditional parametric techniques, historical average (Smith and Demetsky, 1997),
smoothing techniques (Williamset al., 1998), and autoregressive integrated moving average
(ARIMA) (Hansen et al., 1999; Lee and Fambro, 1999) have been applied to forecast
transportation demand. Particularly, ARIMA has become one of the common parametric
forecasting approaches since the 1970s. The ARIMA model is a linear combination of time-lagged
variables and error terms. The ARIMA model has been widely applied in forecasting short-term
traffic data such as traffic flow, travel time, speed, and occupancy (e.g., Ahmed and Cook, 1979;
Hamed et al., 1995; Lee and Fambro, 1999). However, the applications of ARIMA or seasonal
ARIMA models are limited because they assume linear relationships among time- lagged variables
so that they may not capture the structure of non-linear relationships (Zhang et al., 1998).
Artificial neural network works in the same way as a human brain does, human brain consist of
number of neurons connected with each other, in the same way ANN consists of artificial neurons,
called nodes in network, connected with each other. The idea of Artificial Neural Network was
presented in late 1943 by Walter Pitts and Warren S.McCulloch as a data processing unit for
classification or prediction problems (F. Rosenblatt, 1958). The back-propagation learning
algorithm (BPLA) were proposed by Rumelhart et al. 1986 Many studies have indicated that
genetic algorithms (GA) can be successfully applied to identity global optimizations of
multidimensional functions (Chung and Alonso, 2004; Chu et al. 2008). GAs are widely applied
in the optimization of the parameters spaces of neural networks.
The aim of the research was to show success of the hybrid model (genetic artificial neural networks
(GANN)) for Railway Passenger Flow Forecasting. This study presents an integrated genetic
algorithm (GA) and artificial neural network (ANN) for Railway Passenger Flow Forecasting
using stochastic procedures. The technique of the genetic algorithm is used for determining the
design of the neural networks. This paper aims at showing the comparison of the hybrid model of
the genetic-neural networks with traditional Box-Jenkins model.
The paper is divided into several sections. In Section 2. will present the mathematical concept
hybrid models. The application of the results of the research will be presented in Ssection 3.
wWhile in Ssection 4. to the conclusions and opportunities are given for further research are given.
Book of Proceedings
72
2. Methodology
2.1 SARIMA model
Seasonal autoregressive integrated moving average (SARIMA) processes have been introduced
in the literature to model time series with trends, seasonal pattern and short time correlations. The
generalized form of ( , , ) ( , , )sSARIMA p d q P D Q model for a series tY can be written as (Box et al.
2008; Cryer and Chan 2008):
( ) ( )(1 ) (1 ) ( ) ( )s d s D s
p P t q Q tB B B B Y B B
(1)
where s is the length of the periodicity (seasonality) and t is a white noise sequence.
2
1 2( ) 1 p
p pB B B B (2)
2
1 2( ) 1 s s Ps
P PBs B B B (3)
are the non-seasonal and seasonal autoregressive (AR) polynomial term of order p and P ,
respectively
2
1 2( ) 1 q
q qB B B B (4)
2
1 2( ) 1s s s Qs
Q QB B B B (5)
are the non-seasonal and seasonal moving average part (MA) of order q and Q , respectively.
(1 )dB is the non-seasonal differencing operator of order d used to eliminate polynomial trends
and the seasonal differencing operator (1 )s DB of order D used to eliminate seasonal patterns. B
is the backshift operator, whose effect on a time series tY can be summarized as d
t t dB Y Y .
2.2 Neural Networks
The main characteristics of the neural networks is their ability to learn when we have a complex
nonlinear relationship between input and output. We use sequential procedures for training and
adapting them to the data (Anil Jain et al. 2000). Back propagation is currently the most widely
used techniques for training neural networks (Randall Sexton and Robert Dorsey, 2000). BP (back
propagation) neural network relies on a gradient algorithm to obtain the weights of the model and
uses back propagation algorithm to minimize the objective function. BPNN (back propagation
neural networks) typically consists of three layers: an input layer, a hidden layer and an output
layer. The most basic treatment processes called artificial neuron in the BN network and simulated
on the basis of biological neurons. The summation function of a neuron is done as follows (Cheng
and Liu, 2014):
j ji i j
i
b T r a
(6)
Book of Proceedings
73
j ji i j
i
y f w x
(7)
l lj j l
j
z g v y
(8)
21
2l l
l
E t z (9)
Where jb is the activation level of neuron j. T is the transfer function, jir is weight value, j is
bias. So the output of hidden layer and output layer described by the following equation (7) and
(8). The error of output neuron is given by Equation (9). Where ix and
lz are the input and output
signals. jy is the output of the hidden layer. jiw is the weight between input neuron j to hidden
neuron i. ljv is the weight between hidden neuron l to output neuron j. j and l are the biases for
the hidden layer and output layer. f and g are transfer functions for hidden and output layers. lt is
the expected output. E is the error between the expected output and calculated output. In back
propagation algorithms resilient BP was used (for details see (Riedmiller and Braun 1993)).
2.3 Genetic algorithm
GA are robust and adaptive methods that can be used for solving combinatorial optimization
problems. The basic structure is a population of individuals, each individual represents a possible
solution in the search space for a given problem (the space of all solutions). In doing so, each
grants fitness function which assesses the quality of the given individual, represented as a single
solution in the search space. GA must provide a way to continuously from generation to generation,
improves the absolute adjustment of each individual in the population, and the average adjustment
of the entire population. This is achieved by successive application of genetic operators of
selection, crossover and mutation, to give all a better solution given the specific problem (Hansen
et al. 1999).
2.4 Hybrid model
Applying this idea of genetic algorithms in neural networks, integration is used so that on the basis
of input data, select a population that represents the number of neurons in the middle. In this way
determine the architecture of neural networks where the work is carried out on the basis of
prediction and prediction error MAE select the best architecture that represents the final output. In
this research, the stopping criterion is of 200 the population (see pseudo code).
Book of Proceedings
74
******************************************************************************
/* pseudo code for the algorithm to predict hybrid GA-ANN */
Input_Data();
Initialization_of_the_Population();
for(i=0; i< Npop; i++)
pi= a_Values_Function();
Fitness_Function();
Selection ();
Intersection ();
Mutation ();
MAEmin = ANNresilientBackPropagation();
while(!Finishing_Criteria _GA() )
{
for(i=0; i< Npop; i++)
pi= a_Values_Function ();
Fitness_Function();
Selection ();
Intersection ();
Mutation ();
MSE=ANNresilientBackPropagation();
If(MAE < MAEmin)
{
MAEmin=MAE;
CurrentArchitecture =ANNresilientBackPropagation();
}
}
Print_the_Output_Data (the current architecture);
******************************************************************************
Book of Proceedings
75
In the GA, reproduction is implemented by a selection operator. Selection is the population
improvement or “survival of the fittest" operator. It duplicates structures with higher fitness values
and deletes structures with lower fitness values.
The crossover, when combined with selection, yields good components of good structures
combining to yield even better structures. Crossover forms n/2 pairs of parents if the number of
population is n. Each pair produces two offspring structures to the mutation stage. The offspring
is the outcome of cutting and splicing the parent structures at various randomly selected crossover
points. The approaches for selecting crossover points are one-point crossover, two-point crossover,
and uniform crossover.
Mutation creates new structures that are similar to current ones. With a small, pre-specified
probability, mutation randomly alters each component of each structure. The reason for using
mutation is to prevent missing some significant information during reproduction and crossover.
This procedure would avoid the local minimum.
In this research, where GA neural network model is implemented in the programme language C#,
the population represents the input time series, which is the output of SARIMA model obtained as
the output of SPSS. Using GA, the best generation is received which represents the structure of
neural network for which MSE (mean absolute error) is the smallest.
3. Results
We used the data representing monthly passenger flows on all lines of Serbian railway network
provided by the Statistical Office of the Republic of Serbia. The learning period of the GA-neural
network and traditional Box and Jenkins was from January, 2006 to July, 2013. The prediction
was performed the following months (from July 2013 to March 2014.) (Figure 1 and Figure 2)
Figure 1. Shown SARIMA prediction passenger flows from July 2013 to March 2014. Red line
presented observed values while blue line presented forecast.
Figure 2. Shown GA-ANN prediction passenger flows from July 2013 to March 2014. Green line
presented prediction values used Hybrid model and blue line presented observed values.
Book of Proceedings
76
Comparison GA neural network is made with the traditional Box and Jenkins method and the
prediction success is shown through the prediction errors (see the table 1).
Table 4 Shows the performance of the prediction through all errors. In this case you can see the
advantage of the hybrid model predictions compared to the traditional, as errors of prediction
significantly less.
Models Errors
MAE RMSE MAPE
SARIMA 19.18 26.37 4.01
Hybrid model 12.35 16.31 2.21
4. Conclusion
The study shows that the performance of railway passenger flow forecasting can be significantly
enhanced by using proposed hybrid method. These results show that the proposed model can be
useful tool for railway passenger flow, better then SARIMA prediction. By comparing the
prediction values with the real value of the railway passenger flows and by calculating MAE,
RMSE and MAPE the satisfactory prediction results have been achieved. These results show that
the proposed hybrid model can be useful tool for the railway passenger flows prediction then
traditional Box-Jenkins.
Book of Proceedings
77
Acknowledgement
The work presented here was supported by the Serbian Ministry of Education and Science (project
No. III44006 and No. 036022).
References
Ahmed, M.S., Cook, A.R., 1979. Analysis of freeway traffic time-series data by using Box-Jenkins
techniques. Transportation Research Board Record 722, 1–9.
Anil K Jain, Robert PW. Duin, Jiangchang Mao. Statistical pattern recognition: A review. IEEE
transaction on pattern analysis and machine intelligence. 2000; 22(1): 4-37.
B. Chu, D. Kim, D. Hong, J. Park, J. T. Chung, J.-H. Chung and T. H. Kim, GA-based fuzzy
controller design for tunnel ventilation systems, Automation in Construction, vol.17, no.2, pp.130-
136, 2008.
Box G.E., Jenkins G.M. and Reinsel G.C., 2008. Time Series Analysis, Forecasting and Control,
New Jersey: John Wiley and Sons.
Cryer J.D. and Chan K.S., 2008. Time Series Analysis: with Application in R, New York: Springer.
F. Rosenblatt, “The perception: A probabilistic model for information storage and organization in
the brain”, Psychological review, 65(6):386, 1958.
H. S. Chung and J. J. Alonso, Multi objective optimization using approximation model-based
genetic algorithms, The 10th AIAA/ISSMO Multidisciplinary Analysis and Optimization
Conference , vol.1, pp.275-291, 2004.
Hamed, M.M., Al-Masaeid, H.R., Bani Said, Z.M., 1995. Short-term prediction of traffic volume
in urban arterials. Journal of Transportation Engineering 121 (3), 249–254.
Hansen, J.V., McDoald, J.B., Nelson, R.D., 1999. Time series prediction with genetic-algorithms
designed neural networks: an empirical comparison with modern statistical models. Journal of
Computational Intelligence 15 (3), 171–183.
Lee, S., Fambro, D.B., 1999. Application of subset autoregressive integrated moving average
model for short-term freeway traffic volume forecasting. Transportation Research Board 1678,
179–188.
Li Cheng, Jin Liu, “An Optimized Neural Network Classifier for Automatic Modulator
Recognition”, TELKOMNIKA Indonesian Journal of Electrical Engineering, Vol. 12, No. 2,
February 2014, pp. 1343-1352.
Book of Proceedings
78
Riedmiller, M. and Braun, H. (1993) A Direct Adaptive Method for Faster Back propagation
Learning: The RPROP Algorithm. In: Ruspini, H., (Ed.) Proc. Of the ICNN 93, San Francisco,
pp. 586-591
Randall S Sexton, Robert E Dorsey. Reliable classification using neural networks: a genetic
algorithm and back propagation comparison. Decision Support Systems. 2000; 30(1): 11-22.
Rumelhart, D. E., Hinton, G. E., and Williams, R. J. 1986. Learning internal representations by
error propagation. In Parallel Distributed Processing, D. E. Rumelhart, J. L. McClelland, and the
PDP Research Group, eds., Vols. I and II, Bradford Books and MIT Press, Cambridge, MA.
Smith, B.L., Demetsky, M.J., 1997. Traffic flow forecasting: comparison of modeling approaches.
Journal of Transportation Engineering 123 (4), 261–266.
Williams, B.M., Durvasula, P.K., Brown, D.E., 1998. Urban freeway traffic flow prediction:
application of seasonal autoregressive integrated moving average and exponential smoothing
models. Transportation Research Record 1644, 132–141.
Book of Proceedings
79
Inspection of power grid by periodic vehicle routing formulation
Dr. Vasilis Spathis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece.
Ioannis Forlidas
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece.
Erotokritos Skordilis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece.
Dr. Georgios Saharidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece.
Abstract
The inspection of unattended power substations by trained personnel has been surprisingly
important regardless of the enhanced remote supervision today’s technology offers. In this paper,
we present a mixed integer linear programming formulation that examines the
optimized scheduling of these inspections with respect to the particularities and the nature of the
inspection, taking also into consideration the multiple origins of the inspections and the scattered
facilities to be inspected.
Keywords: periodic vehicle routing problem, multi-depot, multi vehicle, network clustering
1. Introduction
The substations that need to be inspected are scattered throughout the country and are essentially
demoted voltage substations. These facilities include equipment that costs several million Euros
and are essential for the stability of the providing power and the quality of service of the power
industry. Specifically throughout mainland Greece are 291 substations connected by 11,300 km of
transmission lines. The particularities of the technical inspection of the equipment and moreover
the importance of the facilities imply the regularly physical inspection by technical personnel.
These inspection have the advantage of the natural presence of highly trained personnel that will
not only observe, but also hear or sense a possible malfunction hours or days before it become
Book of Proceedings
80
crucial for the system. Their report will introduce an early maintenance, or even replacement of
suspicious equipment reducing the probability of a major discontinuance of the power supply.
Every route of inspections will include the departure station, the substation that need to be visited,
the time of inspection at each substation and the total available time for inspection. Our goal is to
reduce the total travel time for the inspection of the substations.
The solution of the problem presented in this paper is based on a more general class of problems,
those of vehicle routing (VRP) and more specifically of the multi depot VRP.
2. Real case study
The mathematical formulation of this problem is based on the formulation of vehicle routing
optimization problems with multiple depots and time windows presented by Dondo et al. [1]. The
formulation was altered to satisfy the requirements of our problem and then applied on real data.
The substations of Central Greece were fed to the algorithm and routes that can cover the
inspection to that substation were calculated. Total number of substation in Central Greece is 26
of whom 24 need to be inspected from inspectors departing from 4 depots, with no more than 5
routes from each depot and total time of each route of inspection not exceeding 400 minutes.
The formulation was realized in C++ code and solved using CPLEX libraries. Due to the large
number of the substations it was not possible to have complete results for the whole problem.
Therefore it was necessary to group the substations (clustering) around a given starting point
(depot) and solve different sub problems (VRP). The grouping took part considering geographical
criteria.
PICTURE 1: Electrical substations of central Greece
Book of Proceedings
81
3. Results
DIAGRAM 2: Re-clustering with respect to partial optimized total solution
The chart above shows the total time required for completion of the inspection of five different
cases. In the first column (un-optimized) is presented the domestic solution which has not been
optimized, already used by the maintenance company. In the second column (fixed clusters) it is
shown the time-optimized solution for the existing clustering, which the company is using. The
last three columns present the results for three different scenarios of clustering carried out. It should
be noted that under each column the CPU time and the total progress of the solution is displayed.
Examining the above figure it is clearly shown that the time for the inspection without optimization
is about 3000 minutes, and the optimized routing decreases the total time to 1900 minutes.
Similarly we observe that for scenarios A, B and C we have no important improvement compared
to the fixed clustering solution.
Lastly we tested the complete problem and after 48 hours of execution and without leaving the
100% GAP, the best solution provided had an improvement of 60 minutes (1 hour) compared to
the clustering scenarios tested.
Book of Proceedings
82
4. Conclusions – Further Investigation
It is clear that the optimized solution provided by the model discussed in this paper can provide
savings up to 40% in hourly wages and significant savings in driven km. Furthermore, the creation
of organized routes allows smooth implementation of the inspections, solves several personnel
management issues, while making the audit process more efficient without ignoring the factors
and characteristics of the inspection.
One of the real problems that we could improve is the time resolution required by the computer.
This could be accomplished by the following ways:
• Constrains for near-by substations
• Constrains for single route covering more than one substation
• Constrains for incompatible substation
References
R. Dondo, J. Cerda, “A cluster-based optimization approach for the multi-depot heterogeneous
fleet vehicle routing problem with time windows”, European Journal of Operational Research
176 pp. 1478- 1507, 2006.
J-F. Cordeau, M. Gendreau, G. Laporte, “A Tabu Search Heuristic for Periodic and Multi-Depot
Vehicle Routing Problems ”, European Journal of Operational Research 119 pp. 105–119, 1997.
S. Nanda Kumar, R. Panneerselvam, “A Survey on the Vehicle Routing Problem and Its Variants
” , European Journal of Operational Research 74 pp.66-74, 2012.
Vehicle Routing Problem from Wikipedia, the free Encyclopedia (last access 22/6/2014).
Book of Proceedings
83
Environmental performance evaluation using a fuzzy aggregation-
disaggregation approach
Zouboulia Sbokou
School of Production Engineering and Management, Technical University of Crete, University
Campus, GR73100 Chania, Greece email: [email protected]
Evangelos Grigoroudis
School of Production Engineering and Management, Technical University of Crete, University
Campus, GR73100 Chania, Greece.
Michael Neophytou
Mills of Crete, 40 Eth. Venizelou, Souda, GR73200 Chania, Greece.
Abstract
An important tool for the evaluation and the documentation of a successful environmental
management system is the Environmental Performance Evaluation (EPE). The EPE is defined as
a continuous internal process and a management tool that uses indicators in order to evaluate the
environmental management system of a business organization and to compare past and present
environmental performance. International standards ISO 14031-14032 describe the categories of
performance indicators; however they do not determine a specific framework for the development
and measurement of these indicators. The main aim of this study is to present an EPE methodology
based on a fuzzy multicriteria analysis approach. In particular, the Fuzzy UTASTAR method is
applied in order to evaluate the environmental performance of a mill industry. It is an extension of
the well-known UTASTAR method capable to handle both ordinary (crisp) and fuzzy evaluation
data. To evaluate the environmental performance of the industry, the production processes are
analyzed and the environmental indicators related to the environmental impact of the industry are
defined. The environmental indicators are related to the products of the industry, the material
consumption, the consumption of natural resources and waste management. The five groups of the
indicators are: air emissions, solid waste, natural resources and energy, environmental education
and third parts, recycling and improvement measures. The main steps of the presented approach
include the following: Criteria assessment (definition of the final set of indicators and their
measurement units), Definition of fuzzy sets (fuzzy values that reflect the low, medium and high
performance of each indicator), Development and ranking of alternative scenarios for each group
of indicators, Application of the Fuzzy UTASTAR model (estimation of fuzzy utility functions,
overall utilities for scenarios). The actual value of each indicator is used in the estimated utility
function of the corresponding indicator and the result is normalized in order to calculate the
environmental performance of the industry. The environmental performance of the industry can
be measured per dimension (group of indicators) or per total.
Keywords: Environmental Performance Evaluation, Fuzzy UTASTAR, Environmental
Management Systems, Aggregation-Disaggregation Approach.
Book of Proceedings
84
1. Introduction
According to the international standard ISO 14031 (ISO, 1999) environmental performance
evaluation (EPE) is a process to facilitate management decisions regarding an organization’ s
environmental performance by selecting indicators, collecting and analyzing data, assessing
information against environmental performance criteria, reporting and communicating, and
periodically reviewing and improving this process. Environmental performance indicators (EPIs)
measure the current or past environmental performance of an organization and compare it to the
targets set by the organization’s management (Jasch, 2008). International standard ISO 14031
describes the categories of performance indicators; however it does not determine a specific
framework for the development and measurement of these indicators.
Multi-Criteria Analysis (MCA) methodologies may be considered as tools that can handle a
set of indicators for the environmental evaluation. MCA applications are applied mainly to
sustainability appraisals; that is, they include indicators from economic, social and environmental
categories, as well as technical criteria (Herva and Roca, 2013).
The main aim of this study is to present an EPE methodology based on a fuzzy MCA approach.
Although fuzzy methods have been applied in the examined problem, the presented study is the
first attempt in the context of preference disaggregation.
Fuzzy UTASTAR is a method for inferring fuzzy utility functions from a partial preorder of
options evaluated on multiple criteria. It is an extension of the well-known UTASTAR method
(Siskos and Yannacopoulos, 1985) capable to handle both ordinary (crisp) and fuzzy evaluation
data (Patiniotakis et al., 2011). Fuzzy UTASTAR builds fuzzy additive value functions taking as
input a partial preorder on a subset of the options, called reference set, along with their associated
scores on the criteria. The resulting fuzzy utility functions can subsequently be used to estimate
the (fuzzy) utility of each option, thus allowing their ranking, prioritization, selection or
classification. The ranking of the options in partial preorder is as compatible as possible to the
original one (Patiniotakis et al., 2011).
2. Environmental Performance Evaluation in a Mill Industry
2.1 Environmental Indicators
In order to evaluate the environmental performance of the industry, the production processes
are analyzed and the environmental indicators related to the environmental impact of the industry
are defined considering the input-output analysis (Figure 4), the environmental aspects and the
environmental policy of the industry. The environmental indicators are related to the products of
the industry, the consumption of materials, the consumption of natural resources, and the generated
wastes. The five groups of indicators are: air emissions, solid waste, natural resources and energy,
environmental education and third parts, recycling and improvement measures.
2.2 Criteria Assessment
It is necessary to reduce the large set of indicators in order to be applied and controlled by the
industry. For this reason, a technique based on the assessment of environmental impacts of the
selected indicators is applied (Karavias, 2008). The first three categories of indicators related to
Book of Proceedings
85
environmental pollutants (air emissions, solid waste, natural resources and energy) are evaluated
on five criteria: 1) the severity of the impact, 2) the likelihood to happen, 3) the frequency of
occurrence, 4) whether the indicator is controllable, and 5) if there is legislation demanding to
measure or not the indicator. The other two categories of indicators (environmental education and
third parts, and recycling and improvement measures) are evaluated on three criteria: 1)
reportability, i.e., how and the frequency of exposure of the various documents and processes, 2)
the interest from various stakeholders and third parts, and 3) if the indicator is in the organization’s
objective. A five-level scale is used for the criteria of severity, likelihood, frequency,
controllability, reportability, and interest from stakeholders (with 1 the lowest and 5 the highest
value). A two-level scale is used for the other two criteria of legislation and organization’s
objective (with 1 for positive and 0 for negative response). The decision-maker (DM) scored the
indicators and then the impact of each indicator is calculated, as shown in Table 5 and Table 6.
The impact of each indicator is calculated as follows:
3
415
1 11 1 1
3 124 3 4 3
ttx x
Impact x
for the three first categories of indicators and
6 7
8
1 11 1 1
3 4 3 4 3
x xImpact x
for the two last categories of indicators,
where ix , 1,2, ,8i are the criteria of severity, likelihood, frequency, controllability,
legislation, reportability, interest from stakeholders, and organization’s objective, respectively.
Inputs Outputs
Raw materialsFuel
EnergyWater
Packaging materials
FlourAnimal feed
Energy consumptionAir emissionsSolid wasteWastewater
NoiseHeat release
Figure 4. Input-Output analysis of the mill industry
The indicators with the highest impact of each category are chosen for the final set of indicators
(bold and underline values in Table 5 and Table 6). There is no single threshold for the impact so
each category will have at least two indicators. The final set of indicators consists of 17 indicators.
Book of Proceedings
86
Indicator Criteria Impact
Severity Likelihood Frequency Controllability Legislation
Amount of CO2 5 5 5 1 1 0.505
Amount of NOx 4 4 4 2 1 0.489
Amount of SOx 4 4 4 2 1 0.489
Amount of SS 4 4 5 1 1 0.430
Amount of CO 3 3 3 3 0 0.188
Amount of PM 3 3 3 2 1 0.438
Table 5. Impact assessment of air emissions
Indicator Criteria Impact
Severity Likelihood Frequency
Quantity of auxiliary materials from recycled materials 3 4 0 0.417
Annual quantities of recycled products 3 2 0 0.250
Number of products with environmental friendly instructions 4 2 1 0.667
Percentage of environmental goals achieved 2 2 0 0.167
Number of vehicles with eco - friendly technology 1 2 1 0.417
Number of planned audits and inspections completed 2 1 1 0.417
Number of findings in inspections per year 2 1 1 0.417
Number of emergency exercises that have taken place 2 2 1 0.500
Response time for the corrective actions 3 4 1 0.750
Costs due to penalties and fines from infringements 1 2 1 0.417
Cost of environmental improvement actions/the total budget 2 2 1 0.500
Table 6 Impact assessment of recycling and improvement actions
2.3 Assessment of Fuzzy Sets and Development of Scenarios
In cooperation with the management of the mill industry, fuzzy values that reflect the low,
medium, and high performance of each indicator are defined. Next, different scenarios with fuzzy
values for each indicator are developed for each category of the selected indicators. An example
for the case of air emissions is shown in Table 7. In order to apply the Fuzzy UTASTAR method,
the DM is asked to rank these scenarios.
Book of Proceedings
87
2.4 Application of Fuzzy UTASTAR
Based on the aforementioned scenarios, the Fuzzy UTASTAR method is applied in order to
estimate the fuzzy value of each scenario and the utility function of each indicator.
The fuzzy value of each scenario is a triangular fuzzy number ( , , )x y zU U U , which is converted
to a crisp number according to following formula: 1 2 3( ) ( 2 ) 4R a a a a , where ( )R is a
ranking function and 1 2 3( , , )a a a is a triangular fuzzy number. The calculation of ( )R is necessary
in order to obtain comparable values for all the examined scenarios. Table 8 shows the utilities of
the alternative scenarios, using the previous formula.
The estimated value functions for the criteria in the category of air emissions are shown in
Figure 5. All these value functions refer to decreasing criteria, since they are related to undesirable
outputs of the production process. As it can be observed, the most important indicator for the
environmental performance of the mill in this particular category is the amount of SOx.
No. of scenario Amount of CO2
(tn CO2 / month)
Amount of NOx
(kg NOx / month)
Amount of SOx
(kg SOx / month)
Ranking by the DM
1 (30, 40, 46) (75, 80, 85) (140, 150, 160) 1
2 (47, 50, 55) (55, 60, 63) (162, 175, 182) 2
3 (31, 35, 37) (64, 70, 74) (185, 190, 200) 3
4 (38, 40, 46) (55, 60, 63) (185, 190, 200) 4
Table 7 Alternative scenarios for the category of air emissions
No. of scenario Ux Uy Uz R
1 0.422 0.689 0.916 0.679
2 0.199 0.260 0.372 0.273
3 0.103 0.210 0.293 0.204
4 0.013 0.160 0.223 0.139
Table 8 Estimated overall utilities for the scenarios of air emissions
Book of Proceedings
88
Figure 5 Value functions for the criteria of air emissions
2.5 Overall environmental performance
The actual value of each indicator, as measured in the mill industry, is used in the estimated
utility function of the corresponding indicator and the result is normalized in order to calculate the
environmental performance of the industry. The environmental performance of the industry can
be measured per dimension (group of indicators) or overall. For example, Table 9 presents the
calculations in the case of air emissions. As shown, based on the estimated utility functions ( iX
corresponds to the values of the criteria scale, while iY refers to the corresponding utility), the
utility of the current value of the industry is calculated through linear interpolation, since Y are
piecewise linear functions. The overall performance score is the sum of these utilities, and can be
used to normalize the estimated value of each indicator.
Following the previous procedure, the environmental performance of the industry may be
estimated. Figure 6 shows the EPE results in the category of air emissions, as well as the overall
performance of the industry. As it can be observed, the results in the dimension of air emissions
are relatively high, while the category of resources and energy appear to have the lowest
performance.
Indicator Value Utility Current
value
Linear
interpolation
Normalized
value X1 X2 X3 Y1 Y2 Y3
Amount of CO2 31 43 55 0.120 0.000 0.000 35 0.080 0.667
Amount of NOx 55 70 85 0.043 0.043 0.000 75 0.029 0.674
Amount of SOx 140 170 200 0.837 0.260 0.000 160 0.452 0.540
Overall value 0.561
Table 9 Calculation of environmental performance evaluation
Book of Proceedings
89
Figure 6 Environmental performance of the mill industry (air emissions and overall)
3. Conclusions
The applied method is an easy to handle and flexible tool for evaluating the environmental
performance of a business organization. This method can be applied in any type of organization,
as each organization can choose the suitable combination of evaluation criteria that are appropriate
for its own operations and activities. Also, the results are able to determine the strong and the weak
points, as well as potential improvement action regarding the environmental management system.
Another important advantage is that the proposed approach may take into account the DM’s
preferences (environmental strategy of the organization) and it can help the DM to handle the
uncertainty of the data. Finally, the method is able to give a clear picture of the rate of the
environmental objectives and targets achieved from the total environmental objectives and targets
set by the senior management.
References
Herva, M. and E. Roca. “Review of combined approaches and multi-criteria analysis for corporate
environmental evaluation”. Journal of Cleaner Production, Vol. 39, 2013, pp. 355-371.
ISO. Environmental management-Environmental performance evaluation-Guidelines: ISO 14031.
International Standard Organization, Geneva, 1999.
Jasch, C. “Environmental performance evaluation and indicators”. Journal of Cleaner Production,
Vol. 8 No. 1, 2008, pp. 79-88.
Karavias, P. Development of a methodology for the evaluation of environments performance. MSc
Thesis, School of Production Engineering and Management, Technical University of Crete,
Chania, 2008 (in Greek).
Book of Proceedings
90
Patiniotakis, I., D. Apostolou D., and G. Mentzas. “Fuzzy UTASTAR: A method for discovering
utility functions from fuzzy data”. Expert Systems with Applications, Vol. 38 No. 12, 2011, pp.
15463-15474.
Siskos Y. and D. Yannacopoulos, D. “UTASTAR: An ordinal regression method for building
additive value functions”. Investigação Operacional, Vol. 5 No. 1, 1985, 39-53.
Book of Proceedings
91
Rationalizing electricity production investments from renewable energy
sources in Greece using a synergy of multicriteria methods
Eleftherios Siskos
National Technical University of Athens, 9, Iroon Polytechniou Str., 157 80, Athens, Greece
Dimitrios Peronikolis
National Technical University of Athens, 9, Iroon Polytechniou Str., 157 80, Athens, Greece
John Psarras
National Technical University of Athens, 9, Iroon Polytechniou Str., 157 80, Athens, Greece
Abstract
The necessity for disengagement from conventional energy sources, along with the increasingly
strict measures of the European Union (EU) towards this direction, lead to the promotion of the
renewable energy sources (RES). In particular, Greece, although having high potential in
electricity production from RES, mainly hydroelectric, wind and solar, is still behind in
comparison to other EU countries in the area of RES adoption. Nevertheless, the incentives
provided by Europe and the Greek government, during the last few years, for investing on
electricity production from RES, are multiple and significant. The aim of this paper is to evaluate
and rank medium-scale investments on electricity production from RES, of approximately 10MW,
in the broader area of Greece. Specifically, a multicriteria evaluation system is elaborated, based
on four points of view: (i) technical (ii) economic (iii) social, (iv) environmental, and (v) political.
The investments assessed are categorized with respect to the type of RES invested upon (i.e. solar,
biomass, geothermal, etc.) and the area of implementation (mainland, islands or offshore). The
selection is supported by the ELECTRE IS method, which takes into account the presence of
pseudo-criteria through the proposed evaluation system. The criteria weights are elicited with the
aid of a revised Simos methodology. The overall objective of this research work is to support
energy policy decision making in Greece and trigger sustainable development.
Keywords: Multicriteria decision support, Simos method, Renewable energy sources, Energy
policy measures, Electricity production investments.
Book of Proceedings
92
1. Introduction – Scope of research
The necessity for disengagement from conventional energy sources, along with the increasingly
strict energy policy measures applied by the European Union (EU) in this direction, lead to
the promotion of renewable energy sources (RES). In particular, Greece despite having
enormous natural wealth, and all forms of RES, is still behind in matters of electricity production,
compared with other EU countries in this area.
As said, geologically and by its climate, Greece can support almost all types of electricity
production from RES, increasingly attracting investors. There are, though, different investment
opportunities, which are differentiated by their type, scale, location and other characteristics. The
challenge therefore, is to investigate what energy power investment by RES is the most appropriate
at the moment, based on the national needs and objectives.
Until now, this selection problem has been approached by various surveys and studies, assessing
the dynamics of each RES per category and quality, which deviates from reality. However, few
studies have been engaged to the evaluation of RES investments with the use of Multicriteria
Decision Analysis (MCDA).
This paper addresses the problem of the selection of the optimal investment for electricity
production from RES for the case of a public investor, such as the Greek government. Initially, an
analysis of the possible investment scenarios in electricity production from RES in Greece is
examined and elaborated. Nine alternatives are chosen and evaluated on five points of view. These
points of view are (i) technical (ii) economic (iii) social, (iv) environmental, and (v) political each
of whom consists of multiple dimensions. In the following phase the MCDA method, Electre IS,
is applied. The parameters of the problem, used by the method (weights, thresholds e.t.c.) are
determined by the Investor/Decision maker (DM). The method results in an outranking graph of
the alternatives, the results of which are examined prior to selecting the most preferred candidate,
based on the investor’s preferences.
2. Literature Review
A number of studies have been published in the field of evaluation of different energy planning
solutions. These works, which approximate the current study, can be categorized as follows:
Quantitative or qualitative comparison of all type of alternative energy sources
Qualitative assessment of RES investments
Quantitative evaluation of RES investments
Evaluation of different technologies of a particular RES investment
Multicriteria evaluation of energy planning investments
Book of Proceedings
93
Table 1 presents a general sample of the literature review on evaluations of energy planning
alternatives, with a small description and the method that has been applied.
Table 1: Indicative applications
Scientific study Description Method used
Kaya and
Kahraman, 2010
Energy planning evaluation using MCDA (all
energy sources) TOPSIS
Polatidis et al.,
2006
Analysis and comparison of multicriteria
methods over their suitability on energy
planning from RES.
Outranking, utility
based models, goal
programming
Tsoutsos et al.,
2008
Evaluation of the energy planning of Crete
using MCDA Promethee Ι & ΙΙ
Karakosta et al.,
2012
Comparison of RES technologies and
investments to nuclear ones
Qualitative
evaluation
Georgopoulou et
al., 1996
Evaluation of RES technologies for the island
of Crete using MCDA Εlectre ΙΙΙ
Kahraman et al.,
2008 Multicriteria evaluation of all energy sources AHP
Naim et al., 2001 Multicriteria evaluation of alternative power
plant technologies
Additive value
model using variable
weights
Burton and
Hubacek, 2007
Multicriteria evaluation of small scale power
plants
Macbeth & Cost-
benefits
Cavallaro, 2005 Energy planning evaluation Promethee Ι & ΙΙ
For an elaborative literature review, on sustainable energy decision making using multicriteria
decision making methods, the reader is prompted to read the paper of Wang et al. (2009).
3. Description of the problem
The purpose of this study is to select the preferentially optimal among the electricity production
investments (scale of 10 MW) from RES in Greece. Firstly, the alternative options of investments
that can be supported by Greece’s geological, climatic and technical conditions and limitations are
Book of Proceedings
94
distinguished. Based on the previews studies the following alternatives are finally selected (see
Table 2). To simplify the procedure that follows, each alternative is represented by a letter.
Table 2: Alternatives
Photovoltaic station (Interconnected) a
Photovoltaic station (Not
interconnected) b
Wind plant (Interconnected) c
Wind plant (Not interconnected) d
Small hydroelectric plant e
Solar thermal plant (Interconnected) f
Solar thermal plant (Not interconnected) g
Geothermal power plant h
Biomass power plant i
These alternatives are evaluated based on a consistent family of criteria that is built according to
the classical modeling methodology (Roy, 1985). In fact, a set of dimensions is grouped in five
points of view which in turn are all sub-aggregated to lead to a set of nine evaluation criteria (see
Table 3) (Løken, 2005). The values in the parentheses show the type of each criterion and their
worst and best levels. The criteria fabrication process was implemented after careful and thorough
planning in order to cover all different aspects that may affect the disposition of the investor
towards his final decision.
Table 3: The evaluation system of RES investments
Points of View Dimensions Criteria
Social
Jobs creation g1: Social criterion
(quality scale, 1-5) Social acceptance
Additional social benefits
Economic
Investment cost g2: Cost criterion
(M €/ M,W 2-18,1) Operational and maintenance cost
Electricity selling price g3: Revenue criterion
(€/ΜWh ,123- 406 ) % Subsidy
Technical Efficiency g4: Effective operation
(quality scale, 1-5) Compatibility
Book of Proceedings
95
Reliability
Know how g5: Expertise
(quality scale, 1-5) Operational safety
Preparation time g6: Project cycle
(years, 18-48) Life Cycle
Environmental
Effect on soil
g7: Environmental
criterion
(quality scale ,1-5)
Effect on water
Noise pollution
Landscape degradation
Required area g8: Required area
(w/m2, 4-60)
Political Penetration margin g9: Political support
(quality scale, 1-5) Stability & policy bureaucracy
4. Methodological Frame
The MCDA method that was selected for the purposes of this decision problem is the ELECTRE
IS, due to the fact that it that can also manage pseudo-criteria. For a pair of actions a,b the primacy
relationship is defined (see Figueira et al., 2005):
aSb ↔ (a , b)
The concordance check is described by two sets (Roy and Bouyssou, 1993):
Js = { j ∈ J / gj (a) + qj[gj(a)] ≥ gj(b)}
And JQ = { j∈ J / gj(a) + qj[gj(a)] < gj(b) ≤ gj(a) + pj[gj(a)]}
It is positive when C (a,b) = ∑ 𝑤𝑗𝑗𝜖𝐽𝑠 + ∑ 𝜑𝑗𝑤𝑗𝑗𝜖𝐽𝛺 ≥ 𝑠
Book of Proceedings
96
where φj= 𝑔𝑗(𝛼) + 𝑝𝑗[𝑔𝑗(𝛼)]−𝑔𝑗(𝑏)
𝑝𝑗[𝑔𝑗(𝛼)]−𝑞𝑗[𝑔𝑗(𝑎)]
w: criterion importance weight
g: value of each alternative of a criterion
p: preference threshold
q : indifference threshold
The discordance check is positive when gj(b) – gj(a) ≤ vj[gj(a)] – ηjqj[gj(b)]
V: veto threshold
Where nj= 1−𝐶(𝑎,𝑏) −𝑤𝑗
1−𝑠−𝑤𝑗
For the elicitation of the criteria weights, the Simos method is used. It is a simple but effective
method that facilitates the Decision Maker express easily his/her preferences over the importance
of criteria, with the use of a deck of cards, (see Simos 1990a, for more details on the method).
5. Implementation
Before the implementation of the algorithm the parameters of the model need to be determined.
First of all, the weights, which are calculated based on the Simos method (Simos, 1990a and Simos,
1990b), the results of which are presented in Table 4. Figure 1 below illustrates the deck of criteria
cards, as arranged by the investor in an ascending order of importance. It should be noted here,
that the after the arrangement of the cards, the weights were not calculated using the mathematical
procedures of Simos, due to a number of robustness issues they bear (see Figueira and Roy, 2002,
and Siskos and Tsotsolas, 2014). Instead, the system of inequalities was solved maximizing the
most importance criterion, namely the revenues.
Book of Proceedings
97
Figure 7: Simos results
Secondly, the veto thresholds, indifference and preference levels are estimated after dialogue with
the DM. All these values are presented along with the criteria values in Table 4.
Table 4: Criteria values and thresholds
Alternative g1 g2 g3 g4 g5 g6 g7 g8 g9
a 4 2,0 406 3 5 18 4 15 3
b 4 2,2 377 3 3 15 4 15 3
c 2 5,5 123 2 5 8 4 5 4
d 3 6,3 129 3 4 5 4 5 4
e 3 7,7 127 5 4 48 2 25 2
f 3 4,0 385 2 1 10 2 40 4
g 3 4,7 357 3 1 10 2 40 4
h 3 15,2 139 3 1 48 3 60 3
i 3 18,1 218 4 2 38 3 4 4
Weights 0,072 0,172 0,224 0,112 0,132 0,132 0,072 0,032 0,052
Indifference
threshold - 1,5 20 - - - - - -
Preference
threshold - 5 70 - - - - - -
Veto
threshold - 10 170 {2,5} {1,5} - - - -
Book of Proceedings
98
After the execution of the two steps of the algorithm, namely the concordance and discordance
checks, we obtain the following matrices, (see Tables 5 and 6).
Table 5: Concordance check (s =0,7)
a b c d e f g h i
a x 0,82 0,72 0,70 0,55 0,89 0,75 0,50 0,53
b 0,70 x 0,73 0,69 0,42 0,90 0,87 0,61 0,53
c 0,43 0,43 x 0,82 0,62 0,43 0,43 0,48 0,29
d 0,41 0,43 0,74 x 0,65 0,54 0,43 0,48 0,29
e 0,45 0,58 0,74 0,88 x 0,69 0,69 0,67 0,48
f 0,47 0,48 0,74 0,60 0,37 x 0,82 0,48 0,38
g 0,46 0,48 0,74 0,74 0,48 0,97 x 0,48 0,38
h 0,50 0,50 0,74 0,74 0,68 0,72 0,65 x 0,41
i 0,47 0,47 0,71 0,71 0,52 0,69 0,62 0,76 x
Table 6: Discordance check
Pairs of alternatives g2 g3 g4 g5
a,b 1 1 1 1
a,c 1 1 1 1
a,d 1 1 1 1
a,f 1 1 1 1
a,g 1 1 1 1
b,a 1 1 1 1
b,c 1 1 1 1
b,f 1 1 1 1
b,g 1 1 1 1
c,d 1 1 1 1
Book of Proceedings
99
d,c 1 1 1 1
e,c 1 1 1 1
e,d 1 1 1 1
f,c 1 1 1 1
f,g 1 1 1 1
g,c 1 1 1 1
g,d 1 1 1 1
g,f 1 1 1 1
h,c 0 1 1 1
h,d 1 1 1 1
h,f 0 0 1 1
i,c 0 1 1 1
i,d 0 1 1 1
i,h 1 1 1 1
All the pair values in the concordance check, surpassing s =0,7 take part in the discordance check.
In particular four pairs (red colored) exhibit positive check.
The rest of the outranks between the alternatives are presented at the outranking graph below (see
Figure 2). The Figure also showcases the core of Electre IS (Π) including the best alternatives,
which are not outranked by any other.
After dialogue with the investor and a final analytical review of the specific values of the four best
alternatives over the evaluation criteria, the alternative b is rejected. Finally, the decision maker,
after having been informed about all the information for a,e and I, selected the photovoltaic
(interconnected) as the best alternative. The procedure followed is presented in Figure 3.
Book of Proceedings
100
Figure 8: Outranking graph
Figure 9: Decision support procedure and final selection
6. Conclusions
The study presented in this short paper, achieves the modeling and solution of the power generation
from RES investment evaluation problem, using a synergy of MCDA methods. It also achieves an
effective implementation of the Simos method to infer indirectly the criteria weights. The
Book of Proceedings
101
evaluation system (points of view, dimensions and criteria) includes all the parameters that affect
an investment on electricity production.
The optimal alternative, as resulted using the preferences of a fictitious decision maker/public
investor, is the interconnected Photovoltaic power plant. The use of an MCDA method, such as
the ELECTRE IS, that is able to handle pseudo-criteria, gives rational results, taking into account
the indifference in minor differentiations of the values of the alternatives on certain quantitative
criteria. The final results emerged after the appropriate implementation of the synergy of
multicriteria methods, namely the ELECTRE IS and the Simos. They were then set under careful
and meaningful dialogue with the investor, who expressed her/his preferences regarding the
importance of the criteria and the final selection of the one among the four final optimal
alternatives. As a result, the solution surfaced as an algorithmic-procedural solution combined with
the decision maker’s judgment.
Regarding some potential future perspectives of this study, it could be implemented for the case
of another country or area of interest, where it would be interesting to test how different data and
parameters diversify the final results. Also a different modeling of the problem or differentiation
of the objective (i.e. selection of the optimal investment portfolio) can be applied or even an
extension of the alternatives by adding incorporation of RES power plants, energy saving actions,
cogeneration of heat and electricity and nuclear power. Finally, the problem could also account
the preferences of multiple DMs, with different viewpoints and magnitudes in their opinions. In
that case, it is critical to integrate adequately these preferences, using elaborative mathematical
models, and result to a compromise solution that is to be accepted by all.
References
Burton, J. Hubacek, K. “Is small beautiful? A multicriteria assessment of small-scale energy
technology applications in local governments”, Energy policy, 35, 2007, pp. 6402-6412.
Cavallaro, F. “An Integrated Multi-Criteria System to Assess Sustainable Energy Options: An
Application of the Promethee Method”, (2005).
Retrieved online: http://hdl.handle.net/10419/74297.
Figueira, J., S. Greco, and M. Ehrgott (Eds.). “State-of-Art of Multiple Criteria Decision
Analysis”, Dortrecht: Kluwer Academic Publishers, 2005.
Figueira, J. and Roy, R. “Determining the weights of criteria in the ELECTRE type methods
with a revised Simos’ procedure”, European Journal of Operational Research, 139, 2002, pp.
317-326.
Book of Proceedings
102
Georgopoulou, E., Lalas, D., Papagiannakis, L. “A multicriteria decision aid approach for energy
planning problems: The case of renewable energy option”, European journal of Operational
Research, 103, 1996, pp. 38-54.
Kahraman, C., Kaya, I., Cebi S. “A comparative analysis for multiattribute selection among
renewable energy alternatives using fuzzy axiomatic design and fuzzy analytic hierarchy
process”, Energy, 34, 2008, pp. 1603-1616.
Karakosta, C., Pappas, C., Marinakis, V., Psarras, J. “Renewable energy and nuclear power
towards sustainable development: Characteristics and prospects”, Renewable and Sustainable
Energy Reviews, 22, 2012, pp. 187-197.
Kaya, T., Kahraman, C. “Multicriteria decision making in energy planning using a modified
fuzzy Topsis methodology”, Expert systems with applications, 38, 2010, pp. 6577-6585.
Løken, E. “Use of multicriteria decision analysis methods for energy planning problems”,
Renewable and sustainable energy reviews, 11, 2005, pp. 1584-1595.
Naim, H., Afgan, M., Carvalho, G. “Multi-criteria assessment of new and renewable energy
power plants”, Energy, 27, 2001, pp. 739-755.
Polatidis, Η., Haralambopoulos, A., Munda, G., Vreeker, R. “Selecting an appropriate multi-
criteria decision analysis technique for renewable energy planning”, Energy sources, Part B:
Economics, Planning, and Policy, 1, 2006, pp. 181-193.
Roy, B. Bouyssou, D. Aide multicritère à la décision: Méthodes et cas, Paris: Economica, 1993.
Simos, J. “L’évaluation environnementale: Un processus cognitif négocié”. Thèse de doctorat,
DGF EPFL, 1990a, Lausanne.
Simos, J. “Evaluer l’impact sur l’environnement: Une approche originale par l’analyse
multicritère et la négociation”, 1990b, Presses Polytechniques et Universitaires Romandes,
Lausanne.
Siskos, E., Tsotsolas, N., Christodoulakis, N. “Elicitation of criteria importance weights through
Simos method: A robustness concern”, Under Revision.
Tsoutsos, T., Drandaki, M. Frantzeskaki, N., Iosifidis, E., Kiosses I. “Sustainable energy
planning by using multi-criteria analysis application in the island of Crete”, Energy policy, 37,
2009, pp. 1587-1600.
Wang, J-J., Jing, Y-Y., Zhang, C-F., Zhao J-H. “Review on multi-criteria decision analysis aid in
sustainable energy decision-making”, Renewable and sustainable energy reviews, 13, 2009, pp.
2263-2278.
Book of Proceedings
103
Orisma(c): Optimizing long term fleet wide crew assignment
Takis Varelas
Danaos Research Centre | Akti kondili 14, Piraeus 18545, Greece | [email protected]
Sofia Archontaki
Danaos Research Centre | Akti kondili 14, Piraeus 18545, Greece | [email protected]
Myrto Livadioti
Danaos Shipping Co Ltd | Akti kondili 14, Piraeus 18545, Greece | [email protected]
Abstract
We present the unique long term maritime crew planning and assignment optimization that Danaos
Corporation envisaged, Danaos Management has implemented and Danaos Shipping deployed as
an enrichment of its ORISMA5 (Operation Research In Ship Management) toolkit. The major
novelty in this system is the extension of the two coordinates, the number of vessels from one to
whole fleet and the time horizon from couple of weeks to several months. Another initiative is the
addition of a third dimension the teamwork index of the vessels’ management team. As assignment
optimization problem the definition of an objective assignment function that should be optimized
is required. We analyzed the problem and found out the formulas and the variables that are needed
for the calculation of coefficients in the identified individual objectives that are combined in a
weighted multi-objective assignment penalty function. The extension of coordinates and the team
working dimension increases the problem complexity and is hard to achieve optimal solution with
conventional heuristics. So we combine operation research genetic and muliti-index axial integer
models, efficient assignment algorithms and, new developed ones into one model adjusted to
specific problem requirements.
System also supports strategic decisions regarding the depth determination of the availability
officers’ pool, the entries’ attributes such as rank, performance, availability and nationality
avoiding in one hand unfeasible solutions and keeping on the other the safety pool level as less as
possible. Furthermore an alert mechanism generates the appropriate triggers for actions whenever
safety levels are reached and suggests mitigation plan.
5 ORISMA is awarded with the highest distinction in Operation research from INFORMS, the Institute of Operations Research and Management Science, the Franz Edelman finalist award for 2012.
Book of Proceedings
104
From design point of view the most important novelty of the ORISMA approach is the usage of
ship officers’ quadruplet as the monitoring entity instead of the individual ship officer.
Keywords
Assignment problems, Integer programming, quadratic assignment, multi-criteria analysis
1. Problem description
Compared with other human resources management systems maritime crewing is highly complex
because of particularities of marine profession. Seafarers usually are not work permanently and
are employed on period basis. A typical employment period is on about 4 months. Crew synthesis
may be multi-national and multicultural. Company compiles crew scheduling taking into account
the maximization of the crew retention factor. Furthermore a crew appraisal scheme is utilized to
evaluate the seafarers’ performance as well as the vessel-seafarer suitability. The main goal is to
find the proper person for the proper vessel in right time with the highest confidence, keeping the
seafarers availability pool as small as possible as well as maximizing the seafarers’ satisfaction
and reducing the non-productive cost of idle time between the ESO (estimated sign off) and AF
(available from) dates. In our days the most known conventional approach is the monitoring of
the debarkation list and the assignment from the stand-by (availability list) of the most suitable
seafarer per vessel. Additionally crew operator, based on his expertise, may take another decision
even shifting in or out the debarkation date of the onboard officers, even shifting in or out the
seaman availability date. He also takes into account the co-existence coefficient of the eventual
synthesis that may raise communication and/or performance problems. Despite the fact that is
approach is quite adequate for one vessel in a small time horizon that we need is to view the whole
picture to support the holistic better solution. In this direction we develop, implement and deploy
the ORISMA(c) solution with impressive results specifically when the fleet population is larger
than 20 vessels. Indicatively the rejoin cost may be reduced on about 2M $ /year for a fleet with
50 vessels.
Companies utilize their own appraisal schemas to evaluate the performance of the officers.
Sophisticated schemas incorporate several assessors and assessment criteria. Evaluation grade
may be expressed qualitatively or quantatively. The evaluation quantitative grade for each officer
may be formulated as weighted multi-criteria function or by simple assignment.
Book of Proceedings
105
Anyway an officer has a performance grade based on his
professionalism, experience and other attributes. Additionally
each officer has a different suitability factor for different vessels.
Finally each officer has different coexistence potential with
other seafarers. Hereafter is substantial to formulate the
assignment index taking the aforementioned factors the
individual officer performance, the officer-vessel suitability and
the officer-officer coexistence level (the teamwork index). The
ship-officer suitability depends on nationality, experience in the
same, sister or similar vessel type or engine type in case of engineers. Significant suitability factor
is the relation of EOS (estimated sign on-off) of the officer, who is going to be replaced and the
AF the availability date for embarkation.
There are several objectives in crew scheduling. One of them is to keep the retention factor (RF)
in the highest level. According to TMSA (tanker management self-assessment) a more than 80%
crew retention is required to achieve the highest assessment stage (Crew retention KPI key
performance indicator=4).
Another important objective is the definition of officers’ pool per rank size. The size, or otherwise
the depth, of each pool should be as small as possible but adequate enough to assure the
embarkation requirements fulfillment. The smaller feasible size assures the higher crew
confidence level for next in proper time employment. Of course important object is the crew
assignment cost in monetary terms that may be formulated according to crew management policy.
All the above objectives (optimization of retention, satisfaction, cost and quality) are combined in
one goal programming function adjusted to each company that should be optimized.
2. Conclusions
The solution has been tested successfully. System is utilized from several companies and the
feedback evaluation is impressive in terms of tangible as well intangible benefits. The system
power originates from the efficiency of the incorporated algorithms which provide the assignment
map within seconds. So operators are able to run the program with different scenarios using a user
friendly interface. In these scenarios may select criteria such as vessels, vessel nationalities, fleets,
ranks and any combination of them. They may also alter system proposals. This interactivity
provides the appropriate feedback for system calibration. System is also customizable. In the
setup the availability window may be changed and, penalty tables may be altered. Finally an
intelligent module may provide not only the optimum solution but proposed solutions in
descending order based on specific criteria.
A build-in multi criteria analysis model would be useful in the definition of weights of the used
individual attributes of the objective function coefficients.
vessel
officer
oficer
assignment index
Book of Proceedings
106
3. Objective, Indexes and formulas
3.1 Teamwork index (The highest index the lowest co-existence)
Four senior officers (n=4) are employed at any time in each vessel. The seniors’ quadruplet q for
a given period consists of the Captain (m), the Chief Engineer (c), the chief officer (o) and the
second engineer(s). Each of them has his individual g grade as an integer value (1 to 4). There are
three correlation matrices mc; mo. and cs were the value of each matrix cell is the penalty of co-
existence of the two senior grades as coordinates. We assume the penalty coexistence index as the
sum of the three correlation cells:
(1) For each q{m,c,o,s) { pq= mc(g(m),g(c))+mo (g(m),g(o))+cs(g(c),g(e)) }
The quadruplet is changed whenever a senior demarcates and is replaced. Assuming that
employment period is fixed (f) then we will have at most n+1 different q’s. Their corresponding
time intervals are as follows:
(2) ed0 = current date: edn+1=ed1+f : For i=1 to n+1 { di=edi-edi-1 }
The team-working index from initial to final stage when all the quadruplet members will have
replaced should take into account the duration of each q and is calculated as
(3) P= i=1Σn+1(Pi* di)/ (ed1+f - ed0)
3.2 Availability index
Whenever we have an open position for embarkation among two candidates with the same
specification we reasonably select the seafarer who is first in the waiting list. So we define an
availability bonus and consequently an associated penalty of short waiting time. If a seafarer is
waiting more than one month it is urgent to find an employment for him. On the other hand in case
we are unable to find an available officer for an open decision operator may contact a seafarer who
is in vacation period to assess the possibility for embarkation. So we need to incorporate in the
availability penalty index this alternative with some penalty of course.
4. The ORISMA(c) optimized mode
The developed model improves dramatically the elapsed time and at the same time assures the
optimum solution. The main initiatives are as follows:
Step-1: Generate variables
For each v vessel in a fleet there are feasible i alternative vessel officers’ quadruplets qi with
assignment penalty piv. Feasible quadruplets are generated from the entries of the four (one for
each rank) availability pools that could be assigned to each vessel. With this filtering q’s number
is much less of the generated n4 quadruplets from the combinations of all elements of the
mentioned normalized pools (n=max of pools depth). Consequently the number of variables x is
iv and denote the of the corresponding v-q assignment truth (x=1) or not (x=0).
Book of Proceedings
107
Step-2: Formalize objective
𝑜𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 ∶ 𝑚𝑖𝑛∑∑𝑝𝑖𝑗 𝑥𝑖𝑗
𝑞
𝑗=1
𝑣
𝑖=1
Step-3: Formalize assignment constraints one for each officer
The sum of variables, where an officer or is involved, should be 0 or 1 and not only 1 or in other
words a candidate officer may be assigned once or may not be assigned. It is in practice explainable
because usually we have more candidates that open positions. So we replace the right hand side of
all assignment constraints for all officers from “=1” to “<=1”.
∀ 𝑜𝑟 𝜖 𝑞𝑖𝑗 | 𝑖 = 1: 𝑣 , 𝑗 = 1: 𝑞(𝑖) 𝑥𝑖𝑗 ∈ { 0,1} & 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑟 = 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑟 + 𝑥𝑖𝑗
∑𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑟 ≤ 1
𝑟
Step-4: Formalize assignment constraints one for each vessel
It is obligatory to assign on quadruplet to each vessel. But in practice whenever availability pools
have been kept small sometimes we may not be able to find quadruplets to all vessels without
conflicts and the mentioned model will only notify infeasibility without explanations. This
infeasibility is too hard to be analyzed manually. To overcome this weak point we introduce in the
objective function a dummy xio =x[v]000000 where v is the vessel number with a high value h
coefficient and we insert this variable in the corresponding vessel assignment constraints.
4.1 Final model
q (i) : number of feasible q’s for vessel i, h: high value penalty
pij : assignment penalty of quadruplet j to vessel I
xio : dummy variable for vessel i, v : number of vessels, or : officer r
𝑜𝑏𝑗𝑒𝑐𝑡𝑖𝑣𝑒 ∶ 𝑚𝑖𝑛∑ ∑ 𝑝𝑖𝑗 𝑥𝑖𝑗 + ∑ ℎ𝑣𝑖=1
𝑞𝑗=1 𝑥𝑖0
𝑣𝑖=1
∑ 𝑥𝑖𝑗 + ∑ 𝑥𝑖0 𝑉𝐼=1
𝑞(𝑖)𝑗=1 = 1 ∀ 𝑖 = 1: 𝑣
Q={qij |∀ 𝑜𝑟 𝜖 𝑞𝑖𝑗 | 𝑖 = 1: 𝑣 , 𝑗 = 1: 𝑞(𝑖) 𝑥𝑖𝑗 ∈ { 0,1} & 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑟 = 𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑟 + 𝑥𝑖𝑗
∑𝑐𝑜𝑛𝑠𝑡𝑟𝑎𝑖𝑛𝑡𝑟 ≤ 1
𝑟
Book of Proceedings
108
References
Balas, E. and Saltzman, M.J. (1991), An algorithm for the three index assignment problem.
Operations research 39:150-161
Hahu, P, Grant, T., and Hall, N. (1998), Solution of the quadratic assignment problem using the
Hungarian method. European Journal of Operational Research, 108:629-640
Haln, P.and Grant, T. (1998) Lower bounds for the quadratic assignment problem based upon a
dual formulation. Operations Research 46:912-922
Karp,R.M (1980) An algorithm to solve the m x n assignment problem in expected time o(mn
logn). Networks, 10:143-152
Varelas O, Archontaki S (2011) Intelligence in crew option systems. 3rd International.
Symposium on. Ship operations, Management and Economics. (Society of Naval Architects and
Marine Engineers (SNAME), Jersey city, NJ pp 291-296j
Varelas P et al. (2013): Optimizing ship routing to maximize fleet revenue at Danaos Interfaces
43(1), pp 1-11
Book of Proceedings
109
Simulation analysis of a pilot handling system for the rail transport of
conventional semitrailers
A. Ballis
Department of Transportation Planning and Engineering
National Technical University of Athens
5, Iroon Polytechniou, GR-15773, Athens, Greece
Tel: +30.210.772 1235 Fax: +30.210.772 2404 E-mail:
Abstract
The transport of semi-trailers by rail is of foremost importance due to the fact that it allows for the
smooth shifting from road haulage to the intermodal/combined transport operations. ISU is a pilot
system allowing for the loading/unloading of semi-trailers on trains by picking them from their
wheels (therefore by applying the same forces as when they are sitting-on during road transport).
Within the European project CREAM effort was given to the analysis and improvement of the
operations and the configuration of the ISU system aiming to develop the “next generation” of this
equipment. Field observations and a simulation - based analysis were performed by the NTUA
research team. The scope of the current presentation is to outline the methodological framework
used for the above simulation analysis as well as to present the early results of the research work.
Keywords SIMULATION, HANDLING SYSTEM, RAIL TERMINAL, SEMI-TRAILERS.
1. INTRODUCTION
Semi-trailers are essential components of the road traffic and thus also of the pre- and post-haulage
legs of the intermodal/combined transport chains. The use of semi-trailers as loading units in the
railway transport has a long record, either as cranable semi-trailers (that are loaded on special
wagons by cranes or reach stackers) or as conventional (non-cranable) semi-trailers that are loaded
on trains using various horizontal handling techniques (rolling motorways, the Modalohr system
or the ISU system). The scope of the current presentation is to outline the methodological
framework used for the simulation analysis of the ISU system and to present the early results of
the research work. Section 2 presents the above conventional and innovative systems revealing the
need for research in this sector. In Section 3 the methodological approach (field observations,
model structure, scenarios investigated) is presented together with the early results of the analysis.
The last Section hosts the conclusions and the future steps of the research.
2. THE NEED FOR INNOVATION IN THE RAIL TRANSPORT OF SEMITRAILER
A semi-trailer is a “goods road vehicle with no front axles designed in such a way that part of the
vehicle and a substantial part of its loaded weight rests on the road tractor” (i). The use of semi-
Book of Proceedings
110
trailer as a loading unit in the European railway transport has a long record, yet while in road
transport the number of semi-trailers has been constantly growing, in rail transport the share of
semi-trailers is declining (ii,iii). The rail transport of semi-trailers is of primary importance due to
the fact that allows for the smooth shifting from road haulage to the intermodal/combined transport
operations, given that certain organizational problems are solved(iv). To this aim, legislative
measures can be used (and are used) to reduce the transport of semitrailers by road (e.g. lorry
tolling systems according to heavy goods vehicle emissions, strict monitoring of maximum driver
working hours (v,vi), etc. In parallel, research is carried out in order to improve technical and
organisational aspects that improve the efficiency of the rail transport of semi-trailers. One critical
technicality concerns the methods used to load/unload semi-trailers on railway wagons. The most
known methods are:
o Cranable semi-trailers. The term cranable semi-trailer (or grapple arm semi-trailer) stands for a
special semi-trailer type having adequate strength to be engaged in a lifting operation by a crane
having a grapple-arm spreader. Cranable semi-trailers are in accordance with the normal road-
going specifications and legislative dimensions (length: 13.6m, height: 4m, width: 2.50-2.60
m) but in addition, are being reinforced to withstand the stresses of being lifted in a laden state
from road into rail and vice versa. For this purpose they have strengthen chassis, modified lift
suspension, hinged underrun protection device and are fitted with lifting pockets in the
underside to host the crane’s grapple arms. The disadvantages of the cranable semitrailers in
comparison with the conventional ones are the higher cost and (mostly) the higher tare weight
that reduces the maximum permitable cargo weight by about 500 kg.
o Rolling Motorways. A “Rolling Motorway” or “Rollende Landstrasse” or Ro-La is a system
where complete road vehicles are driven onto special rail wagons with small wheels that are
forming a long platform where road vehicles are rolling on and off(vii). Rolling motorways are
used in the European transalpine services across Switzerland and Austria which restricted the
number of heavy vehicles transiting the country each year. The disadvantage of Rolling
Motorways is that (a) the lorry must also travel with the semitrailer and (b) that the small wheels
are wearing much more than the wheels of the ordinary wagons.
o The Modalohr system. This innovative system can carry trucks or semi-trailers by use of
specifically designed low-floor articulated railway wagon. The upper deck of each wagon can
rotate so that semi-trailers can roll on by use of road tractor(viii). The Modalohr system was used
in the France-Italy transalpine corridor and in the Luxembourg to Perpignan (French) route.
The disadvantage of this system is that it requires special expensive wagons and dedicated
terminals (with ramps allowing the roll on and off of semi-trailers to wagon decks) in both ends
of the journey.
o The ISU System. This innovative system was designed to allow the transport of existing
conventional semi-trailers (non-cranable) by picking them from their wheels. A prototype has
been developed and demonstrated in Wien Nordwest terminal as well as in Wells terminal in
Austria. The pilot run of the system has been performed on the Austria –Turkey route as a part
of activities performed within the European CREAM project (xi). The detailed description of
ISU system is included in the following Section.
Other innovative ideas (mostly on conceptual stage or as drawings) exist, yet the thorough
presentation of all concepts and ideas published about semi-trailer handling equipment, is outside
the scope of the current presentation.
Book of Proceedings
111
Figure 10: Conventional tractor-semitrailer combination (upper part) and components of ISU
system
3. TRAIN SERVICE PROCESSES AND SIMULATION OF ISU SYSTEM
ISU is the abbreviation of the German “Innovativer Sattelanhaenger Umschlag” that means
“innovative semi-trailer transhipment”. ISU system was invented by two German rail experts(ix)
that reached an agreement with Rail Cargo Austria for the pilot use of the system. The system was
presented in the framework of the BRAVO project (x) and further was investigated and improved
within CREAM project (xi). To allow for the transport of conventional semi-trailers ISU is picking
them from their wheels, e.g. by applying the same forces as when they are sitting-on during road
transport. The system (see Figure 1) requires a specific (but existing) wagon type, a number of
wheel-packer elements, a transverse beam and an adapter (an auxiliary frame that can be mounted
to any spreader, hoisting ropes are connected to a traverse beam and wheel-packer elements). The
wheel-packer elements are located in a special ramp or in the ground (ISU-ramps); the semi-trailer
is moved into the wheel-packer elements and is lowered to the jockey wheels. After the ropes have
been mounted by the groundsman the vertical transhipment is performed by the crane into the
pocket wagon, where the ropes are detached and loading is completed(x). The wheel-packer
elements remain in the wagon to facilitate the unloading procedure at the destination terminal.
The handling system of ISU deviates a lot from the common conventional handling operations
(due to the necessity for wheel-packer elements and the transverse beam that must be moved
between wagons and the ISU ramp). The service cycle of ISU is largely dependent on the initial
conditions (wheel grippers in ramp slots or in the wagon, reach stacker near the ramp or near the
railway line) as well as by the sequence of handling requests (loading or unloading operations).
Field observations and preliminary analysis revealed that any complex train service operations can
be analysed in combinations of 4 elementary processes. Figure 2 presents graphically the initial
conditions and operations of ISU systems handling a loaded truck after having served another
loaded truck. Different operations exist in other cases (loaded truck followed by empty truck,
empty truck followed by empty truck, empty truck followed by loaded truck).
In order to investigate the parameters affecting the performance of ISU system and to propose
improvement for ISU equipment, a simulation-based analysis has been performed by a research
team of the National Technical University of Athens (NTUA) participating in the CREAM project
consortium. Field observations and evaluation of the analysis was facilitated and supported by a
research team of Rail Cargo Austria that also participated in the consortium.
Book of Proceedings
112
The simulation model was developed in ARENA software with certain complex modules in
Simon language. The model has advanced animation capabilities: (a) tractors and trailers are
presented using computer graphics (see Figure 2, middle part) while handling operations are
presented using videos. Each time a handling operation is performed the programme retrieves and
displays the relevant video recording of the pilot ISU operation in Wells. Video playing time is
synchronised with the simulation clock so that tractor and trailer movements are synchronised with
video pictures. The simulation analysis revealed that the performance of the system is largely
dependent on tractor arrival sequence therefore, proper queuing discipline rules have been
introduced to improve system’s performance. Furthermore, the following technical modifications
have been simulated for various tractor arrival patterns.
o Special gripping apparatus to reduce handling time and labour fatigue
o Stand alone spreader with supporting legs. This way Reach Stacker can decuple from the
spreader (that remains on the ramp) to perform other tasks using a second ISU spreader. This
option affects positively the service cycle time.
o Anti-sway system
o New ramp design allowing for fast repositioning in various terminal locations
The embedded Table in Figure 3 outlines the results of the analysis. The “Moderate revision”
option has a fair cost and good performance yet the additional weight of the extra apparatus (legs,
anti-sway) increases the weight of the ISU spreader and therefore reduces the maximum weight
that can be handled by the Reach Stacker. To overcome this disadvantage, a new lighter frame
design is required to compensate for the additional weight. The “Major revision” option includes
all proposed modifications at the expense of higher cost.
ISU ramp Vehicle
Reach
Stacker
Initial Conditions
for the current
operation
Empty slots
Tractor
with Semi-
trailer
Near
railway
line
Final Condition for
the current
operation
“Wheel
grippers”
and “lifting
beam” in
place
Departs
Near
railway
line
1. Empty Travel
Previous
operation that
sets the initial
conditions
(semitrailer to
be unloaded)
Current
operation
(semitrailer
to be
unloaded)
Book of Proceedings
113
Figure 2: Initial conditions and operations of ISU systems handling a loaded truck after having
served another loaded truck. Different operations exist in other cases (loaded truck
followed by empty truck, empty truck followed by empty truck, empty truck followed by
loaded truck).
3. Reach Stacker detaches “wheel grippers” and “lifting
beam” in the associated ramp slots
4. Tractor with semi-trailer rides the ramp
5. The driver disconnects the semi-trailer from the tractor
6. Handling (Reach Stacker picks up semi-trailer and moves
to the pocket wagon)
7. Tractor departs
2. Reach Stacker picks
up “wheel
grippers” and
“lifting beam”
8. Handling (Reach Stacker lifts
semi-trailer into the pocket
wagon)
Book of Proceedings
114
Figure 3:
Proposed
modifications, simulation analysis and early results
4. CONCLUSIONS
Semi-trailers are important elements of the intermodal transport system as they allow for the
smooth shifting from road haulage to the intermodal transport chain. The rail transport of semi-
trailers have been implemented in Europe through various methods, yet more research is required
as the existing handling techniques suffer from certain drawbacks. ISU system is a new proposal
that is currently in pilot stage. Field observations and simulation-supported analysis performed
Minor Revision Moderate Revision Major Revision
Spreader with supporting
legs
Anti – sway
Gripping apparatus
New frame design
Mobile Ramp
Service cycle -1% -30% -35%
Labour fatigue -25% -50% -50%
Loading capabilities Reduced
Cost Fair high
Book of Proceedings
115
within the framework of a European research project concluded to a number of modifications that
will improve significantly the performance of the system. The research teams of NTUA and Rail
Cargo Austria have agreed to continue the research towards the development and testing of a
prototype derived from the current analysis.
ACKNOWLEDGEMENT
The current research was performed as a part of the European research project CREAM (2006 –
2001) which was financed by the 6th EC Framework Programme.
References
1 United Nations, Glossary for transport Statistics, 3rd edition 2003 1 S A I L project. Semitrailers in Advanced Intermodal Logistics, Analysis of the State of the
Art Deliverable 1, November 2000, http://www.zlw-ima.rwth-
aachen.de/forschung/projekte/sail/index.html. 1 UIRR. Statistics 2007, UIRR International Union of combined Road-Rail transport
companies, http://www.uirr.com 1 National Technical University of Athens, UIRR and CEMAT. Investigation of Greek
Transport demand for the Combined Transport corridor Greece-Italy-Germany, Pilot Project
Financed by European Commission DG VII, 1994. 1 RoadTransport.com EU Drivers' Hours explained, 06 May 2008,
http://www.roadtransport.com 1 EC Regulation 561/2006 on the Harmonisation of Certain Social Legislation Relating to
Road Transport and Amending Council Regulations (EEC). No 3821/85 and (EC) No
2135/98 and Repealing Council Regulation (EEC) No 3820/85, 2006. 1 Lowe D. Intermodal Freight Transport, Elsevier, 2005. 1 Lohr Groupe, Modalohr Presentation. Available at: http://www.oevg 1 Official site of ISU system at: http://isusystem.de 1 BRAVO project. Information available at: www.bravo-project.com . 1 CREAM project “Customer-driven Rail-freight services on a European mega-corridor based on Advanced business and operating Models”.
Final Report available at: http://www.cream-project.eu/home/index.php
Book of Proceedings
116
Research on internet sufficiency of websites concerning women agricultural
co-operatives in Greece: A multicriteria approach
Athanasios Batzios
Lab. of Agricultural Informatics, School of Agriculture, Faculty of Agriculture, Forestry and
Natural Environment, Aristotle University of Thessaloniki, Greece, [[email protected]]
Thomas Bournaris
Lab. of Agricultural Informatics, School of Agriculture, Faculty of Agriculture, Forestry and
Natural Environment, Aristotle University of Thessaloniki, Greece, [[email protected]]
Zacharoula Andreopoulou
Lab. of Forest Informatics, School of Forestry and Natural Environment, Faculty of Agriculture,
Forestry and Natural Environment, Aristotle University of Thessaloniki, Greece,
Christos Batzios
Laboratory of Animal Production Economics, School of Veterinary Medicine, Faculty of Health
Sciences, Aristotle University of Thessaloniki, Greece, [[email protected]]
Basil Manos
Lab. of Agricultural Informatics, School of Agriculture, Faculty of Agriculture, Forestry and
Natural Environment, Aristotle University of Thessaloniki, Greece, [[email protected]]
Abstract
This paper deals with the research of the internet sufficiency of websites concerning women
agricultural co-operatives, through empirical research for the assessment of criteria/characteristics
of relative websites. Towards this direction, the basic criteria/characteristics of a website were
identified and then, an empirical research was performed in a sample of 30 websites of women co-
operatives. These websites evaluation was accomplished through specific navigation
characteristics, design, interactivity, accessibility, e-services and usefulness of the information
provided, that reflect the internet sufficiency of these websites. The data that derived from the
empirical research were used to evaluate the fulfillment rate of the various criteria/characteristics
for the internet sufficiency of the websites of women co-operatives. Multicriteria analysis was
further performed aiming to hierarchically classify and rank the websites towards the total net flow
of internet sufficiency. The results of the research out sketch the “profile’ of internet presence and
promotion of women agricultural co-operatives. The conclusions of the paper can comprise useful
consulting tools and contribute in a more rational organization of websites for the promotion of
women co-operatives and generally in the effective development of women entrepreneurship in the
agricultural production sector.
Key words: women agricultural co-operatives, websites, website evaluation criteria, multicriteria
analysis
Book of Proceedings
117
1. Introduction
Women agricultural cooperatives constitute an agricultural business sector that functions under
specific rules and legislation due to its collaborative character, that differentiates that sector from
other business sectors (Gidarakou et.al., 2000). However, women co-operatives often lack in
effective management and they face problems on product packaging, quality control, brand name,
etc. affecting the marketing of the products. The problem of products disposal seem also to be
important for most of women co-operatives due to their distance from urban centers and their
markets.
Within the current economic recession and the high unemployment rates in Greek economy,
women co-operatives can play an economic and social role as a factor for women employment,
local economy and agro-tourism. Lately, rural agro-business tend to adopt new flexible ways for
their promotion, while their corporate website is the most contemporary means for their internet
promotion.
In order to successfully accomplish a contemporary and competitive internet presence for the
women agricultural cooperative business, effective web design and structure is a prerequisite
within the context of powerful cooperative promotion. Today, women’s co-operatives presence in
the internet is not satisfactory, as their majority does not even have a website or they are hosted in
other local websites while there is recent research findings and proposed models for website
evaluation (Andreopoulou et.al. 2014, Chatzinikolaou et.al, 2013; Tsekouropoulos et.al, 2012;
Kargioti et.al, 2006; Patsioura et al., 2004).
The necessity for analytical research on the characteristics of internet sufficiency of women co-
operatives websites in combination to their special role in the socio-economic status of our country
was the trigger for that research. The aim of the paper is the research on the internet sufficiency of
websites of women agricultural co-operatives, through empirical analysis and evaluation of the
websites’ criteria/characteristics. The research focus in the following targets: Identification of the
basic criteria/characteristics for the evaluation of women agricultural co-operatives, Evaluation of
the accomplishment rate for the various criteria /characteristics for the internet sufficiency of
websites and Multicriteria analysis and hierarchical classification and ranking of these websites
towards their sufficiency.
2. Research Methodology
Based on bibliography, some basic identified criteria/characteristics were chosen and grouped in
six basic multi-criteria categories (Table 1). The research on the internet of women co-operatives
has been made using specific relevant key-words and their combinations through «Goοgle» search
engine. According this, the proper population of active women co-operatives with (proper) website
is Ν=39 co-operatives, in a total of Ν΄=159 co-operatives active today in Greece. Through random
sampling, 30 websites were collected to form the research sample using random numbers. In the
form-questionnaire for the research evaluation were also included general info questions and 30
questions about the basic criteria/characteristics of internet sufficiency structured in 6 categories
in YES/ NO type.
The total accomplishment rate for the various criteria/characteristics is evaluated through the
evaluation form. Further, multicriteria analysis PROMETHEE II was performed to generate a total
ranking of the websites in the sample.
Book of Proceedings
118
Table 1: Criteria for the evaluation of women agricultural co-operatives websites
3. Results
Bellow are the results of Statistical Data analysis in Criteria/ Characteristics of:
Navigation: Average accomplishment Rate: 41,34 ±28,85%
Web-design: Average Accomplishment Rate: 57,34 ±30,40%
Interactivity: Average Accomplishment Rate :38,02±31,77%
Accessibility: Average Accomplishment Rate 52,00±45,49%
E- Services: Average Accomplishment Rate : 8,02±5,07%
Information Usefulness: Average Accomplishment Rate: 31,34±25,09%
The total ranking of the websites using multicriteria analysis (PROMETHEE II) was based on the
score of total net flow of their internet sufficiency (Doumpos & Zopounidis, 2004; Roy, 1991).
PROMETHEE II, applies a linear function having as criteria the websites’ characteristics that
reflect “navigation”, “web design”, “interactivity”, “accessibility”, “e-services” and “information
usefulness” (M. De Marsico and S. Levialdi, 2004).
The results from multicriteria analysis using PROMETHEE II are presented in Table 2.
AA Website Rank
Total Net Flow:
Φ (ki) = Φ + (ki) - Φ - (ki)
1 Women Cooperative Arnissas “Voras”– P. Pella 3,048903816
Book of Proceedings
119
2 Women Agro-touristic Cooperative Mesotopos Lesvos 2,380252853
3 Women Agricultural Cooperative Agras Lesvos 1,387823219
4 Women Agricultural Cooperative Leonidion 0,884727459
5 Women Cooperative Kokkinogia Dramas 0,801005854
6 Women Agro-touristic Cooperative Zagoras – P. Magnesia 0,783774446
7 Women Cooperative Gerakari – P. Larissa 0,766838257
8 Women Agricultural Cooperative Glossa Skopelos 0,730568981
9 Women Agro-touristic Cooperative Portaria, Pelion 0,60900681
10 Women Cooperative Kato Asiton “Traditional Asitians dishes”– P. Heraklion 0,451379668
11 Women Cooperative Rachon Ikaria 0,326048469
12 Women Agro-touristic Cooperative Pteleus “Ftelia”- P. Magnesia 0,171278392
13 Women Cooperative Ioannina 0,155310682
14 Women Agricultural Cooperative Aigionion “Armonia” – P. Pieria 0,080639115
15 Women Agro-touristic Cooperative Parakila, Lesvos -0,063181556
16 Women Cooperative “Erganos” – P. Heraklion -0,161566716
17 Women Agricultural Cooperative “Agios Antonios” – P. Thessaloniki -0,322747685
18 Women Agro-touristic Cooperative Gardikion “Anemona” -0,401456658
19 Women Agro-touristic Cooperative Sesklon “Ftasma” -0,419811111
20 Women Cooperative “Archanon Geuseis” – P. Heraklion -0,479698322
21 Women Cooperative Kalon Agron Dramas “Kaloagritissa” -0,487282322
22 Women Agricultural Cooperative Argolida “Kianon Erga” – P. Argolida -0,498689517
23 Women Agro-touristic Cooperative Trigonon – P. Evros -0,677711752
24 Women Agricultural Cooperative Velvendus – P. Kozani -1,001721831
25 Women Agricultural Cooperative Zakron “Mehlion” – P. Lasithi -1,109463217
26 Women Cooperative “Poroion” – P. Serres -1,17893897
27 Women Cooperative “Melissanthi” – P. Heraklion -1,273688259
28 Women Agricultural Cooperative “Archontissa tou Aigaiou” – P. Hydrousa, Andros -1,288287049
29 Women Agricultural Cooperative Anatolikon, Valmada - P. Thessaloniki -1,436200488
30 Women Agricultural Cooperative Gimnotopos – P. Preveza -1,77711257
Table 2: Multicriteria analysis PROMETHEE II for the ranking of the websites in the sample,
based on the total net flow of their internet sufficiency.
Book of Proceedings
120
4. Conclusions
The websites lag significantly as to their characteristics of Navigation and they present a high
accomplishment rate only for criterion “fast website loading” (86,5%), while the average
accomplishment rate in this category is estimated to 41,34±28,85%. Satisfactory Accomplishment
is found for the criteria of the Design, only according to “neutral website background“(96,7%) and
the “color combination” while all the other criteria are accomplished in low rate. A lag is observed
in Interactivity characteristics except of the criterion “Telephone and/or E-mail” (86,7%), with
another accomplishment rate estimated to 38,02±31,77%. Accessibility criteria achieve a low
accomplishment rate (52,00±45,49%), e.g.: 70% of the Websites do not support “other languages”
and “connection through social media “ while they do not have “mobile view ή mobile
application”. However, all the Websites support the criteria “various navigation platforms” and
“full access no log-in required”. According to E-services criteria the results are totally
disappointing (8,02±5,07%). Finally according to Information Usefulness criteria a satisfactory
rate is accomplished for the criteria “co-operative history info (73,3%)”.
In terms of total net flow of internet sufficiency of the web sites, with a low positive total net flow
are found 14 women co-operatives. Optimum Website is of women co-operative in Arnissa, named
Voras in prefecture of Pella (total net flow= 3,0489), and then following 2 Websites in Lesvos.
The rest 16 Websites of the sample present a negative sign in relation to the total net flow of their
internet sufficiency.
Findings of the research show that the internet presence of women co-operatives is nor sufficient
neither functional, with the majority being characterized of poor content and design. The majority
in the sample have only created a basic website with certain characteristics. “e-shop with shopping
cart”, “audiovisual content- music/sound/video/slideshow”, “Social media connection”, “multiple
language support”, “online payment security”, “mobile view/application e.g. Android” are some
of the basic characteristics found in a contemporary women cooperative website, though they were
found in limited websites. Only a few cases of websites follow the website’s structure and design
rules causing a positive interest to internet users/potential clients. Based on this, we have the belief
that conclusions can become useful consulting tools for a functional structure of similar websites,
in the context of effective promotion of women entrepreneurship.
Results of research, lead us to make the following suggestions:
Improvement of the existing websites, functional enrichment with attractive content
according to the co-operative character.
Professional design of the website.
Independent website promotion in the internet.
Links in relative portals such as Ministry of Agricultural Development, Chambers,
Prefecture, etc.
Consulting through government stakeholders and motivation to co-operatives without
website to proceed in uploading their functional website.
Initiatives for informing the co-operatives on the benefits of their internet promotion and
probably organization of short term seminars on web design, internet marketing and
entrepreneurship.
Book of Proceedings
121
References
Andreopoulou, Z., Koliouska, C., Lemonakis, C., & Zopounidis, C. National Forest Parks
development through Internet technologies for economic perspectives. Operational Research, 1-
27
Chatzinikolaou, P., Bournaris, T., & Manos, B. (2013). Multicriteria analysis for grouping and
ranking European Union rural areas based on social sustainability indicators. International
Journal of Sustainable Development, 16(3), 335-351
Doumpos M., and Zopounidis K. (2004). Multicriteria Analysis of Decisions.
Methodological approaches and applications, ISBN: 9789608105706. New Technologies
Publications, 2004
Gidarakou I., Xenou A. and Theofilidou K. (2000). Farm Women's New Vocational Activities:
Prospects and Problems of Women's Cooperatives and Small On-Farm Businesses in Greece.
Journal of Rural Cooperation, 28(1): 19-37
Kargioti E., Vlachopoulou M. and Manthou V. (2006). Evaluating customers’ online satisfaction:
The case of an agricultural website. Proceedings of the International Conference on Information
Systems in Sustainable Agriculture, Agroenvironment and Food Technology (HAICTA2006).
ISBN 960-8029-43-0, Ed. Nikolas Dalezios, University of Thessaly Publications, 2006
M. De Marsico and S. Levialdi (2004). Evaluating Web Sites: exploiting user’s expectations, Int.
J. Human-Computer Studies, 60 (2004), 381-416.
Patsioura, F., Vlachopoulou, M., & Manthou, V. (2004). Evaluation of an agricultural web site.
Proceedings of International Conference on Information Systems & Innovative Technologies in
Agriculture, Food and Environment, (pp. 28-37), Thessaloniki, March 2004.
Roy B. (1991). The outranking approach and the foundations of ELECTRE methods. Theory and
Decision, Vol. 31, pp 49-73.
Tsekouropoulos, G., Andreopoulou, Z., Seretakis, A., Koutroumanidis, T., & Manos, B. (2012).
Optimising e–marketing criteria for customer communication in food and drink sector in
Greece. International Journal of Business Information Systems, 9(1), 1-25
Book of Proceedings
122
Adaptation of ITA for project portfolio selection within a group of decision
makers
Olena Pechak *,
George Mavrotas,
Danae Diakoulaki,
Laboratory of Industrial and Energy Economics, School of Chemical Engineering,
National Technical University of Athens, Iroon Polytechniou, 9, 15780 Athens
John Psarras
Energy Policy Unit, School of Electrical and Computer Engineering,
National Technical University of Athens, Iroon Polytechniou, 9, 15780 Athens
Abstract
Project portfolio selection is a problem of selecting a subset of projects from a wider set,
optimizing one or more criteria and satisfying specific constraints. Unlike in financial problems,
these projects are integer variables which are not divisible. Multiple Criteria Decision Analysis
and mathematical programming are most common tools to model such problems. When selection
process takes place within a group, preferences of multiple decision makers (DMs) are not unique
and negotiations should take place to find a needed balance between different points of view. In
present work we use a version of the Iterative Trichotomic Approach (ITA) adjusted to group
decision making with the focus on convergence. Group-ITA provides a possibility to draw
conclusions about the consensus over each individual project as well as on the final portfolio. The
basic idea is a classification of projects into three sets: the green projects (selected by all decision
makers in the “consensus” portfolio), the red projects (rejected by all decision makers from the
“consensus” portfolio) and the grey projects which are selected by some (but not all) decision
makers. Then a mathematical model is developed, where preferences of decision makers are
incorporated and a process of step-by-step convergence of these preferences takes place. As the
iterative process moves from one round to the next one, green and red sets are enriched while the
grey set shrinks. The emptiness of the grey set means end of calculations. Final outcome is the
consensus portfolio of projects, as well as the degree of consensus on each project and the
consensus index for the whole portfolio according to the convergence path. The Consensus Index
expresses the easiness to arrive at a final conclusion within a group. The more green projects we
have from early rounds the greater is the degree of concordance among DMs. On the contrary, if
the majority of green projects is identified on last rounds, it means the need to elaborate in the
convergence process in order to agree at selected projects. In other words, the consensus is hardly
attained. Besides the Consensus Index, we can extract the degree of consensus for each project
according to the round that it enters or exits the final portfolio. The method is illustrated with an
example based on real data for renewable energy projects.
Keywords: Project Portfolio Selection, Multiple Criteria, Integer Programming, Group Decision
Making, Consensus
*corresponding author, e-mail: [email protected]
Book of Proceedings
123
1. Introduction
In current research, project portfolio selection is defined as problem of selecting one or a subset
from a wider set of proposals, in other words a subset of projects is considered as a “portfolio of
projects”. In past, the usual approach was to rank projects using one or more criteria and choose
the top ranked ones that cumulatively satisfy a budget limitation. However, the existence of
various limitations to be satisfied by the final selection destroys the independence of projects,
which is one of main assumptions in Multiple Criteria Decision Analysis (MCDA) ranking (see
e.g. Belton and Stewart (2002)). Specifically, the top ranked projects may only by chance satisfy
imposed constraints. On the other hand, Integer Programming (IP) is an appropriate tool for such
combinatorial problems where 0-1 (binary) variables express incorporation (Xi=1) or exclusion
(Xi=0) of ith project in/from the portfolio.
Involvement of several decision makers leads to even more complicated selection process.
In an attempt to address the selection problem within a group, we modify and test the Iterative
Trichotomic Approach (ITA) under the title Group-ITA. The ITA method has been originally
developed for project portfolio selection in order to deal with uncertainty either in problem’s
parameters (Mavrotas and Pechak, 2013a) or in preference parameters (Mavrotas and Pechak,
2013b).
2. The ITA Method
The core idea of ITA method is a separation of initial set into three parts (trichotomy) according
to the projects’ membership in the final portfolio, given the current level of information. The
selection is performed iteratively in computation rounds (R), until the process converges to a
final set. A predetermined number of these rounds may be initially set and every round feeds its
subsequent until the final selection is built. According to ITA the initial set of candidate projects
is divided into three subsets (classes): the “green” projects that are present in final portfolio
under all circumstances, the “red” projects that are absent from final portfolio under all
circumstances, and the “grey” projects that are present in some final portfolios. From round to
round the grey set is reduced as a result of convergence or uncertainty reduction. Such process
flow helps the Decision Maker (DM) to identify and focus on projects that are really at stake.
The “sure” projects (green and red sets) are determined and the attention can be shifted towards
“ambiguous” projects (the grey set). Similar approach of splitting initial set of projects on 3
subsets has been proposed by Liesiö et al. (2008), however, the membership principles were
different. Our method provides quantitative and qualitative information that cannot be acquired
using e.g. just expected values of distributions. In the latter case, the DM is provided only with
final optimal portfolio where the information about certainty degree of selected projects is
missing.
For the group project portfolio selection we adopt the combination of Multiple Criteria
Decision Analysis and Integer programming (MCDA – IP). At the beginning we use MCDA to
assign scores to projects and then feed these evaluations in objective function as coefficients in
the IP model that incorporates constraints of the project selection problem. In presence of
multiple DMs we assume that each DM expresses his preferences by assigning his own weights
Book of Proceedings
124
to the criteria of project evaluation. Therefore, each DM calculates his own multi-criteria scores
for projects. In general this means that he usually has an objective function that differs from the
others’. As a consequence, obtained optimal portfolios are usually different. In the so called
Group-ITA the membership of each project in the green, red or grey sets is determined according
to the concordance of decision makers. Namely, the green set includes projects that are present in
the final portfolio according to all DMs, the red set those ones that are absent from the final
portfolio according to all participants, and the grey set includes projects that are present in the
final portfolios according to some group members. Assume that there are N projects, P DMs and
K criteria of evaluation. Therefore the weight of importance that p-th DM assigns to k-th
criterion is wpk with p=1..P and k=1..K. For each group member we calculate multi-criteria
scores mspi for every project i=1..N. The objective function of the IP problem for the p-th DM is
then:
max1
N
pi i
i
ms X
where Xi is the binary variable that indicates if the i-th project is selected (Xi=1) or rejected
(Xi=0). Solving the formulated P integer programming problems we obtain at most P different
optimal portfolios (some of them may be identical). Subsequently, the members of the green, red
and grey sets are identified. Items of green set are those projects present in all P optimal
portfolios (green projects). Accordingly, within red set are projects that are absent from all P
optimal portfolios (red projects) and the grey projects (i.e. members of grey set) are those that are
present in some of P optimal portfolios.
3. Case study
Group-ITA is applied in a group decision making problem dealing with energy projects. There
are 133 energy projects from three RES technologies (wind - W, small hydro - SH, photovoltaic -
PV) distributed across 13 regions of Greece. The 5 criteria for evaluation are Regional
development, Employment, Economic Performance (expressed with IRR), CO2 emission
reduction and Land use. All but the 5th objective are to be maximized. The data for this problem
are available in Makrivelios (2011). There are also specific policy constraints for the project
portfolio selection problem that must be respected, namely:
Available budget is 150 M€ (the total cost of the 133 projects is 659 M€),
Cost of projects in Central Greece should be less than 30% of the total cost,
Cost of projects in Peloponnese should be less than 15% of the total cost,
Cost of projects in East & West Macedonia, Northern & Southern Aegean, Epirus should
be greater than 10% of the total cost ,
Number of projects by technology should be between 20% and 60% of selected projects,
Total capacity of selected projects should be greater than 300 MW.
Book of Proceedings
125
projects
MW
(>=300)
Cost
(<=150
M€)
STE
(<=30%)
PEL
(<=15%)
Other
(>=10%)
W
(20%-
60%)
SH
(20%-
60%)
PV
(20%-
60%)
round 0 73 185.5 96.5 25.5% 3.2% 22.5% 15.1% 35.6% 49.3%
round 4 74 202.6 102.9 23.9% 9.2% 21.1% 16.2% 35.1% 48.6%
round 6 77 222.6 114.7 25.6% 8.2% 20.8% 16.9% 33.8% 49.4%
round 8 78 235.2 119.5 28.6% 7.9% 20.0% 17.9% 33.3% 48.7%
round
10 83 301.3 149.8 29.4% 10.3% 25.6% 20.5% 32.5% 47.0%
Table 1. Characteristics of consensus portfolio (i.e. green projects only).
The results and particular characteristics of the portfolio created by green projects in each
round (consensus portfolio) are shown in Table 1. By studying it decision makers may decide to
select a consensus portfolio prematurely, i.e. before arriving to Round 10, if they accept the
respective constraints violations, which are denoted with bold fonts.
Further, we develop a measure of consensus for the final portfolio according to the degree
of concordance among group members, which actually expresses the easiness of convergence to
final consensus. In order to calculate a consensus index (CI) we draw the so called consensus
chart where the percentages of green projects that have already been found in rth round are
plotted as a function of respective decision round. The resulting curve is called consensus curve.
In Figure 1 observe that until round 3 there are no new projects added in the green set. This may
happen especially when the maximum number of rounds (R) is relatively high. In addition, the
DM is aware of projects’ prioritization given that he knows in which round a project enters the
green set.
0 1 1 2 1
10
1
10
1
( ... ) /2 2 2
[ ] /2 2
1[ ] /
2 2
R R
RR
r
r
R
r
r
a a a a a aCI R
a aCI a R
aCI a R
CI takes values between 0 and 1
and is calculated using the trapezoid rule for piecewise linear functions according to the
following equations where aR is the percentage of green projects that have already been found in
rth round: 0.88 1
[ 3 0.88 2 0.892 2 0.928 2 0.94 ] /10 91%2 2
CI
80%
82%
84%
86%
88%
90%
92%
94%
96%
98%
100%
0 1 2 3 4 5 6 7 8 9 10
% o
f gre
en p
roje
cts
in t
he
fin
al p
ort
folio
Rounds
α1 α2 α3
α4
α8α7α6
α5
α9
α10
α0
Fig. 1. Consensus chart for the application.
Book of Proceedings
126
4. Conclusions
Project portfolio selection is a complex problem that usually involves numerous criteria and
constraints (budget, policy, allocation etc) that should be satisfied. Moreover, multiple decision
makers from different positions, with different backgrounds and usually with conflicting views
participate in the selection process. In present paper we propose a systematic procedure towards
convergence of different points of view. For this method we assume that the preferences of group
members are expressed with a set of weights used in the MCDA for overall project evaluation.
The cornerstone of the iterative process is the systematic way for weights convergence that
guarantees the convergence of the whole process to a final portfolio. The outcome of Group-ITA
is not merely the final portfolio, but also the “course” towards it that may provide fruitful
information about the project selection problem and may be used to reconsider some of the initial
assumptions. After all calculation we not only converge to the final portfolio, but also measure
the degree of consensus for each project that is selected or rejected. Moreover, it provides a
measure of consensus for the final portfolio as a whole.
References
Belton, V., Stewart, T.J. “Multiple Criteria Decision Analysis: An Integrated Approach”. Kluwer
Academic Publishers, Boston 2002.
Liesiö J., Mild P., Salo A. “ Robust portfolio modeling with incomplete cost information and
project interdependencies”. European Journal Operations Research, Vol. 190, 2008, pp. 679–
695.
Makrivelios E. “Multi-criteria evaluation of projects based on renewable energy sources in
Greece”. Master thesis in “Management of Industrial Systems”, joint program of National
technical University of Athens and University of Piraeus, Greece, 2011.
Mavrotas G., Pechak O. (a) “The trichotomic approach for dealing with uncertainty in project
portfolio selection: Combining MCDA, mathematical programming and Monte Carlo
simulation”. International Journal of Multicriteria Decision Making, Vol. 3 (1), 2013, pp. 79-97.
Mavrotas G., Pechak O. (b) “Combining Mathematical Programming and Monte Carlo
simulation to deal with uncertainty in energy project portfolio selection”. In: F. Cavallaro (Ed),
Assessment and Simulation Tools for Sustainable Energy Systems, Springer-Verlag, London
2013, pp. 333-356.
Book of Proceedings
127
F.W. Lanchester’s combat model application in a supply chain in a duopoly
Miltiadis Chalikias
Applied Economic Statistics and Operations Research Laboratory, Department of Business
Administration, School of Business and Economics, Technological Education Institute of
Piraeus, 250 Thivon & P. Ralli, 12244, Egaleo, Greece, email: [email protected]
Michalis Skordoulis
Management Information Systems and New Technologies Laboratory, Department of Business
Administration, School of Business and Economics, Technological Education Institute of
Piraeus, 250 Thivon & P. Ralli, 12244, Egaleo, Greece email: [email protected]
Abstract
The purpose of this study is to investigate the possibility of applying some of the most widely
known mathematical theories of war in firms. In this research, Frederick William Lanchester’s
combat models were examined that seemed to be particularly useful in the U.S. Army at the Pacific
campaign against the Japanese fleet during World War II. These mathematical models were based
on differential equations and its main purpose was to predict the outcome of battles.
Keywords: Frederick William Lanchester, mathematical theories of war, differential equations,
supply chain in duopoly.
1. Introduction
Frederick William Lanchester was born in 1868 in London and studied engineering (Ricardo,
1948). In 1916, he invented the operations strategy for the Royal Air Force of England, formulating
model based on differential equations, the two models estimate the forces that are required for
winning in a military battle (Bracken, 1995).
In the first Lanchester combat model, it is considered that two forces with the same martial
ability, R(t) and G(t), initiate a military conflict between them at the time (t). R(t) neutralizes g
number of soldiers, and G(t) neutralizes r number of soldiers respectively (MacKay,2006). The
numbers g and r, are called efficiency ratios of forces R and G respectively (Daras, 2001). We
have an initial situation where applies the following system:
dGgG
dt
dRrR
dt
(1)
Apart from the first case, there exists the more complex mathematical case, Lanchester’s second
model, where two forces participate at the military conflict, one of which has greater military
capacity than the other, creating the so-called asymmetric warfare (Lanchester, 1956).
Book of Proceedings
128
As far as business administration is concerned, Lanchester’s combat models have been applied
in several cases. A new approach to Lanchester’s combat models, was applied for the first time in
Japan by Taoka and Yano in marketing strategy (Oudrhiri, 2005).
2. Construction of the mathematical model
We consider two competing firms A and B which coexist in a common market and sell the product
P. This market is characterized as an oligopoly, as we deem that there aren't other similar firms,
while the input of such firms is very difficult. Because in this oligopoly we consider the existence
of only two firms, we are eventually also led to the special form of duopoly just as in the case of
the implementation of Richardson’s arms race model (Chalikias & Skordoulis, 2014). It is, at first
phase, assumed that the technology that is used by the two firms is the same. The firms know each
other's moves.
Let x(t) to be the number of available product units for sale of firm A and y (t), the number of
available product units for sale of firm B at time t. During the operation of firms A and B that are
competitive with each other, the rate of change of the quantities x(t) and y(t) equals the rate of
growth of refueling at the distribution points, minus the rate of their reduction.
The rate at which the available for-sale product units are increasing and decreasing is denoted by
f(t) for A and g(t) for B respectively. The rate of available for sale product units for firm A equals
ay(t) and for firm B with bx(t), where a and b are appropriate positive constants. As in the case of
the two warring conventional forces that is analyzed by Daras (2001), the mathematical model that
is based on Lanchester's combat models and describes the above situation is the following:
dxay f(t)
dt
dybx g(t)
dt
(2)
Any solution of the system of differential equations (2) for 0 0x and
0 0y will be given by
the formulas:
0
02 2
tx x x xa e e e ex(t) y ( abt) ab(t s) f(s)ds
b
0
02 2
tx x x xb e e e ey(t) x ( abt) ab(t s) g(s)ds
a
(3)
(4)
The mathematical model may be applied on several examples of supply chains in duopoly. It is
proposed that this model is applied on the market of cola type drinks in Greece. The competition
between Coca-Cola and Pepsi has motivated the interest of many researchers who have studied
and implemented models like this. Chintagunta & Vilcassim (1992), used Lanchester’s combat
models in order to examine the effects of advertising expenditure on consumer demand for Coca-
Cola and Pepsi in the level of competition between these two firms. A similar model using
statistical data on advertising expenditure of Coca-Cola and Pepsi in order to analyze advertising
strategies that are used by them was also applied by Erickson (1992). In the same context, Wang
& Wu (2001) resulted in the fact that consumers respond to the advertisements of Coca-Cola and
Pepsi in the same way. Finally, using Lanchester’s combat models, Fruchter & Calish (1997)
Book of Proceedings
129
described the competition between Coca-Cola and Pepsi and solved the problem of determining
the optimum advertising strategy for maximum profits. In their entirety all the researches have not
taken into account the possible influence of other firms that sell cola type products.
As far as the previous literature analysis is concerned, it would be very interesting that the
constructed model will be applied in the duopoly of Coca-Cola and Pepsi in the Greek market.
3. Conclusion
From the above research, it was concluded that the mathematical model that was constructed
based on Lanchester’s combat models may be applied in the case of duopoly that was examined.
In a next step, it is proposed the construction of similar models that take into account several
factors such as the price and quality of products that may affect consumer preferences and
ultimately the firms production.
References
Bracken, J. (1995). Lanchester models of the Ardennes campaign. Naval Research Logistics.
42(4): 559-577.
Chalikias, M. & Skordoulis, M. (2014). Implementation of Richardson’s arms race model in
advertising expenditure of two competitive firms. Applied Mathematical Sciences. 8(81): 4013-
4023.
Chintagunta, P. & Vilcassim, N. (1992). An empirical investigation of advertising strategies in a
dynamic duopoly. Management Science. 38(9): 1230-1244.
Erickson, G. (1992). Empirical analysis of closed-loop duopoly advertising strategies.
Management Science. 38(12): 1732-1749.
Fehlmann, T. (2008). New Lanchester theory for requirements prioritization. In: Proceedings of
the Second International Workshop on Software Product Management. Barcelona, September
2008. Bacelona: I.E.E.E, pp. 35-40.
Fruchter, G. & Kalish, S. (1997). Closed-loop advertising strategies in a duopoly. Management
Science. 43(1): 54-63.
Lanchester, F.W. (1956). Mathematics in warfare. The World of Mathematics. 4: 2138-2157.
McKay, N. (2006). Lanchester combat models. Math Today. 42: 170-178.
Oudrhiri, R. (2005). Six Sigma and DFSS for IT and Software Engineering. The Quarterly Journal
of the TickIT Software Quality Certification Scheme. 4: 7-9.
Ricardo, H. (1948). Frederick William Lanchester. Obituary Notices of Fellows of the Royal
Society. 5(16): 756-766.
Book of Proceedings
130
Wang, Q. & Wu, Z. (2001). A duopolistic model of dynamic competitive advertising. European
Journal of Operational Research. 128(1): 213-226.
Book of Proceedings
131
An optimization modeling approach for the establishment of a bike-sharing
network using Monte Carlo Simulation and stochastic demand: a case-study
of the city of Athens
Zygouri E.
Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos
Athinon, Pedion Areos, 38834 Volos, Greece, Email: [email protected]
Fragkogios A.
Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos
Athinon, Pedion Areos, 38834 Volos, Greece, Email: [email protected]
Saharidis G.K.D.
Department of Mechanical Engineering, School of Engineering, University of Thessaly, Leoforos
Athinon, Pedion Areos, 38834 Volos, Greece and Kathikas Institute of Research & Technology,
Paphos, Cyprus Email: [email protected]
Mavrotas G.
Department of Process Analysis and Plant Design, School of Chemical Engineering, National
Technical University of Athens, Zographos Campus, 15772 Athens, Greece, Email:
Abstract
This study introduces a novel mathematical formulation that addresses the strategic design of a
bicycle sharing network with stochastic estimated demand. The developed pure integer linear
program takes into consideration the available budget of a city for such a network and optimizes
the location of bike stations, the number of their parking slots and the the bicycle fleet needed in
order to meet as much demand as possible and to offer the best services to the users. The
methodology used for the simulation of demand is the Monte Carlo simulation, which is combined
with the Iterative Trichotomic Approach (ITA). The proposed method is implemented on the very
center of the city of Athens, Greece.
Keywords: Bike, Sharing System, Monte Carlo, Integer
Book of Proceedings
132
1. Introduction
Bike-sharing networks have received increasing attention during the last decades and especially in
the 21st century as a no-emission option in order to improve the first/last mile connection to other
modes of transportation, thus facilitating the mobility in a densely populated city. The bike-sharing
network consists of docking stations, bicycles and information technology (IT) interfaces that have
been recently introduced to improve the quality offered to the users.
This expanding trend of bike-sharing networks necessitates their better planning and design in
order that they are successful. The goal of this paper is to propose a novel mathematical
formulation to design such networks incorporating stochastic estimated demand, the fixed costs of
infrastructure, the proximity and density of stations, as well as their size. Given a set of candidate
locations of stations and with a predefined available construction budget the model decides the
number and the location of the stations, how large they will be and how many bikes should they
have in order to meet the assumed demand.
2. Model Formulation
Given a set of candidate locations of bike stations and the value of demand for bikes at these
locations it is necessary to know where to place the bike stations and how many parking slots and
bikes should each one have. The available budget of a city for the construction of the whole bike-
sharing system is predefined and so are the costs of a single bike, a single parking slot and a single
station. So it is a matter of optimization for the model to decide how many stations, bikes and
parking slots it will include in its solution. The walking time between the locations is another
parameter of the problem used to ensure the proximity of the constructed stations as far as this is
possible.
As regards demand in each location, it is split into “Demand for Pick-Ups” and “Demand for Drop-
Offs”. The first one depicts how many users would like to take a bike from a station and the second
one shows how many riders would like to leave a bike at a station.
In Figure 1 the thorough consideration of the problem is explained. N locations i are predefined
together with their “Demand for Pick-Ups” and “Demand for Drop-Offs”. The walking time
between these N locations is, also, known. It is a matter of optimization how many bike stations
will be established and where, so that every location has a nearby station. The locations k, where
stations are established, is a subset of the locations i.
Figure 10 Network structure of bike-sharing system.
Book of Proceedings
133
If budget is not enough to construct stations at all N locations, at some locations i there will be no
station. These locations should have a nearby station k no more than a specific walking time away
and only a percentage of their demand is considered to be passed to this station k. The rest of their
demand is not served supposing that this part of citizens will not take a bike due to the distance of
the station k from their location i. In this way, it is assumed that location i is served by station k.
On the one hand, this transfer of the demand is inevitable as the restricted budget does not allow
stations to be built at all locations. On the other hand, it is not desirable because it means that the
users of the network will have to walk from location i, where they would rather a station to be
present, to the established station k and vice versa. This would result into poor service quality
offered to the users of the bike-sharing network, as some potential customers will not eventually
use the network.
This consideration is accomplished through the following objective function. The objective
function of the model is a minimization of three terms. The first term expresses the amount of
demand that is transferred from a location i to its allocated station k, which are a specific walking
time away from one another. Thus, the model will propose a dense distribution of stations,
establishing no station at locations with low demand ensuring that they are as close to a station as
possible. This term is multiplied by the penalty unit cost to differ its importance from the other
two terms. The second and the third term of the objective function are introduced in order to
minimize the unmet demand. There is a difference between the parameters that express the
“Demand for Pick-Ups” from location i during time interval t and the “Demand for Drop-Offs” at
location i during time interval t and the variables that express the number of bicycles that are
available at station k at the beginning of time interval t and the number of bicycles that could leave
station k during time interval t. The former express the users who would like to pick up and drop
off a bike from and to a candidate station location respectively. However, the station k may not
have the required bikes or free parking slots to meet these two types of demand respectively. So
the number of bikes that eventually leave or arrive at a station k is expressed by the two mentioned
variables. Both the parameters and the variables refer to each time interval t. These two terms are
multiplied by the same penalty unit cost meaning that no different importance is given to either of
them.
In the mathematical model there is a constraint, which warrants that the total available budget is
not exceeded. Other constraints ensure that the bicycle parking slots at each constructed station are
between the permissible minimum and maximum value and that each station cannot have more
bikes than the number of its parking slots.
Some other constraints guarantee that a location i cannot be served by location k, if a station is not
built in location k, also, that if a station is constructed at location k this location will be served by
its own station. Furthermore, that each location i may be served by exactly one bike station k and
that a constructed station k can serve only locations which are located within a maximum walking
time from it.
Finally, among others there are some constraints, which guarantee that the bicycles that can leave
the station can be no more than the available ones and the bikes that can come to a station can be
no more than the free parking slots.
3. Monte Sarlo Simulation and Iterative Trichotomic Approach
In order to find the optimal solution the Iterative Trichotomic Approach is applied, which consists
of decision rounds. Each of these rounds includes a series of 1000 Monte Carlo simulations - IP
Book of Proceedings
134
optimizations, from which 1000 optimal solutions arise. In each of the simulations values for every
demand of every candidate location are randomly generated within the probability distributions
that are estimated to describe them. In each decision round the candidate locations are categorised
into green, red and grey subsets, depending on how many times they are included in the optimal
solution for the establishment of a bicycle sharing station.
Specifically, if a candidate location is selected as a station in more than 990 iterations of a decision
round, it is allocated to the green subset. Respectively, if a candidate location is selected as a station
in less than 10 iterations of a decision round, it is allocated to the red subset. If a candidate location
is selected as a station in more than 10, but less than 990, iterations of a decision round, it is
allocated to the grey subset. At the end of each decision round, the green and red locations are
fixed with constraints in order to be selected and not selected respectively for the establishment of
bike stations and the variance of the demand of the grey locations is reduced, assuming that their
demand becomes less stochastic (more information is gained for these stations). The Iterative
Trichotomic Approach is completed when all candidate locations are categorised into the green
and red subsets and no grey locations exist, as depicted at Figure 2.
Figure 2 The Iterative Trichotomic Approach
Of particular importance is the method used to estimate the stochastic demand characteristics of
each candidate location. Originally, the hourly usage of an already established Bike Sharing
Network, “Velib’” in Paris, is used as a basis. The candidate locations are categorized into clusters,
whose demand is multiplied by standard rates adjusted on the center of Athens. However, each
location is evaluated for 11 parameters that affect the expected demand for shared bike rental.
These 11 parameters are Population Density, Universities (facilities, student dormitories, etc.),
Jobs Density, Retail Jobs Density, Tourist Attractions, Parks and Recreation areas, Metro, train
stations, Buses, trolleys stations, Bicycle lanes, Bicycle friendly roads and Topography. Thus, the
upper and lower limits of the probability distributions of every demand of each location are
obtained. These distributions are used as the basis of the stochastic Monte Carlo simulations.
Book of Proceedings
135
4. Athens Case-Study
4.1 Data Settings
The authors chose 50 candidate locations in the 1st Municipality District of Athens, where bike-
sharing stations could be constructed. These candidate locations were categorized into four clusters
(“Housing”, “Employment”, “Subway” and “Spare Time”) depending on their location. The
stochastic demand is assumed to follow the uniform distribution. The upper and lower limits of
the probability distributions of every demand of each candidate location are obtained through the
evaluation of the location on the 11 preciously mentioned parameters.
The walking time between these locations was calculated using Google Earth. As regards the costs
of the network, two already implemented networks were taken into account, the first one in Greece
(Karditsa) and the second one in Cyprus (Nicosia). Examining the budget and the dimensions of
each city and its network the following data were assumed for the case of Athens. The cost of
establishing a station is € 12,000. The cost of each slot in a station is € 900. The cost of a bike is €
500 and the total available budget is € 1,000,000. Furthermore, it is assumed that a location with
no station cannot stand off a location with a station within more than 7 minutes of walking time.
The minimum and the maximum parking slots that a station can have are as many as in the Velib’
network (between 8 and 70 per station).
4.2 Results
Decision rounds of the Iterative Trichotomic Approach were executed, in each of which 1000
Monte Carlo simulations were made. In each decision round the candidate locations are
categorised into in the green, red and grey subsets, as mentioned above. The convergence is
obtained after four decision rounds.
The above decision making process determined the locations where bike stations will be
established. However, when designing a Bike Sharing Network, it is necessary to determine the
number of parking slots of each station and the number of bikes in the whole network. The
methodology followed for this calculation each station is as follows: Initially, having added
constraints for establishing stations on the green and not the red locations, another round of 1000
Monte Carlo iterations was performed, where bikes demand was stochastic for all sites (uniform
distribution with the largest range of values as in the 1st decision round). The 1000 iterations with
stochastic demand led to a distribution of parking slots for each station. Afterwards, the less
probable values of slots at each station were excluded and the reduced range that remained are
included in the mathematical model as constraints. Finally, the mathematical model is solved for
the last time and the demand of every location equals the centre value of its uniform distribution.
Thus, one optimal solution is gained with the parking slots and bikes of every station.
Figure 3 depicts the proposed established bike stations. The shape of each dot corresponds to the
station’s cluster, while its size represents the number of parking slots each station should have.
The total number of docking stations is 38 and the number of parking slots is 470 making a mean
value of 470/38=12.4 slots per station. Looking at the parking slots of each station, one can notice
Book of Proceedings
136
that the larger stations belong to the cluster “Subway”, which is typical of the increased demand
in the metro stations. The total number of bikes in the network is 242.
Of particular importance is the knowledge of the decision round, in which each location is included
in the green or red subset. The sooner a location enters the green or red subset, i.e. the sooner her
condition ("go" or "no go") is concluded the larger the degree of certainty of that decision. This
information is very useful for decision-makers given the uncertainty of demand of each location.
Figure 3: The established stations of the solution of the case of Athens categorized in clusters
and with their size
5. Conclusions
Solving the proposed mathematical model many times with different values of stochastic demand
each time, can lead to the optimal design of a Bike Sharing Network so as to meet as much demand
as possible during its usage afterwards. The knowledge gained from the already implemented
networks can and should be used for the design of future ones. This model reclaims the usage data
from the Velib’ network of Paris to predict demand in Athens and designs a suitable bike-sharing
network to meet that demand.
However, some parameters could be altered to notice how the solution changes. Such parameters
could be the available budget or even the demand profiles to approximate the seasonal differences
(winter-summer) or the week differences (weekdays-weekend). The larger application of the
model is, finally, another work to be done concerning, for example, the whole Municipality of
Athens.
Housing Employment Subway Spare Time
Book of Proceedings
137
References
DeMaio, P., 2009, “Bike-sharing: History, Impacts, Models of Provision, and Future”, Journal of
Public Transportation, Vol. 12, No 4, pp. 41-56
Mavrotas, G. and Pechak, O. , 2013, “The trichotomic approach for dealing with uncertainty in
project portfolio selection: combining MCDA, mathematical programming and Monte Carlo
simulation”, Int. J. Multicriteria Decision Making, Vol. 3, No. 1, pp.79–96.
Lin, J-R., T-H. Yang., 2011, “Strategic design of public bicycle sharing systems with service level
constraints”, Transportation Research Part E, Vol. 47, pp. 284-294
Martinez, M. L., L. Caetano, T. Eiro, F. Cruz., 2012, “An optimization algorithm to establish the
location of stations of a mixed fleet biking system: an application to the city of Lisbon”, Procedia-
Social and Behavioral Sciences, Vol. 54, pp. 513-524
Krykewycz G. R., Puchalsky C. M., Rocks J., Bonnette B., & Jaskiewicz F., 2010, “Defining a
Primary Market and Estimating Demand for Major Bicycle-Sharing Program in Philadelphia,
Pennsylvania”, Transportation Research Record, 117-124
Lathia N., S. Ahmed, L. Capra., 2011, “Measuring the impact of opening the London shared
bicycle scheme to casual users”, Transportation Research Part C, Vol. 22, pp. 88-102
Etienne C., L. Oukhellou., 2012, “Model-based count series clustering for Bike-sharing system
usage mining, a case study with the Velib’ system of Paris”, Transportation Research-Part C
Emerging Technologies, Vol. 22, pp. 88
Book of Proceedings
138
Evaluating new service development effectiveness in tourism: An ordinal
regression analysis approach
Fotis Kitsios
School of Information Sciences, Department of Applied Informatics, University of Macedonia,
P.O. Box 1591, GR54006 Thesaloniki, Greece, email: [email protected]
Evangelos Grigoroudis
School of Production Engineering and Management, Technical University of Crete, University
Campus, GR73100 Chania, Greece.
Abstract
Innovation development is an important factor for the viability and profitability of service
businesses operating in modern markets. The importance of the service sector in developing
economies and the specific characteristics of services compared to tangible products require
further investigation in the New Service Development (NSD) process and effectiveness assessment
when innovations applied. The innovation development process will be significantly improved and
the contribution of innovations in company’s goals will be substantial. The purpose of this study
is to evaluate the effectiveness of the NSD process in the tourism economy and in particular the
Greek hotel sector. For this purpose, factors influencing the process of developing new services in
the hospitality sector were explored and correlated with the financial results of the hotel enterprises
through an ordinal regression analysis model. The model adopts a mathematical programming
approach in order to estimate the efficiency of this process. In the presented study the Greek
tourism industry and its importance to the national economy is discussed. The study explores in
detail the factors influencing the NSD process. The questionnaire developed for the purpose of the
survey included a large number of variables related to all the stages of the NSD process (from idea
generation till the service launch). All variables are measured in a 5point Likert type scale and data
was collected using in depth structured and questionnaire-based interviews with 77 hotel managers
for 147 new services in a representative sample of 99 hotels in Greece. Several financial ratios
covering different aspects of the business (e.g., profitability, liquidity, activity) are used in order
to evaluate the NSD process for three years after the services innovation had been launched. The
main results of the ordinal regression model include the estimated contribution of each factor to
the financial performance of the hotels studied.
Keywords: New Service Development, Efficiency Evaluation, Ordinal Regression Analysis,
Business Performance, Service Innovation Strategies
Book of Proceedings
139
1. Introduction
The rapid changes in today's business environment such as globalization and competition have a
direct impact on the conditions in which businesses operate. The evolution of management science,
the development of technology, and the fact that customers over time becomes increasingly
demanding, creating the need for companies to constantly seek new ways to differentiate what they
offer to the market. They aimed to gain a competitive advantage in order to get profitable and
sustainable. The ability to create innovative products and services is the key to sustainability and
growth. The rapid development of technology, the management systems that ensure quality of
products and services, the pretentiousness of consumers, the competition from non-traditional
sources, such as the Internet, and the emerging hybrid industries, increase the need for business
innovation.
In recent years there have been many discussions and research to develop new products and
services and to evaluate their effectiveness. In the field of products, new product development is
accepted as requirement for business development and prosperity, and the effectiveness is
measured and reflected in the company's turnover. However, new service development (NSD) is
not so widely studied, given the particular characteristics of the service industry (e.g., differences
between products and services, deficient knowledge available in NSD).
The purpose of this paper is to evaluate the effectiveness of the NSD process in the Greek
hospitality industry. For this purpose an exploration of the factors influencing the process of
developing new services in the hospitality sector is conducted and an ordinal regression analysis
linking these factors with the financial results of hotels is applied.
2. Literature Review
The development of new services and the orientation towards innovation are components of
success for modern enterprises. In this context, Dolfsma (2004) emphasized that service firms are
more profitable when they are innovative. In addition, a service business innovator may also have
high performance in non-financial assets, such as reputation, trust building, and good relations
establishment with existing and new customers (Avlonitis et al., 2001). A strong correlation
between innovation strategy and financial outcomes has been found by Zahra and Covin (1994),
who suggested that companies should avoid investing in innovations that do not fit with the
strategic goals of the business. Moreover, they found that the relation between financial
performance and different types of innovation may vary. The relation between innovation and
business performance may be studied through several variables, like patents acquired by
enterprises, innovations in processes, and particularly investment in R&D departments (Nås and
Leppãlahti, 1997). All previous research efforts emphasize that innovation is always associated
with the business profile, the philosophy adopted, as well as the sector and the size of the
organization.
The evaluation of business performance can be based in different datasets. For example, some
variables that measure the profitability of a business-oriented innovation are: efficiency, growth,
profit, liquidity, success / failure, and market share.
Book of Proceedings
140
Generally, all studies report a significant correlation between innovation and business
efficiency. However, the adoption of innovations may result to business competitiveness, only if
the company can defend itself in the market against competitors. In addition, innovations may lead
to better flexibility, which is an extremely important advantage in market conditions with strong
competition.
Numerous studies show that innovative firms may have improved financial (assets, sales,
exports, etc.) or business performance results (growth, number of employees, etc.) (see for
example Thwaites and Wynarczyk, 1996; North and Smallbone, 2000). However, all these
researches focus on the relationship between innovation and business performance, without
measuring the efficiency of adopted innovations, especially in the service sector. Furthermore,
most studies are cross-sectoral, and thus presented findings cannot be easily extrapolated to the
service sector. This justifies the necessity of conducting more sector-oriented studies in order to
have comparable and representative results (Kitsios, 2005).
3. Methodology
a. Data and Variables
As far as the Greek hotel sector is concerned, a first and thorough approach was conducted in 2005,
aiming to record and comprehend the decision-making process followed by hotel managers
(Kitsios, 2005). Collecting determinant factors of success in NSD defined in Greek and
international literature, this research formed a 126 factor questionnaire that was applied in
interviews with 99 Greek hotels of a wide geographical range. The study used a Likert-type data
collection process and applied several statistical analysis methods. The initial large set of factors
reduced in 24 determinant new factors, 6 statistically significant. These factors have been included
in a predictive model which may be used as a guide by the hotel managers.
Based on the aforementioned framework, this study analyzes the efficiency of NSD in the hotel
performance. For this reason, two sets of variables are used:
Drivers: These variables are based on the NSD process and can be considered as the causes
of financial results. As shown in Table 11, the 24 variables used in the study are categorized
into 6 main groups: 1) Enterprise’s behavior for the service innovation, (2) Idea generation
sources for the provided service, (3) Actions for developing the provided service, (4)
Organizational structure impact, (5) Enterprise’s resources allocation impact, and (6)
Market impact (see details in Kitsios, 2005; Kitsios et al., 2009)
Outcomes: These variables are based on the financial balance sheets of the examined
hotels. A total of 8 financial ratios are used in this study, covering profitability, turnover,
efficiency, as well as solvency ratios (Table 12).
The final dataset of the presented study consists of 77 hotels and a total of 153 new service
projects, both successes and failures. Data were collected by direct in depth interviews with the
hotel managers.
Book of Proceedings
141
b. Ordinal Regression Analysis
The applied model is an ordinal regression analysis approach, assuming a set of explanatory and a
set of response variables. In particular, given a set of result variables jY and a set of driver
variables iX , the main principle of this approach is that the weighted average of the “values” of
jY can be written as a weighted average of the “values” of iX , according to the formula:
Dimension Variables Dimension Variables
Strategic
Focus
1. The strategic objectives had been
clearly identified
Integrated
Market
Launch
15. An integrated promotion plan had been
implemented (e.g. brochures,
advertisements, direct sales, exhibitions,
conferences and seminars for clients)
2. Expression of objectives as
contribution to the income of the
company
After
Launch
Review
and
Assessment
16. The measurements and forecasts had
been successful for the performance of the
new service
3. The areas of strategic focus were
clearly identified
17. The advertising, promotion and
communication efforts were targeted to the
right customer segment
4. The strategic action plans were
clearly identified
Market
Potentiality
18. Previous knowledge of the potential
market size
Idea
Generation
5. There was a mechanism and a
systematic effort to capture and collect
new ideas for development
Market
Synergy
19. An analysis of how the product meets
customers’ needs was conducted as
opposed to competing products
Preliminary
Market
Assessment
6. Preliminary market assessment had
been undertaken prior to any major
investment
20. The service was aligned with the
overall image of the hotel
7. Enough time and money were spent
on preliminary market assessment
21. The potential needs of customers were
appreciated in the commercialization stage
of the new service
8. A clear and focused definition of
the target market was given during the
preliminary market assessment
22. The customers’ purchase decision
process and behavior was clearly
perceived by the hotel
Operational
Analysis
9. A realistic business analysis had
been carried out
23. There were strong support for the new
product after its launch
10. A comprehensive analysis of the
competition had been carried out
24. Potential customers had showed a great
need for this class of product
11. Forecasts of expenses and sales
had been conducted
12. Discount cash flow analysis
Book of Proceedings
142
13. Breakeven and return on
investment analysis
14. Informal analysis (guesses and
estimates)
Table 11 NSD variables (drivers of efficiency)
Dimension Variable
Profitability 1. Gross profit margin (Gross profit/Sales)
2. Net profit margin (Net income/Sales)
Turnover 3. Asset turnover (Sales/Total Assets)
4. Equity turnover (Sales/Equity)
5. Liability turnover (Sales/Liabilities)
Efficiency 6. ROE- Return On Equity (Net income/Equity)
7. ROA - Return On Assets (Net income/Total Assets)
Solvency 8. Solvency ratio (Equity/Total Liabilities)
Table 12 Financial variables (efficiency outcomes)
* *
1 1
1 1
1
m n
j j i i
j i
m n
j i
j i
w Y b X
w b
(16)
where *
jY and *
iX are the value functions of jY and iX , respectively, ib and jw are the weight
coefficients, n and m are the number of factors. It should be noted that *
jY and *
iX are piecewise
linear, monotone value functions, normalized in the interval [0, 1], thus: **1
**1
0, 1 for 1,2,...,
0, 1 for 1,2,...,
j
i
j j
i i
y y j m
x x i n
(17)
where *k
jy and *k
jx are the values of the k
jy and k
jx level, j and i are the number of scale levels
of functions *
jY and *
iX , respectively. The scales j and i are defined by the analyst and, when
necessary, linear interpolation is used in order to calculate in-between values.
According to the previous assumptions, and using a goal programming approach, the ordinal
regression analysis equation may take the following form (Wagner, 1959; Siskos, 1985):
* *
1 1
m n
j j i i
j i
w Y b X
(18)
Book of Proceedings
143
where and are the overestimation and the underestimation error, respectively.
The parameters of the model may be estimated using linear programming techniques, where
the objective is to minimize the sum of errors, subject to the following constraints:
Equation (18) for each case (in our application for each new service project)
Monotonicity constraints of *
jY and *
iX .
Normalization constraints of ib and jw .
Non-negative constraints for all the variables of the model.
This model is similar to canonical correlation analysis, under the philosophy of ordinal
regression. A detailed presentation of ordinal regression analysis principles, including a discussion
about model stability may be found in Grigoroudis and Siskos (2002, 2010).
c. Results
The most important results of the ordinal regression model refer to the estimation of importance (
ib and jw ) and performance (average of *
jY and *
iX ) of the examined variables. These results are
normalized in [0, 1] and thus it easy to identify the strong and weak points of the innovation drivers
and results. It should be noted that the ordinal regression model has been applied in three different
periods (in 2004, the year that innovation has been developed, and in 2005 and 2006 after one or
two years), in order to examine potential hysteresis in the relation between the variables.
Table 13 presents the weights and the average performance of the main NSD dimensions. As
it can observed, operational analysis is the most important dimension in all years, it appears to
have one of the lowest average performance indices. In general, the importance, as well as the
performance, of these NSD factors appear unvaried the examined period.
Similarly, the weights and the average performance indices of the financial variables are shown
in Table 14. Regardless of the influence of the general economic environment, it seems that the
most important impacts refer to the improved performance of profitability and turnover. Analytical
results of the ordinal regression model may be found in Charalambous (2009).
Dimension Weights (%) Performance (%)
2004 2005 2006 2004 2005 2006
Strategic Focus 7.22 6.29 8.26 80.88 80.00 81.11
Idea Generation 1.45 1.38 1.79 74.67 74.67 74.67
Preliminary Market Assessment 7.07 5.34 5.33 58.07 54.72 52.75
Operational Analysis 50.64 69.90 51.57 35.66 39.46 36.90
Integrated Market Launch 1.31 1.31 1.24 33.17 33.17 33.17
After Launch Review and Assessment 3.41 3.02 3.91 75.38 74.73 76.02
Market Potentiality 1.57 1.39 1.54 69.77 69.77 69.77
Book of Proceedings
144
Market Synergy 27.32 11.38 26.36 87.92 86.08 87.20
Table 13 Weights and average performance for the drivers
Dimension Weights (%) Performance (%)
2004 2005 2006 2004 2005 2006
Gross profit margin 8.77 10.41 23.14 38.92 68.53 75.60
Net profit margin 3.40 4.57 3.53 32.35 31.49 45.63
Asset turnover 28.91 3.55 7.94 91.09 46.24 79.20
Equity turnover 19.38 13.48 25.81 34.62 76.20 25.68
Liability turnover 3.98 13.34 4.03 62.95 88.23 64.05
ROE 10.58 34.03 5.65 19.00 7.34 48.33
ROA 3.38 3.59 3.45 37.65 55.94 50.56
Solvency ratio 21.60 17.03 26.44 77.88 84.07 72.97
Table 14 Weights and average performance for the outcomes
4. Conclusions
This study examined the linkage between drivers and outcomes of NSD applying an ordinal
regression model. The results reveal specific factors that seem to play an important role in
innovation efficiency performance. For example, operational analysis appear as the most important
NSD dimension in all of the examined years. This category refers to one of the initial stages of
NSD and can be characterized as a test of project feasibility and profitability. Market synergy is
also an important dimension. It is not a particular stage of the NSD process, but represents the
control of the market, targeting mainly to the consumer. Market synergy concerns the
harmonization of new services with the market and the customer desires and needs. Another
important dimension refers to strategic focus, which is the first stage of the NSD process, where
the innovation strategy is formulated in agreement with business strategy and objectives. Its
importance is justified by the relevant literature, since a company that identifies appropriate areas
of interest, can set long term goals in the market. Regarding the financial variables, the most
important ratios refers to equity turnover and solvency ratio, which appear very important in all of
the examined years. Other important variables, although their weights may vary during the
examined period, include: gross profit margin, asset turnover, liability turnover, and ROE.
These findings reveal the importance of financial liquidity and managerial efficiency for the
hotel industry (i.e., the ability of a firm to use available resources in order to achieve specific sale
goals). The aforementioned variables can determine how quickly and effectively assets are
converted to cash. In general, the findings show the emphasis that should be given on the one hand
to the customer needs, and on the other to the effective management of a NSD project.
Book of Proceedings
145
References
Avlonitis, G., P. Papastahopoulou, and S. Gounaris. “An empirically-based typology of product
innovativeness for new financial services: success and failure scenarios”. Journal of Product
Innovation Management, Vol. 18 No. 5, 2001, pp. 324-342.
Charalambous, M. Evaluation of innovation efficiency in tourism businesses. Diploma Thesis,
Technical University of Crete, Chania, 2009 (in Greek).
Dolfsma, W. “The process of new service development: Issues of formalization and
appropriability”. International Journal of Innovation Management, Vol. 8 No. 3, 2004, pp. 319-
337.
Grigoroudis, E. and Y. Siskos. “Preference disaggregation for measuring and analysing customer
satisfaction: The MUSA method”. European Journal of Operational Research, Vol. 143 No. 1,
2002, pp. 148-170.
Grigoroudis, E. and Y. Siskos. Customer satisfaction evaluation: Methods for measuring and
implementing service quality, Springer, New York, 2010.
Kitsios, F. Innovation management in new service development, PhD Thesis, Technical University
of Crete, Chania, 2005 (in Greek).
Kitsios, F., M. Doumpos, E. Grigoroudis, and C. Zopounidis. “Evaluation of new service
development strategies using multicriteria analysis: Predicting the success of innovative
hospitality services”. Operational Research: An International Journal, Vol. 9 No. 1, 2009, pp. 17-
33.
Nås, S.O. and A. Leppãlahti. Innovation, firm profitability, and growth, STEP Report 1/97, The
STEP Group, Oslo, 1997.
North, D. and D. Smallbone D. “The innovativeness and growth of rural SMEs during the 1990s”.
Regional Studies, Vol. 34 No. 2, 2000, pp.145-157.
Siskos, J. “Analyses de régression et programmation linéaire”. Révue de Statistique Appliquée,
Vol. 23 No. 2, 1985, pp. 41-55.
Thwaites, A. and P. Wynarczyk. “The economic performance of innovative small firms in the
South East region and elsewhere in the UK”. Regional Studies, Vol. 30 No. 2, 1996, pp. 135-149.
Wagner, H.M. “Linear programming techniques for regression analysis”. Journal of the American
Statistical Association, Vol. 54, 1959, pp. 206-212.
Zahra, S.A. and J.G. Covin. “The financial implications of fit between competitive strategy and
innovation types and sources”. The Journal of High Technology Management Research, Vol. 5
No. 2, 1994, pp. 183-211.
Book of Proceedings
146
New Technologies & Labor Market
Nikolaidis K.
E-mail of author: [email protected]
Abstract
Today the new working relationships have major changes. The new forms such as telecommuting,
workers by lease, part-time employment, the fourth shift give another dimension to the workplace.
Telecommuting is any form of work that includes electronic data processing and the use of media
for multiple/cross communication so that the employee can produce the work he was asked in an
area outside the space where the business is located. There are alternative names for telecommuting
in the relevant bibliography such as teleworking at home or distance working.
Keywords TELECOMMUNICATION, TELEWORKING, Labour Market
1. INTRODUCTION
What led to the development of telecommuting? This question can be answered by the
development of IT and telecommunications. The new forms of communication open new
possibilities in computing via high-speed transmission of the data with the VDSL at the speed of
50 mbps. At that speed we can have perfect image in HD and high quality sound. So the potential
weakness in the communication of the past have been overcame. The globalization of the economy
is another fact which has led to the development of telecommuting. In our day and time, the
economy and consequently the firms operate globally, with the result that workers face a flexibility
issue. Both the business forms of work have changed. The new trends that have appeared nowadays
in the field of teleworking regarding the new forms of telework are as follows: The international
literature identifies the following types of telework :
● Home Based Teleworking : Teleworking is made home-based (exclusively or on a regular basis).
An area of the house converted into office with the proper equipment (computer, telephone,
modem, fax and stationery).
●Satellite Centers : These centers are used by the employees of the same organization and are
located in remote areas near the homes of the workers.
Book of Proceedings
147
●Telework Centers : They are well organized spaces with access to telecommunication and
electronic equipment, in the form of offices used by the employees of different companies, or
employees of the same company-who belong to different fields of work-or even by self-employed
ones with some basic lease.
●Televillages : It is the modern form of telecottage. Entire villages are equipped with the
appropriate technological apparatus, whose houses are “wired” to be able to communicate with
each other and with other villages.
●Teamwork from distance: Some examples are as telemedicine, tele-education, e-commerce and
research from a distance.
2. SECTION Ι
The employer- employee working relationship may be based on a contract of dependent
employment where the former has at his disposal the entire labor power of the worker, who is
considered employed. In any other type of contract the worker is self-employed. Based on the
above we have 3 types of employment relations.
Full-time employment is perfomed at home and concems one employer by this term we mean that
the work is done entirely at home and is not related to working hours.
In part-time employment, work is carried out partly at home and the rest at the employer’s
premises.
The self-employed type is more flexible as the employee works at home for more than one
employers. In this case the worker has his own working model and determines his employment
relationship.
Another division regarding the forms of teleworking is related to the use of IT, and whether it is
necessary for the worker or to be online offline. In the case online work the worker is online with
the company and there is not enough freedom in the time and pace of their work, which means that
he should abide by company’s actual working hours. In offline work the employee has greater
freedom and flexibility in the management of his work since he can link to the company’s network
only when necessary. So he can manage his time the way he thinks is the best. Whether the contact
is online or offline is also a very important factor in the process of telecommuting. Both styles
have positive and negative effects on this process.
In online communication, the worker directly depends on the understanding of the presentation
and discussions as they are conducted and by whether he takes good notes or has a good memory.
Book of Proceedings
148
During the same period, the contributions of the project leader or keynote speaker as well as those
of the participants are almost spontaneous.
On the other hand, in offline communication, the distance worker has more time to think about his
contribution and less pressure to respond immediately. Which form of communication is most
appropriate depends on what kind of activities it will support. For example, offline communication
is better suited for file transfer, information retrieval, etc., while realtime communication is very
useful for the communication and discussion of specific implementation issues or employment
problems. Thus, depending on the case, both forms of communication may be used by an entity
that implements telecommuting procedures.
Many different disciplines and fields are now ripe for the implementation of flexible working
arrangements. The general factors that can serve as criteria for diversification and broad axes of
direction to introduce teleworking schemes in business operations are generally working without
personal contact, task management through profit results and tasks related to the management and
electronic processing of data. These factors generally cover effectively the organizational and
physical side of the work, but work as a social institution has a social dimension too. So when we
analyze key factors for conducting business through flexible working arrangements we should also
include on the analysis level those factors which are directly related to social interaction (such as
sales, insurance) .
In general, the global literature indicates that the characteristics that make a job suitable for
integration into telecommuting shapes are:
1. The ability to be handled without constant personal contact and interaction with other people.
2. The ability to organize the necessary social contacts on a periodic basis. The work that currently
require daily meetings can be reoganised aiming at the integration of partial work at home into the
working pattern.
3. The ability to be manageable through profit results or by agreeing to meet specific objectives in
a given time. Experience has shown that teleworkers need short-term goals if they are to work
effectively.
4., A possible access from distance (either electronically or by telephone conference) via remote
device / PC or with a permanent connection to a specific database in cases where access to
information is needed on a daily basis.
5. The possibility of a task, where a job is directly dependent in cases where a job on the time of
delivery of the products or to be delivered by electronic means or "hand in hand " or via courier.
On the basis of the above factors, the key sectors that have already adopted some forms of flexible
working procedures in Greece and are expected to continue such practices are summarized as:
Distance education , Telemedicine
Book of Proceedings
149
Marketing & advertising.
At this point, it must be stressed that teleworking should not be considered the ideal working
method for all occupations. Occupations having as a prerequisite personal contact or manual labor
cannot implement telework.
As far as enterprises are concerned the main advantage of telecommuting is the increase in
productivity. This increase is mainly due to lower vacation periods during labor and greater
concentration of workers, increased motivation and job satisfaction, greater commitment to work
in the absence of lost time and hassle when traveling.
Arguably, the shift to teleworking and systematic use creates and projects the image of an
innovative and advanced company which streamlines the organization of work, benefits from the
information society and adjusts to modern developments.
The greatest and most immediate benefit is the gain from the reduction of operating costs. ¨ We
save in salaries and travel costs, we limit the need for premises and therefore the fewer the
buildings the less the maintenance costs.
With the implementation of telework we greatly contribute to the reorganization of enterprises by
increasing the workers productivity and by improving the management of their tasks. All these
have as a result the increased competitiveness of the company.
With this new form of work, the company has more flexibility in the rational management of staff.
The term "office" as a fixed spatial point ceases to exist. The company is no longer defined by the
offices occupied, but as a network of relationships , which are connected through
telecommunication networks. In this way, the opportunity for access to the labor market is given
even to geographically remote areas.
The desire for greater self-determination and control of time that the employee has leads to the
adoption of flexible forms of work. The possibility offered by telecommuting to workers not to
make unnecessary movements,or need to communicate with their colleagues in the narrow sense
of an office and a specific timetable makes telework very attractive to a large number of workers.
2.1. Sub Section Ι
At this point we can see what the resulting problems resulting of telework are ¨ We have problems
with the educational system, the level of penetration and the use of technologies. There are also
many questions about the social context of work which are not sufficiently circumscribed.
Another important problem is that companies have not integrated their information infrastructure
due to the large initial financial cost needed for the initial installation.
Book of Proceedings
150
The employer cannot control and supervise the employees due to the fact that they are not
constantly at the workplace.
At this point a significant question arises. Whether companies can continue to require commitment
from employees while they themselves do not commit; I believe the answer to this question is not
easy especially nowadays when in Europe there is massive unemployment, plaguing mostly young
people aged up to 35 years. Let’s not forget that teleworking and the new forms of work under
discussion refer mainly to younger age groups of workers.
This research we are conducting ends up with some suggestions that, I think, will improve the
employment frame of telework and the new forms of work related to telecommunications and
computing. These proposals do not necessarily cost money but they are important in order to
improve and institutionalize necessary and sufficient conditions for the proper implementation of
telework. So, I will indicatively mention the implementation of teleworking in the public
administration. As well as the education of young people on flexible types of work even at school,
for example in the subject of professional orientation.
Of course, there are some suggestions related to businesses. The field of teleworking is new and
thus there is considerable margin for optimization as well as proposals that will contribute to the
better implementation of new technologies. For instance, the application of pilot programs, the
study of all the matters related to human resources and new technologies.
Another point that we will focus our proposals in the field of telework on has to do with the
employee. It is equally important to make suggestions and take steps in industry associations
aiming at the collective representation of workers. Teleworkers are entitled to claim the same rights
as employees who are on the premises.
2.1.1. Sub Section ΙΙ
With basis to the investigations Eirobserver, “Social Partners sign teleworking accord”, Ecat-IST
Programme Key Action II, SBIS General Population Surveys, 2002 EMERGENCE 2000-2003,
which took place in 2002, we will present some tables and statistics presenting the situation in E.E.
from 2002 until today. First we should note that the Netherlands and the Nordic countries are those
that the precede in Europe. The Great Britain is above the average, followed France, Italy and
Spain, while Germany is very close to average.
COUΝΤRY Home Based
Teleworking (full
time)
Home Based
Teleworking
(additional work)
Total of
teleworking
AUSTRIA 2 4.7 6.7
BELGIUM 2.2 5.3 7.5
DENMARK 2.6 15.1 17.7
FINLAND 4.7 11.0 15.7
Book of Proceedings
151
FRANCE 2.2 2.3 4.4
GERMANY 1.6 6.3 7.9
GREECE 2.1 3.9 6.0
IRELAND 0.5 5.5 6.0
ITALY 0.8 1.7 2.5
LUXEMBOURG 0.9 2.4 3.3
HOLLAND 9.0 11.6 20.6
PORTUGAL 0.5 1.1 1.6
SPAIN 0.3 2.0 2.3
SWEDEN 5.3 9.5 14.9
GREAT BRITAIN 2.4 8.5 10.9
AVERAGE E.E 2.1 5.3 7.4
From the above table we see that Greece is close to the average of E.U. The Nordic countries have
greater penetration to telework in relation to the countries of the South. Then the thesis will try to
study the behavior of employees in relation to the evolution of telecommunications and the
development of the internet from 1 Mbps up to 100Mbps. We will study how the labor cost is
affected by the speed of the internet and the possible scenarios that arise from this study.
3. Conclusions
In this paper we propose a methodology that may be useful at improving the current framework as
an additional tool in the sector of telecommunication. At this point we can see what the resulting
problems of telework are. We have problems with the educational system, the level of penetration
and the use of technologies. There are also many questions about the social context of work which
are not sufficiently circumscribed. Another important problem is that companies have not
integrated their information infrastructure due to the large initial financial cost needed for the initial
installation. The employer cannot control and supervise the employees due to the fact that they are
not constantly at the workplace.
This thesis will contribute to scientific research approaching the following questions.
Is there potential improving for the way to use the telecommunications?
How can we make the substitution of labor by telecommunications?
Is there clear evidence that this resource will be used by this research will reduce the overall cost
of labor.
Book of Proceedings
152
References
(1) Ali, M.S. (2002). Information resource centre : mainstream for the flow of information for lifelong
learning. Paper presented at the XV annual conference of the Asian Association of Open Universities
(AAOU), 21G23 February 2002, New Delhi, India.
(2) Anastasiades P.A. (2003). ‘Distance learning in elementary schools in Cyprus: the evaluation
methodology and results’. Computers & Education, 40(1), pp. 17G40.
(3) Bates, A.W. (1993). ‘Theory and practice in the use of technology in distance education’, in Keegan,
D. (Ed.), Theoretical Principles of Distance Education , London: Routledge, pp.213G233.
(4) Garrison, D. R. (2000). ‘Theoretical challenges for distance education in the 21st Century: A shift from
structural to transactional issues’. International Review of Research in Open and Distance Learning 1(1)
(pp. 7G13),
(5) Dabholkar, P.A. (1994), "Technology-based service delivery: a classification scheme for developing
marketing strategies", in Swartz, T.A., Bowen, D.E., Brown, S.W.(Eds),Advances in Services Marketing
and Management , JAI Press, Greenwich, CT, Vol. Vol. 3 pp.241-71.
Book of Proceedings
153
Regression modeling for spectral data sets: A multi-objective genetic
approach
Loukas Dimos
Agricultural University of Athens, Department of Food Science & Human Nutrition, Laboratory
of Microbiology and Biotechnology of Foods, Iera Odos 75, Athens 11855, email:
Ropodi Athina
Agricultural University of Athens, Department of Food Science & Human Nutrition, Laboratory
of Microbiology and Biotechnology of Foods, Iera Odos 75, Athens 11855
Nychas George-John
Agricultural University of Athens, Department of Food Science & Human Nutrition, Laboratory
of Microbiology and Biotechnology of Foods, Iera Odos 75, Athens 11855
Abstract
In prediction problems, the finding of the best regression model in terms of quality-of-fit and
parsimony is of great importance. Thus, the selection of a subset of the most informative and
uncorrelated variables (Variable Selection problem-VS) is critical for the model's performance.
This work focuses on the multi-objective modeling of the VS problem especially for spectroscopic
applications. The proposed methodology considers the VS for the Partial Least Squares Regression
(PLS-R) modeling as a two objective task, minimizing the number of selected variables as well as
the Mean Square Error MSE) of prediction. The model selection is a two step procedure: (i) the
NSGA-II genetic algorithm is applied in order to generate the frontier of Pareto-optimal solutions
with respect to the multi-objective formulation of the problem, and (ii) a decision making process,
based on information metrics, enables the selection of the final regression model. The
aforementioned method was applied on spectral data related to the microbiological quality of
minced meat samples, which were acquired by means of a spectroscopic instrument (Fourier
Transform InfraRed - FTIR).
Keywords: multi-objective optimization, genetic algorithm, variable selection, spectral data, meat
quality
Book of Proceedings
154
1. INTRODUCTION
VS is one of the most critical steps in statistical modeling. Suppose Y a dependent variable and
X={X1,..,Xp} a set of potential predictors, are vectors of n observations. The problem of variable
selection, or subset selection, arises when one wants to model a relationship between Y and a
subset of X, but there is uncertainty about which subset to use (George, 2000). This becomes harder
in cases of high complexity or irrelevant information(noise).
In regression modeling, VS can viewed as a special case of the model selection problem,
where each model under consideration corresponds to a distinct subset of X of the most 'significant'
variables. The 'best' model corresponds to the most parsimonious model, regarding the bias-
variance trade-off (Hastie et al., 2009). More specifically, the low complexity models result in
high bias in regressors and an underfit model that fails to identify all important variables, while
the high complexity models result in high variance in regressors and an overfit model that cannot
be generalized beyond the observed sample data. Typically we need to choose the 'best' balance
between complexity(to be minimized) and quality-of-fit (to be maximized). However, it is
impossible in practice to optimize simultaneously both objectives and we usually select a good
subset of variables instead of the optimal one.
VS is mainly applied in areas for which datasets with hundreds or thousands variables are
available. In biological sciences, high dimensional multivariate data sets are frequently generated
by means of analytical instruments, or 'sensors', that are based either on vibrational spectroscopy
or surface chemistry. Among the most popular spectroscopic techniques is FTIR, which involves
the rapid acquisition of absorbance values within a specific range of 3700 wavenumbers (400 cm-
1-4000 cm-1) of the IR spectrum. The information contained in the spectral data can be used for the
prediction of the molecular fingerprint of the sample (Nychas et al., 1998). Supervised linear
modeling techniques such as PLS-R (Marten & Naes, 1989) are ideal for the analysis of
spectroscopic data. PLS-R is based on latent variables and produces reliable full-spectrum models
that are almost insensitive to noise (Leardi, 2003). Nevertheless, VS can be highly beneficial for
the regression modeling when the number of variables to select is large compared to the number
of samples, which is the typical biological case.
The simplest method for model selection would be to examine all possible combinations
of variables by means of an exhaustive search. However, for large number p of initial variables the
VS problem is known to be NP-hard with time complexity O(2p) (Amaldi & Kann, 1998). A large
number of heuristics methods have been proposed for providing solutions in a reasonable amount
of time, most of the them can be roughly categorized as following: (i) Information criteria methods
(AIC; Akaike, 1974), (BIC; Shwarz, 1978), (ii) Stepwise selection methods (forward selection,
backward elimination and stepwise regression), (iii) Evolutionary Algorithms (Leardi, 2003; Jarvis
& Goodacre, 2005), (iv) Statistical approaches (cross-validation; Shao,1993), (bootstrap;
Efron,1986), (PLS- based methods; Nørgaard et al., 2000) (v) Regularization techniques (Lasso;
Tibshirani, 1996), (vi) Stochastic approaches (Bayesian Model Averaging; Clyde et al., 2011),
(Competitive Adaptive Reweighted Sampling - CARS; Li et al., 2009), and (vii) Decision Trees (
CART; Breiman et al., 1984, Random Forests; Breiman, 2001). For more details, please refer to
the Guyon & Elisseeff (2003), Mehmod et al. (2012) and Fan & Lv (2010).
Although widely used, most of those classical approaches treat VS as a single-objective
problem (i.e. minimizing MSE) with a complexity penalty scheme to represent the trade-off
between empirical risk and model complexity. Due to difficulty of the penalty choice, the
sensitivity of the single solution might be quite critical especially for the case of high-dimensional
data. For this reason, a two-stage multi-objective approach is proposed for the VS by this study.
Book of Proceedings
155
In the first stage, an optimization procedure based on the application of the Multi-Objective
Genetic Algorithm (MOGA) NSGA-II (Nondominated Sorting Genetic algorithm) (Deb, 2001) is
introduced in order to provide the Decision Maker (DM) with the Pareto front of the optimal
solutions in terms of quality-of-fit and complexity. Subsequently, in the decision-making stage,
the AIC criterion is used for choosing the appropriate model among a reduced set of good models.
In the next three paragraphs, we will firstly introduce the multi-objective formulation of
the model selection problem. Secondly, the proposed methodology for PLS-R modeling will be
presented and an application for the FTIR data of a microbiological experiment will be finally
given.
2. MODEL SELECTION AS A MULTI-OBJECTIVE PROBLEM
Regression modeling is a special case of supervised learning. Let X a fixed input space, Y the
output space and S={Xi,Yi} n training samples drown independently and identically distributed
from an unknown distribution D(X,Y). The problem of model selection is to find the appropriate
model Yf X: with minimal error on the training set with respect to D. More formally, the model
is assumed to belong to a predefined hypothesis-space H, which for linear regression is the
following:
Jk
kkkk XxpJxxH ,,...,1/ (1)
H forms a nested structure HHH t ...... 1 where Ht represents the subset of models with t many
variables. By interpreting the model selection problem as finding the 'best' trade-off between
complexity and quality-of-fit , we can formulate the following bi-objective problem where the
objectives are jointly minimized (Sinha et al., 2013):
Definition 1. Let ,: H ),( 21 a bi-objective function where:
i. the first objective dHfdH :min:1 represents the complexity of the
model in terms of the number of variables; and
ii. the second objective
n
i
ii fYn
H1
2
2
1min: X represents the quality-of-fit
of the model in terms of its generalization error.
The optimization problem is given by:
)),(),(()(min 21 fffHf
(2)
s.t. Cf , HC
Both objectives of (2) conflict to each other in a sense that the improvement of the one
leads to deterioration of the other. The proposed multi-objective framework provides us with the
set of the best trade-off solutions, called Pareto optimal solutions. The Pareto optimality concept
is formally defined as follows (Deb, 2001):
Book of Proceedings
156
Definition 2. A model f(1) is said to dominate the other model f(2), denoted f(1) f(2), if
)2()1( ff jj and )2()1( ff jj , 2,1j .
Definition 3. A feasible solution Cf * of problem (2) is called a Pareto optimal solution ,if
Cg that fg . The set of all the Pareto optimal solutions is called the Pareto set: PS=
.:/ fgCgCf
The image of the PS in the objective space is called the Pareto Front: PF= ./ PSff
3. THE PROPOSED MULTIOBJECTIVE METHODOLOGY FOR PLS-R MODELING
Due to its latent origins, PLS-R requires specific treatment. For this reason, the proposed
methodology will be analyzed in two parts. In the first part, the optimization of the &2 problem
will be introduced by means of the application of an hybrid version of the NSGA-II. In the second
part, the overall procedure for the PLS-R modeling will be given by an analytical scheme.
3.1 The hybrid NSGA-II for model selection
In order to avoid overfitting, an out-of sample estimator of the MSE, the MSE of cross-validation
(MSE-CV) has been chosen as the second objective. Cross-validation is a method for model
selection based on the predictive ability of the models. In its simplest form, the leave-one-out
cross-validation (LOOCV; Stone, 1974), it considers an initial set of n data points and splits it in
two parts. The first part contains n-1 points for fitting a model (model construction), whereas the
last one point is used in the second part for the validation of the model. LOOCV selects the model
with the best average predictive ability calculated, based on all the n divisions of the initial data
set (Shao, 1993). Although computational expensive, the hybridization of the NSGA-II with the
LOOCV optimization technique ensures that only the best models will be selected in every step of
the procedure and also that the final solutions will be as accurate as possible in terms of the
generalization error.
The steps of the proposed MOGA for model selection are the following (Deb, 2001; Sinha et
al. 2013):
1. Initialization: A binary vector of the size of the number of the p variables is initially
introduced. It is denoted as a chromosome. If a particular variable is present, the bit value
is 1; otherwise it is zero. The random combination of the zero-one genes formulates an
individual. For the model selection problem, each one individual corresponds to a PLS-R
model for a specific subset of the p variables. A parent population Pt, of size p, is initialized
by picking the regression variables with uniform probability.
2. Crossover: A single point crossover of the binary strings of two randomly chosen members
of the Pt is used for the creation of two offsprings. The procedure is repeated with different
parents until a Ot population of p offspring members is produced.
3. Mutation: A binary mutation on each offspring is performed by flipping the bits with
probability pm=1/p.
4. Evaluation & Non-Domination Sorting: The combination of the Pt and Ot populations
results in an intermediate population Rt of size 2p. Each member of the Rt is evaluated with
respect to the two objective functions. Those fitness results are used for the ranking of the
individuals into different non-dominated fronts. Subsequently, a new parent population
Book of Proceedings
157
Pt+1 is created by choosing firstly individuals of the best ranked fronts followed by the
next-best and so on, until p individuals are obtained
5. Stopping criteria: Steps 2 to 4 are repeated until a number of 500 generations is reached.
3.2 The overall methodology for PLS-R modeling
According to the principles of supervised learning, the initial data set is divided into training and
test set. For the training set, the maximum number of PLS components is firstly extracted by an
application of the PLS-R for the set of the initial variables. Subsequently, by a sequential
application of the MOGA approach of &3.1 and the AIC criterion a single solution is resulted,
which is validated by the test set. The whole procedure is graphically described by the next figure
(Figure 1).
Figure 15: The overall multi-objective methodology for PLS-R modeling
4. APPLICATION
The proposed methodology was applied on experimental data concerning the microbiological
quality of minced meat samples. For this experiment, minced beef meat of normal pH was obtained
from a central retail of Athens. It was divided in portions of 70-80g, placed onto styrofoam trays
and stored aerobically and under Modified Atmosphere Packaging (MAP) at 4 and 10oC.
Microbiological analysis (total viable counts - TVC) was performed every 12 and 24 hours for
samples stored at 10 and 4 oC, respectively. In parallel, Fourier Transform Infrared Spectroscopy
spectra were collected from which the fingerprint area (1800-800 cm-1) of the FTIR spectrum was
selected. In order to deal with the multicollinearity of the data, and also to increase the performance
of the algorithm, we performed sub-sampling (of window 5) and reduced the number of variables
(wavenumbers) to 207. Subsequently, a Savitzky-Golay smoothing was introduced for noise
reduction (Brown & Wentzell, 1999) and autoscaling was selected in order to enhance the variation
of the most uninformative variables (Leardi, 2003).
Autoscaled FTIR data were used for the prediction of microbial counts, regardless of
storage and packaging conditions. For an initial number of 12 PLS components, the application of
the hybrid NSGA-II of the &3.1., provided the DM with a Pareto set of 20 non-nested PLS-R
models (Figure 2a). The resulted variations in complexity (from 3 to 27 variables) and in MSE
(from 0.0327 to 0.2832) of the training data, enhance the DM to understand more thoroughly the
requested trade-off that separates the alternative models. At this point, DM should decide either to
keep the best model in MSE (Model of 27 variables: training (0.0327), test (0.0404)) or further
examine the solutions of the optimal frontier. In second case, the application of the corrected AIC
(AICc) for small samples (n<40):
1
122)log(
kn
kkk
n
RSSnAICc
(3)
Train Set
Test Set
Maximum number
of PLS components
Pareto front of
optimal
solutions
Post-Optimality
Analysis
Final
Solution
Book of Proceedings
158
( nMSERSS and k=number of the model variables) selects the model of 12 variables (Model
12) as the optimum one in terms of the trade-off between complexity and the quality-of-fit. The
application of the PLS-R for the provided 12 regressors results to MSEtest=0.1979 (Figure 2b).
Figure 2: (a) Pareto plot; and (b) Plot of observed versus predicted log values of TVC (Model
12)
5. CONCLUSIONS
This work proposed a methodology for PLS-R model selection for spectral data set. A hybrid
version of the multi-objective genetic algorithm NSGA-II was introduced for the identification of
the Pareto front optimal models. In case of the single solution, a second stage decision making
process which is based on the AIC criterion was suggested. The computational results showed,
verified that the proposed approach constitute a good alternative for the model selection problem.
However, a further validation with many more data sets is required.
ACKNOWLEDGEMENTS
This work has been supported by the project “Intelligent multi-sensor system for meat analysis -
iMeatSense_550” co-financed by the European Union (European Social Fund – ESF) and Greek
national funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework (NSRF) - Research Funding Program: ARISTEIA-I.
REFERENCES
Akaike, H. "A new look at statistical model identification". IEEE Transactions on Automatic
Control, Vol. 19, 1974, pp. 716-723
Amaldi, E., and Kann, V. "On the approximation of minimizing non zero variables or unsatisfied
relations in linear systems". Theoretical Computer Science, Vol. 209, 1998, pp.237-260
Breiman, L., Friedman, J. H., Olshen, R. A., and Stone, C. J. Classification and Regression
Trees. Wadsworth & Brooks, 1984
Book of Proceedings
159
Breiman, L. Random Forests, Berkeley, CA, USA: Statistics Department, University of
California, 2001
Brown, C.D., and Wentzell, P.D. "Hazards of digital smoothing as preprocessing tool in
multivariate calibration". J. of Chemometrics, Vol. 13,1999, pp. 133-152
Clyde, M., Ghosh, J., and Littman, M. "Bayesian Adaptive Sampling for Variable Selection and
Model Averaging". Journal of Computational and Graphical Statistics, vol. 20, no. 1,
2011, pp. 80–101.
Deb, K. Multi-Objective Optimization Using Evolutionary Algorithms, John Wiley & Sons Ltd.,
2001
Efron, B. "How biased is the apparent error rate of a prediction rule?". J. of American Statistical
Association, Vol. 81, 1986, pp. 461-470
Fan, J., and Lv, J. "A selective overview of variable selection in high dimensional feature space".
Statistica Sinica, Vol. 20, 2010, pp. 101-148
George, E.I. "The Variable Selection Problem". Journal of American Statistical Association,
Vol.95, No. 452, 2000, pp.1304-1308
Guyon, I., and Elisseeff, A. "An Introduction to variable and feature selection". Journal of
Machine Learning Research, Vol. 3, 2003, pp. 1157-1182
Hastie, T., Tibshirani R., and Friedman J. The Elements of Statistical Learning (2nd edition).
Springer, New York, USA, 2009
Jarvis, R.M., and Goodacre, R. "Genetic algorithm optimization for pre processing and variable
selection of spectroscopic data". Bioinformatics, Vol. 21, No 7., 2005, pp. 860-868
Leardi, R. "Genetic algorithm-PLS as a tool for wavelength selection in spectral data sets", in
Nature-inspired Methods in Chemometrics and Artificial Neural Networks (Leardi, R., ed.).
Elsevier, Amsterdam, 2003
Li, H.-D., Y.-Z. Liang, Q.-S. Xu, & D.-S. Cao. "Key wavelengths screening using competitive
adaptive reweighted sampling method for multivariate calibration". Anal. Chim. Acta 648, 2009,
pp. 77-84
Marten, H., and Naes, T. Multivariate Calibration, Wiley, Chichester, 1989
Mehmod, T., Liland, K. H., Snipen, L., and Sabo, S. "A review of variable selection
methods in Partial Least Squares Regression", Chemometrics & Int. Lab. Systems, Vol.
118, 2012, pp. 62-69
Nørgaard L., Saudland A.,. Wagner J, Nielsen J.P., Munck L. and Engelsen S.B. "Interval Partial
Least Squares Regression (iPLS): A Comparative Chemometric Study with an Example from
Near-Infrared Spectroscopy", Applied Spectroscopy, Vol. 54, 2000, pp. 413-419
Book of Proceedings
160
Nychas, G. J. E., Drosinos, E. H., and Board, R. G. "Chemical changes in stored meat", in The
microbiology of meat and poultry (Board, R.G., and Davies A. R., eds), pp. 288–326. London:
Blackie Academic and Professional, 1998
Schwarz, G. E. "Estimating the dimension of a model". Annals of Statistics, Vol. 6, Νο. 2, 1978,
pp. 461–464
Shao, J. "Linear Model Selection by Cross-Validation". J. of the American Statistical
Association, Vol.88, No.422, 1993, pp. 486-494
Sinha, A., Malo, P., and Kuosmanen, T. "A Multi-objective Exploratory Procedure for
Regression Model Selection". Journal of Computational and Graphical Statistics, (In Press),
2013.
Stone, M. "Cross-validatory choice and assessment of statistical predictions". J. Roy. Stat. Soc.
B Met, Vol. 36, No. 2, 1974, pp. 111-147
Tibshirani, R. "Regression shrinkage and selection via the lasso". J. Roy. Statist. Soc, Ser. B58,
1996,pp. 267–288.
Book of Proceedings
161
Optimal use of non-collaborative servers in two-stage tandem queueing
systems
Dimitrios Pandelis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece, email: [email protected]
Ioannis Papachristos
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Abstract
We consider two-stage tandem queueing systems with a dedicated server in each queue and a
flexible server that can attend both queues. We assume exponential interarrival and service times,
and linear holding costs for jobs present in the system. We study the optimal dynamic assignment
of servers to jobs assuming a non-collaborative work discipline with idling and preemptions
allowed. We formulate the problem as a Markov decision process and derive structural properties
of the optimal policy. For larger holding costs in the upstream station we show that i) non-idling
policies are optimal, and ii) if there is no dedicated server in the first station, the optimal allocation
strategy for the flexible server has a threshold-type structure. We also provide numerical results
that reveal that under the non-collaborative assumption the optimal policy may have
counterintuitive properties, which is not the case when a collaborative service discipline is
assumed.
Keywords: Tandem queues, Flexible servers, Markov decision processes.
1. Introduction
We study the optimal dynamic server assignment in a two-station tandem queueing system with
one dedicated server for each station and one flexible server that can work in both stations. In
particular, we seek server allocation strategies that minimize linear holding costs for systems with
Poisson arrivals and exponential service times. The problem we consider is motivated by the
increasing use of flexible resources, such as cross-trained workers and reconfigurable machines,
in order to cope with varying demand and changing operating conditions. We refer interested
readers to Hopp and Van Oyen (2004) and Andradottir et al. (2013) for extensive literature surveys
on workforce flexibility.
Because of the complexity of the mathematical models involved, research on the optimal use of
flexible servers with holding costs has focused on two-stage systems. Farrar (1993), Wu et al.
(2006), and Pandelis (2007) considered different versions of a system without arrivals (clearing
system) with dedicated servers in each station and one flexible server that is either constrained to
work in the upstream station or can work in both stations. They showed that the optimal policy is
Book of Proceedings
162
characterized by a switching curve; the flexible server is idled or assigned to the downstream
station if the number of jobs there exceeds a certain threshold. Wu et al. (2008) and Pandelis
(2008a) showed the optimality of such switching-curve policies for systems with arrivals and no
dedicated server in the upstream station. Finally, Pandelis (2008b) studied a model with server
operating costs in addition to holding costs and identified conditions under which the switching-
curve structure of the optimal policy is preserved.
With the exception of Pandelis (2007) (constrained version), a common assumption in all the
papers referred to in the previous paragraph was that different servers could collaborate to work
on the same job, in which case the total service rate was equal to the sum of the individual servers
rates. Moreover, a non-idling discipline for at least the dedicated servers was assumed in all of
these papers. Both of these assumptions were relaxed by Pandelis (2014) in the case of clearing
systems, where the optimality of threshold-type policies was established for certain special cases.
In this paper we also relax the aforementioned two assumptions and show the optimality of non-
idling policies for larger holding costs in the upstream station. Moreover, we show that when there
is no dedicated server in station 1 and the dedicated server in station 2 is faster than the flexible
server, the optimal policy is determined by a switching curve. In addition, we provide a condition
under which priority is always given to station 1.
The paper is organized as follows. In section 2 we formulate the problem as a discrete-time Markov
decision process. The structure of the optimal policies is derived in section 3. Finally, we discuss
our results in Section 4.
2. Problem formulation
We consider a two-stage tandem queueing system where jobs arrive according to a Poisson process
with rate . After their service is completed in the upstream station (station 1), jobs move to the
downstream station (station 2) where they receive additional service and then leave the system.
Each job in station i , 1,2i , incurs linear holding costs at rate ih . There are dedicated servers,
one for each station, that are trained to work only in their corresponding station, and one flexible
server that can work in both stations. We assume that this server can transfer from station to station
instantaneously without any cost. We assume exponential processing times with rates 1 2, for
jobs served by the dedicated server and 1 2, for jobs served by the flexible server in station 1,2,
respectively. At most one server can be assigned to each job, that is, server collaboration is not
allowed, but two servers can work simultaneously on different jobs in the same station. Our
objective is to find a server allocation strategy that minimizes the total expected discounted holding
cost over a finite time horizon.
Allowing preemptions at times of arrivals and service completions, we formulate the problem as a
Markov decision process with state space 1 2 1 2{( , ) : , 0}x x x x , where 1x , 2x are the number of
jobs in station 1 and station 2 respectively, including those in service. Instead of the continuous
time problem we consider the equivalent discrete time problem obtained by uniformization, where
without loss of generality we assume 1 2 1 2 1 . We denote by 1 2( , )nV x x the
minimum n -step expected cost starting from state 1 2( , )x x , discounted by a factor , and by
Book of Proceedings
163
1 2( , )A x x the set of feasible service rates at state 1 2( , )x x . Then, the value function satisfies the
following optimality equations.
1 21 2 1 2
1 2 1 1 2 2 1 1 2 ; , 1 2( , ) ( , )
( , ) ( 1, ) min ( , )n n nA x x
V x x h x h x V x x W x x
,
where
1 2; , 1 2 1 1 1 2 2 1 1 2 1 2 1 1 2( , ) ( 1, 1) ( , 1) (1 ) ( , )n n n nW x x V x x V x x V x x ,
and
0 1 2( , ) 0V x x .
3. The optimal policy
We start this section with a broad characterization of the optimal policy that reduces the search for
the optimal allocation in a subset of the feasible service rates. First, we provide a monotonicity
property of the value function, which can be proved by a straightforward induction on n .
Lemma 1. 1 2( , )nV x x is non-decreasing in its arguments.
Because there is nothing to gain by keeping jobs in the downstream station, it is reasonable to
allocate as much service rate as possible to that station to push jobs out of the system. This is
established in the following proposition.
Proposition 1. For given 1 , 1 2; , 1 2( , )nW x x is minimized by maximizing 2 .
Proof. For 2 2 we have
1 2 1 2; , 1 2 ; , 1 2 2 2 1 1 2 1 1 2( , ) ( , ) ( ) ( , ) ( , 1)n n n nW x x W x x V x x V x x
,
which is positive by Lemma 1 and the proposition is proved. ■
A consequence of Proposition 1 is that the optimal policy does not idle the dedicated server in
station 2 when there are at least two jobs there. Next, regarding the optimal allocation in the
upstream station, we define function 1 2( , )nf x x as follows:
1 2 1 2 1 2( , ) ( , ) ( 1, 1)n n nf x x V x x V x x , 1 21, 0x x .
Note that for some initial rate allocation 1 2, the incentive to allocate additional rate in station
1 is given by
1 2 1 2; , 1 2 ; , 1 2 1 1 2( , ) ( , ) ( , )n n nW x x W x x f x x .
Book of Proceedings
164
Therefore, the sign of 1 2( , )nf x x determines whether we should add service rate to station 1 or not.
This is formalized in the following proposition.
Proposition 2. For given 2 , 1 2; , 1 2( , )nW x x is minimized by maximizing 1 if 1 1 2( , ) 0nf x x
and by minimizing 1 if 1 1 2( , ) 0nf x x .
Proof. For 1 1 we have
1 2 1 2; , 1 2 ; , 1 2 1 1 1 1 2( , ) ( , ) ( ) ( , )n n nW x x W x x f x x ,
which proves the proposition. ■
According to Proposition 2, depending on the sign of 1 2( , )nf x x , the optimal policy should either
allocate as much service rate as possible to the upstream station or not serve that station at all. In
particular, as far as the dedicated server in station 1 is concerned, the optimal policy should not
idle him when 1 2( , ) 0nf x x and there are at least two jobs upstream, and idle him when
1 2( , ) 0nf x x .
When it is not cheaper to have jobs in station 1 compared to station 2, it makes sense not to idle
resources to keep jobs upstream. This is a consequence of the following lemma.
Lemma 2. Let 1 2h h . Then, 1 2( , ) 0nf x x for all 1 21, 0x x .
Proof. The proof is by induction on n . Let * *
1 2, be the optimal allocations in state 1 2( , )x x . Then,
1 2 1 2( , )nf x x h h * * *1 2 2
1 1 2 1 2 1 2; , ;0,( 1, ) ( , ) ( 1, 1)n n n
f x x W x x W x x
*
1 2 1 1 2 2 1 1 2( 1, ) ( , 1)n nh h f x x f x x
* *
1 2 1 1 2(1 ) ( , ) 0nf x x ,
by the induction hypothesis. ■
Based on the properties obtained so far, the optimal server allocations for 1 2h h are either
explicitly determined (in certain cases with one job in one or both stations) or are restricted to two
choices (see Table 1 below). On the other hand, when 1 2h h , the possible optimal allocations are
those of Table 1 with the addition of allocation 2 2(0,max{ , }) if 2 1x and 2 2(0, ) if 2 1x
.
1 1
2 2
1 1
2 2
1 1
2 2
1 1
2 2
Book of Proceedings
165
(1,1) 1 2( , )
1 2( , ) 1 2( , )
1 2( , ) or 1 2( , )
2(1, )x 1 2 2( , )
1 2 2( , ) or 1 2( , )
1( ,1)x 1 1 2( , )
1 1 2( , ) or
1 2( , )
1 1 2( , ) 1 1 2( , ) or
1 2( , )
1 2( , )x x 1 1 2( , ) or
1 2 2( , )
Table 1 Optimal allocations for 1 2h h
In the rest of the section we further characterize the structure of the optimal policy when 1 2h h ,
there is no dedicated server upstream, and the dedicated server in station 2 is faster than the flexible
server. Specifically, we show that the optimal policy is determined by a switching curve and
identify a condition under which it always assigns the flexible server to station 1. This is stated in
the following theorem.
Theorem 1. Let 1 2h h , 1 0 , and 2 2 . Then,
1) For each 1 1x there exists a threshold value 1( ) 2t x such that the optimal policy assigns the
flexible server to station 2 if 2 1( )x t x , and to station 1 otherwise.
2) If 1 1 2 2 2( )h h h , it is optimal to assign the flexible server to station 1.
Proof. For 1 1x , 2 0x we define function
1 2( , )nd x x 1 1 2 2 1 2 1 2( , ) ( , ( 1) ) ( , )n n nf x x V x x V x x .
For 2 1x this function gives the incentive to assign the flexible server to station 1 instead of
station 2 and can be computed by the following recursive expressions:
1 1 1 2( ,0) ( )nd x h h 1 1 2 1 1( 1,0) ( ,0)n nd x d x
2
1 1 1 1 1 2 1 1 1 1( 1,1) ( 1) ( ,0) ( 1,0)n n nf x x V x V x 1 for 1 0x ,
1 1 1 2 2 2( ,1) ( )nd x h h h 1 1 2 1 1( 1,1) ( ,0)n nd x d x
1 1 1 1 2 1 1( 1,2) ( 1) ( ,1)n nd x x d x
1 for 1 0x ,
2 1 1 2 2 2(1, ) ( )nd x h h h 1 2 2 1 2(2, ) (1, 1)n nd x d x
2 1 2 2 1 1 2 2 1 2(1, 1) ( 2) (1, ) (1, )n n nd x x d x d x
1 for 2 1x ,
Book of Proceedings
166
1 2 1 1 2 2 2( , ) ( )nd x x h h h 1 1 2 2 1 1 2 1 1 1 2( 1, ) ( , 1) ( 1, 1)n n nd x x d x x d x x
2 1 1 2 2 1 1 1 2 2 1 1 2( , 1) ( 2) ( , ) ( , )n n nd x x x d x x d x x
1 for 1 1x , 2 1x .
Applying induction on n we can use the above equations to show that i) 1 2 1 2( , ) ( , 1)n nd x x d x x ,
which suffices to prove the first part of the theorem, and ii) 1 2( , ) 0nd x x when 1 1 2 2 2( )h h h
, which proves the second part.
4. Discussion
We considered optimal server allocations in two-stage queueing systems with dedicated servers at
each stage and one flexible server. In contrast to similar models studied in the past we assumed a
non-collaborative work discipline and allowed server idling. We derived properties of server
allocation strategies that minimize the expected discounted holding cost over a finite time horizon.
It can be shown by standard dynamic programming arguments that these properties are also valid
for the infinite horizon discounted cost criterion and, assuming stable systems, for the average cost
per unit time criterion as well.
The results we managed to obtain were for jobs that accrue at least as much holding costs when
present at the first station compared to the second station, in which case we proved that non-idling
policies are optimal. For this case, when there is no dedicated server upstream and the flexible
server is not faster than the dedicated server in the downstream station, we showed the optimality
of a policy that assigns the flexible sever to the downstream station when the number of jobs there
exceeds some threshold value. This value becomes infinite, that is, the optimal policy always gives
priority to the upstream station, when the holding cost saved from a service completion in station
1 is greater than or equal to the cost saved from a service completion in station 2.
When server collaboration is allowed, it has been shown that when there is no dedicated server
downstream and the holding cost saved from a service completion in station 1 is less than the cost
saved from a service completion in station 2, priority should be given to station 2. Although this
is a reasonable property of the optimal policy, it does not necessarily hold when collaboration is
not allowed. For example, consider a system with arrival rate equal to 0.07, service rate for the
dedicated and flexible server in station 1 equal to 0.1 and 0.05, respectively, service rate for the
flexible server in station 2 equal to 0.85, and holding cost rate for station 1 and station 2 equal to
17 and 1, respectively. The minimum average cost policy obtained by the value iteration algorithm
assigns the flexible server to station 1 when there are two jobs in each station.
Book of Proceedings
167
References
Andradottir S., Ayhan H., and Down D.G. “Design principles for flexible systems”. Production
and Operations Management, Vol. 22, 2013, pp. 1144-1156.
Farrar T.M. “Optimal use of an extra server in a two station tandem queueing network”. IEEE
Transactions on Automatic Control, Vol. 38, 1993, pp. 1296-1299.
Hopp W.J. and Van Oyen M.P. “Agile workforce evaluation: a framework for cross-training and
coordination”. IIE Transactions, Vol. 36, 2004, pp. 919-940.
Pandelis D.G. “Optimal use of excess capacity in two interconnected queues”. Mathematical
Methods of Operations Research, Vol. 65, 2007, pp. 179-192.
Pandelis D.G. “Optimal stochastic scheduling of two interconnected queues with varying service
rates”. Operations Research Letters, Vol. 36, 2008, pp. 492-495.
Pandelis D.G. “Optimal control of flexible servers in two tandem queues with operating costs”.
Probability in the Engineering and Informational Sciences, Vol. 22, 2008, pp. 107-131.
Pandelis D.G. “Optimal control of noncollaborative servers in two-stage tandem queueing
systems”. Naval Research Logistics, Vol. 61, 2014, pp. 435-446.
Wu C.-H., Down D.G., and Lewis M.E. “Heuristics for allocation of reconfigurable resources in
a serial line with reliability considerations”. IIE Transactions, Vol. 40, 2008, pp. 595-611.
Wu C.-H., Lewis M.E., and Veatch M. “Dynamic allocation of reconfigurable resources in a
two-stage tandem queueing system with reliability considerations”. IEEE Transactions on
Automatic Control, Vol. 51, 2006, pp. 309-314.
Book of Proceedings
168
GreenYourRoute platform
Georgios K.D. Saharidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece, email: [email protected]
Abstract
The objective of the proposed research is to develop a Decision Support System (DSS) for a web
based platform which will help individuals and companies move commodities in the most
environmental friendly way, minimizing environmental externalities (e.g. CO2 emissions) and
transportation costs. The developed platform which is the final outcome of an FP7 European
research project and the national Operational Program "Education and Lifelong Learning",
referred to as GreenRoute and EnvRouting projects, uses existing information systems (e.g.
geographical, weather, real time traffic information systems) and emission calculation models as
a basis to apply three main scientific outcomes.
Keywords: Green, vehicle routing, web platform
1. Introduction
EU-15 greenhouse gas emissions (GHG emissions) decreased in all sectors between 1990 and
2006 except in the transport sector, where an increase of 26% was documented (EC, 2008). The
carbon footprint of transport has been constantly increasing inside and beyond the EU, currently
reaching nearly a quarter of the overall greenhouse gases. Governments, organisations and
companies want to better monitor and optimise the environmental impact of all the logistics
operations and movements across the supply chain and the transportation corridors of the EU,
reversing the current trends without jeopardizing international trade and movement. The objective
of the proposed research, henceforth referred to as “GreenYourRoute”, is to develop a Decision
Support System (DSS) for a web based platform which will help individuals and companies move
commodities in the most environmental friendly way, minimizing environmental externalities (e.g.
CO2 emissions) and transportation costs. The developed platform which is the final outcome of
GreenRoute and EnvRouting projects will use existing information systems (e.g. geographical,
weather, real time traffic information systems) and emission calculation models (ECMs) as a basis
to apply their main three scientific outcomes, detailed in the following section.
2. Scientific outcomes
The first scientific outcome of the GreenRoute project is the development of a function that assigns
a score to each arc of EU transportation network referred to as the arc environmental externalities
score (EESarc). EESarc approximates the potential environmental externalities if this arc is used.
EESarc is estimated undependably of the vehicle but depended on the environment (i.e.
transportation network) where the vehicle moves. The second scientific outcome is the
development of a revised version of existing emission calculation models where the EESarc
Book of Proceedings
169
function is incorporated and used as a correction coefficient. The revised emission calculation
models predict the emissions at an arc under different conditions (e.g. traffic conditions). The third
scientific outcome is the development of a novel approach for solving the general Vehicle Routing
Problem (VRP) whose objective is to find the most environmental friendly route. The new solution
approach is the first based on decomposition techniques. Two main types of decomposition
techniques are studied: a) structural decomposition and b) mathematical decomposition. These
three scientific outcomes play the role of a black box which is used by the platform. The
GreenYourRoute platform is published to the web and is available with no cost to the user seeking
to find optimal environmental friendly routes for its vehicle fleet.
Modeling approaches to predict the environmental externalities (e.g. estimate CO2 emissions)
coming from a move of a vehicle, are based on the estimation of fuel consumption by the vehicle.
The fuel consumption is based on several factors with the most popular being distance, time, type
of vehicle (e.g. body style, model year, type of engine etc.), weight load and mode of operation
(e.g. different engine management concepts, gear-shift philosophies etc.). Certainly, the vehicle,
which is the source of emissions, is a very important component for the evaluation and estimation
of environmental externalities coming from freight transportation. However, there is also another
important component which affects the amount of emissions produced by freight transportation
that has been taken partially (e.g. the case of traffic by using average speed instead of real time
traffic information) or not at all under consideration (e.g. the case of wind condition are not taken
under consideration for the estimation of environmental externalities). This component is the
characteristics of the transportation network.
The transportation network is the recipient of the emissions but at the same time its specific
characteristics are one of the sources of emissions (the other is the vehicle). The transportation
network is composed by nodes which correspond to intersections and arcs which connect two
nodes. Its arcs are characterised by many factors (from now on referred to as Transportation
Network (TN) factors) which definitely influence the fuel consumption and the emission
production if this arc is used.
The first TN factor taken under consideration in the framework of GreenYourRoute is the traffic
conditions which are taken by real time traffic information systems instead of mean speed. It is
noted that the use of mean speed distributions in emission modeling does not explicitly take into
account the effect of different driving dynamics at a particular mean speed (e.g. constant speed
versus high levels of speed fluctuation) on vehicular emissions. This affects the accuracy of
emission predictions especially for urban areas where the traffic jam is high. The second TN factor
is the infrastructure profile. Even in the case of large-scale considerations, it cannot be assumed
that – for example - extra emissions when travelling uphill are balanced by a corresponding
reduction in emissions when travelling downhill. Finally, the third TN factor, taken under
consideration, is the weather conditions. It is certain, that when, for example, a vehicle travels
against the wind the environmental externalities are higher than when there is no wind or the wind
follows the traveling direction of the vehicle. Summarizing, TN factors taken under consideration
in the framework of GreenYourRoute are (without be limited to): the traffic conditions, the
Book of Proceedings
170
infrastructure profile and the weather conditions. By traffic conditions we mean the real time traffic
conditions, by infrastructure profile we define the road gradient (uphill – downhill), the type of
road (road width, number of traffic lanes) and the traffic lights system and by weather conditions
we consider the wind, the temperature and the humidity conditions of an arc.
A screenshot of the user interface of GreenYourRoute platform is presented in the following figure.
Figure 1: GreenYourRoute platform - User interface
3. Conclusion
In this research work three prototype platforms for three regions of Greece are presented. The
next step of this work is to collect transport information for additional regions of Greece and
introduced them in the database developed in the frame of FindMyWay project. The final outcome
of this research will be the development of a journey planner for the entire Greece connecting all
cities and villages having a population greater or equal to 50.000.
Acknowledgements
The author gratefully acknowledges financial support from the European Commission under the
grant FP7-PEOPLE-2011-CIG, GreenRoute, 293753 and the Action «Supporting Postdoctoral
Researchers» of the Operational Program "Education and Lifelong Learning" (Action’s
Beneficiary: General Secretariat for Research and Technology, Greece), and is co-financed by the
European Social Fund (ESF) and the Greek State.
Book of Proceedings
171
Reducing Waiting Time at Intermediate Nodes for Intercity Bus
Transportation
Dimitropoulos Charalampos
Department of Mechanical Engineering, School of Engineering, University of Thessaly,
Leoforos Athinon, Pedion Areos, 38834 Volos, Greece, Email: [email protected]
Skordilis Erotokritos
Department of Mechanical Engineering, School of Engineering, University of Thessaly,
Leoforos Athinon, Pedion Areos, 38834 Volos, Greece, Email: [email protected]
Saharidis George K.D.
Department of Mechanical Engineering, School of Engineering, University of Thessaly,
Leoforos Athinon, Pedion Areos, 38834 Volos, Greece and Kathikas Institute of Research &
Technology, Paphos, Cyprus Email: [email protected]
Abstract Scheduling of transit networks is one of the most addressed problems in the mathematical
optimization science, due to the increase of public transportation in the last decade. Researchers
have introduced various formulations to address the problem of timetabling, using different
objectives like bus synchronization and passenger demand. In this paper, we present a mixed-
integer linear programming formulation with the objective of minimizing passenger waiting times
at transitional transfer nodes, taking also into consideration high passenger demand that occurs at
certain times.
Keywords: minimizing waiting time, public transportation, mixed integer linear programming,
transitional, nodes.
1. Introduction
The foundation of the Greek public transport bus service consists of 61 regional bus cooperatives
of individual owners of 4,175 vehicles with a public coach license. These cooperatives are called
KTEL, an abbreviation of “Kentrikon Tameion Eppagelmation Leoforion” (roughly translated into
Central Union of Bus Operators). The KTEL companies provide for about 80% of all passenger
transportation in Greece. Interregional transport is provided by most of the KTEL companies.
According to the “Study of passenger transport by coach”, undertaken in January 2009 by the
European Commission [2007], 180 million passenger journeys were conducted by KTEL only in
2002, whereas 5,710 thousand passenger kilometers were covered by KTEL buses in 2004. Despite
the existence of routes and roads, there are still several Greek major and smaller cities not directly
connected by KTEL buses, mostly because of the arrangements made between the companies
about the transportation of passengers from one city to another. As a result, passengers have to
transfer from one bus to another, in order to reach their final destination, leading to substantial
waiting times at the transfer/transfer nodes.
Book of Proceedings
172
The case mentioned above, along with the increasing use of public bus transportation, leads to
the necessity of a coordinating synchronization mechanism, the purpose of which is to minimize
the waiting time for the passengers at the transfer nodes. Increased number of passengers using
KTEL services leads to a large number of passengers waiting at these nodes, so a big issue for
KTEL companies is how to mitigate this waiting time. Unfortunately, at the moment, there is no
synchronization mechanism, although it would be for the benefit of both the passengers and the
KTEL cooperatives.
This paper presents a novel approach for timetabling, in which the center of the formulation is
the minimization of waiting times at transfer nodes of bus networks. In contrast to previous studies,
the bus routes are predetermined and the available number of buses remains unmodified, which is
the most applicable and realistic approach for KTEL companies.
2. Model formulation
One basic assumption to our modeling approach is that travelling time is assumed to be constant
and equal to the travel time of traversing a link at the posted speed limit. One could argue that
various factors (that are volatile like traffic congestions and weather conditions) could affect trip
travel times. These factors were not considered as all buses use national roads where these factors
are not as influential to travel time. The same principles apply to a number of attributes that
describe the bus network in order to make it more flexible and easier to modify. These are the
starting and ending time of the bus routes, the time a bus has to wait at a transfer-transfer node and
the time interval between successive routes. Furthermore, the number of routes between nodes is
also considered as fixed, as the purpose of the formulation is to not to minimize the existing number
of bus routes, but rather to schedule the buses on these routes. Additionally, it is assumed that
transfer/transfer nodes have sufficient capacity to accommodate the demand for bus parking. Thus,
no additional constraint regarding limited number of bus terminals needs to be considered. It is
also assumed that buses travelling have to return to their originating node (similar to the traveling
salesman problem). However, in cases of bi-directional routes, this assumption is no longer
necessary. Additional data will be used to describe some extensions to the primary formulation,
such as the passenger demand pattern.
The formulation described in this paper includes both continuous and binary variables. The
objective function is the minimization of the total waiting time for the first and second available
buses departing from all transfer nodes of the network. Furthermore, during the day, fluctuation of
passenger demand is observed with distinct peak periods. These periods are referred to as “high-
priority periods” and penalty factors are used in the objective function to account and assign higher
priority to routes with this demand. Additionally, time restrictions for departure times for a subset
of buses are considered. This is necessary in order to define the buses departure time window.
There are also certain cases where this time window needs to be defined for specific bus routes to
accommodate the schedule of the bus drivers.
Figure 1 illustrates the concept of the formulation. It presents a layout of how the waiting time
at the transfer nodes is calculated based on the arrival and departure times of all routes concerning
these nodes.
Book of Proceedings
173
Fig. 1 Illustration of the formulation.
For the mathematical formulation a number of constraints were developed. The first constraint
defines the combination of routes to and from an intermediate node as active or inactive, based on
whether the bus from the intermediate node has already left or not. In conjunction to that constraint,
the second constraint calculates the waiting time between the arrivals on each intermediate node
and the first two available departing routes from that node towards the destination node.
Furthermore, a constraint which assures that a bus departing from any node will eventually return
to this node was developed, since our assumption suggests that each bus will return to their
respective starting nodes. There is also a continuity constraint which assures that every itinerary
must begin after the previous one regarding the same destination. Finally, a constraint which
defines the time window in which any bus must depart was also implemented.
Regarding the extended objective function that takes into consideration high passenger
demand on certain time periods during a day, two additional constraints were utilized. The first
one calculates the difference between the departing time of a route and the time the high passenger
demand occurs. The second constraint defines the number of extra buses that need to be used in
order to accommodate this high passenger demand.
3. Numerical examples
This section describes a real-life example of a bus network in Greece. The example is based on the
bus network of the island of Crete. We introduced the data needed for the formulation based on
information obtained from Crete’s KTEL. Note that time values introduced both as parameters and
estimated as decision variables in the models, are expressed in minutes and based on a 24-hour
conversion (e.g. 10:00 am is translated into the 600th minute).
The intercity bus network of Crete consists of 122 nodes, representing cities and villages
accessible through the network. Out of these, 13 are regarded as transfer nodes. Among them are
the four main cities at the island (Chania, Rethymnon, Heraklion, Ag. Nikolaos). The routes
connecting the nodes of the network vary in number, from two to twenty five, depending on the
population density at each node. The formulation was applied to the full extent of this intercity bus
network.
Book of Proceedings
174
Numerical Results & Discussion
In this section, results of numerical examples are presented. These include a comparison between
the total waiting time from the existing bus schedule and the total waiting time produced by the
proposed model. Problem sizes and different values of the objective functions for all different
aforementioned model variations are also presented. Aside from the values of the objective
functions, the total waiting time for each extension is presented as the sum of the waiting times
from all nodes to all nodes and for the total number of routes.
The average waiting time is calculated by dividing the total waiting time that is obtained as the
sum of the minimized waiting times yielded from the objective function, with the total number of
active connections between routes to and from the transfer node. The resulting values of decision
variables, as well as the objective function, are all measured in minutes, accordingly with all the
parameters describing time in the network.
All tables presented below include certain cases regarding the initial objective function and the
extended one, all applied on the bus network of Crete. Case 1 represents the actual waiting times
based on the existing bus schedule and act as a benchmark for the effectiveness of the formulation.
Furthermore, case 2 corresponds to the values of the waiting times yielded by the formulation
without taking into consideration an increased passenger demand. Finally, case 3 accommodates
the results of utilizing the extension for increased passenger demand at all routes that depart from
all transfer nodes. We must again mention that cases 1, 2 and 3 correspond to the minimization of
the sum of waiting times only for the first and second available departing buses from any transfer
node.
Table 1 includes the number of constraints, variables and CPU solution times for all the cases
that will be presented next. Table 2 presents the initial values of the total and average waiting time
and the number of active routes based on the existing timetables, as well as the values extracted
from the formulation. Furthermore, a percentage of improvement derived from the simulation
compared to the existing timetables is also included.
Table 1 Problem’s Size
Decision
Variables
Constraints CPU Solution Time
(seconds)
Gap (%)
Case 1 171206 5573193 0.05 0
Case 2 104440 194267 66117.2 83
Case 3 112508 200629 1774.8 85
Table 2 Numerical Results
Total Waiting
Time
(%)* Total Number of
Active Connections
( ∑ 𝑌𝑖,𝑘,𝑗𝑚,𝑛 )
(%)* Average Waiting
Time
(%)*
Case 1 1994385 - 33503 - 59 -
Case 2 480890 75.9 16769 50.0 28 52.4
Case 3 795840 60.0 24717 26.2 32 45.7
*Percentage of improvement over original bus timetables.
Book of Proceedings
175
In our attempt to approach the existing bus network more accurately, the time window during
which the routes occur was divided into five time zones, each one representing: early morning,
late morning, noon, afternoon and night respectively. Each one of them can be represented by a
specific time of the day around which all existing departing routes are focused. It should be noted
that a typical timetable covers the time window from 5:30 am to 11:00 pm, or 330 to 1380 (in
minutes). The first time zone is between the 330th and the 540th minute of the day, the second is
between the 540th and the 750th minute of the day, the third is between the 750th and the 960th
minute of the day, the forth is between the 960th and 1170th minute of the day and the last one is
between the 1170th and the 1380th minute of the day. For these time zones, table 3 and figure 2
present the number of departing routes that occur during these zones, both for the formulation and
the existing bus schedule. For case 3, we introduced five high passenger demand periods, each one
corresponding to one of representing the five time zones, and we assumed it occurs in the middle
of each zone (i.e. for time zone between 330th and 540th minute, the high demand period occurs at
435th minute). The penalty factor values were deduced from the number of departing routes as
distributed for time zones 1-5 for the existing bus schedule. This will lead to a timetable closer to
original one, although the total waiting time will be increased as a result.
Table 3 Distribution of departing routes
Time zone 1 Time zone 2 Time zone 3
Time zone 4 Time zone 5
Case 1 335 194 339 145 77 Case 2 192 340 246 204 108 Case 3 257 239 323 210 61
Fig. 2 Graph representing cases 1,2,3
It can be deduced from cases 2 and 3, that there is a significant decrease of the waiting
times at all transfer nodes of the bus network, compared to the existing bus schedule (case 1).
Specifically, comparing cases 1 and 2, there is a 75.9% decrease in the average waiting time,
despite the fact that the active number of connections (𝑌𝑖,𝑘,𝑗𝑚,𝑛
) is significantly smaller. Case 3
produced a time table that minimizes the total waiting time, without deviating too much from the
existing time table, since it takes into consideration only the first and second available departing
routes from every intermediate node, leading to more active connections for the passengers to use.
0
50
100
150
200
250
300
350
400
Timezone1
Timezone2
Timezone3
Timezone4
Timezone5
Case 1 Case 2 Case 3
Book of Proceedings
176
Although the total waiting time is increased compared to case 2, it remains smaller compared to
case 1 by 26.2%.
In both cases of objective functions, it is prudent that case 3 yielded increased waiting times
compared to case 2, but greatly decreased compared to the existing bus schedule (cases 1).
Meanwhile, the increased number of active connections in case 3 make the results more suitable
for realistic examples, where the passengers need to have more available options concerning their
departing from a transfer node.
4. Conclusions
This paper has addressed the problem of minimum waiting times when creating bus timetables.
A mathematical formulation was proposed to develop a more desirable and passenger-friendly
transit system. This was achieved by minimizing the waiting time spent by passengers at transfer
nodes on the bus network. This approach was tested by utilizing data obtained from the bus
network of the Greek island of Crete. By comparing existing timetables to the ones extracted from
the proposed model, a definitive improvement was observed. This improvement remained despite
our effort to remain closer to the existing timetables, through the use of the various extensions that
were introduced.
References
1. Steer Davies Gleave. Study of passenger transport by coach. Publication TREN/E1/409-2007.
European Commission, 2009.
2. Ceder, A, Golany B, Tal O. (2001) Creating Bus Timetables with Maximal Synchronization.
Transportation Research Part A. 35:913-928.
3. Eranki, A. “A model to create bus timetables to attain maximum synchronization considering
waiting times at transfer stops”. Master’s thesis. Department of Industrial and Management
Systems Engineering, University of South Florida, 2004.
4. Ibarra-Rojas, O. J., and Y. A. Rios-Solis. “Synchronization of bus timetabling”.
Transportation Research Part B, Vol. 46, 2012, pp. 599-614.
5. Hall R., Dessouky M. and Lu Q. “Optimal Holding Times at Transfer Stations”. Computer and
Industrial Engineering. Vol. 40, 2001, pp. 379-397.
6. Bussieck M. R., Winter T., and Zimmermann U. T. “Discrete Optimization in public rail
transport”. Mathematical Programming. Vol. 79, Issue 1-3, 1997, pp. 415-444.
7. Goverde R. M. P. “Synchronization Control of Scheduled Train Services to Minimize
Passenger Waiting Time”. Transport, Infrastructure and Logistics, Proceedings 4th TRAIL
Congress, 1998.
8. Chen D. and Wu K. “Research on Optimization Model and Algorithm of Initial Schedule of
Intercity Passenger Trains based on Fuzzy Sets”. Journal of Software, Vol. 7, No. 1, 2012, pp.
49-54.
9. Reinhardt L. B., Clausen T., and Pisinger D. “Synchronized dial-a-ride transportation of
disabled passengers at airports”. European Journal of Operational Research. Vol. 225, 2013,
pp. 106-117.
10. Wong R. C. W., Yuen T. W. Y., Fung K. W. and Leung J. M. Y. “Optimizing Timetable
Synchronization for Rail Mass Transit”, Transportation Science, Vol. 42, No. 1, 2008, pp. 57-
69.
Book of Proceedings
177
Innovation management strategies for organizational performance
Dimitrios Mitroulis
School of Informatics, Department of Applied Informatics, University of Macedonia
P.O. Box 1591, GR54006 Thessaloniki, Greece
E-mail: [email protected]
Fotios Kitsios
School of Informatics, Department of Applied Informatics, University of Macedonia
P.O. Box 1591, GR54006 Thessaloniki, Greece
E-mail: [email protected]
Abstract
Innovation has always been a critical factor for every kind of entrepreneurial achievement and
performance. However, most of the organizations, which were supposed to innovate, have focused
on better short term efficiency, ignoring the chance of getting competitive advantage over their
competitors. The organization’s relationship with either customers or competitors could improve
its knowledge over the market conditions and gain market-oriented information. The successful
management of the market-oriented inflow and organizational innovation leads to the
improvement of the organizational performance. The purpose of this paper is to collocate a main
framework which a business or industry could use in order to identify whether its organizational
innovation could be the joint between market-orientation and organizational performance. The
whole research focalizes on the questions that could unveil the organizational performance, by
evaluating its innovation capabilities. Other studies have shown the importance of innovation in
today’s organizations, giving emphasis on market-orientation, better efficiency, innovativeness
and organizational performance. Therefore, in order to evaluate and examine the theoretical
assumptions, a questionnaire, addressed to Greek SMEs, is cited. It is used to examine and evaluate
capabilities, operations and competitive advantages which could lead SMEs to organizational
performance. All variables are measured in a 5point Likert type scale. The results of this study are
examined with multicriteria methods.
Keywords: Market-orientation, Innovation, Performance, Dynamic Hybrid Strategy
Book of Proceedings
178
1. Introduction
Globalization, economic crisis, technological changes, lack of opportunities threaten business
sustainability. In fact, they try to follow the unceasing ongoing changes so that they could gain
stability. Customers seek the best, possible, combination of quality and price for products and
services that are offered. Competition has reached the highest limits and the organizations are now
focused to find a solution to this questionings. How are we supposed to gain competitive
advantage? How are we going to seize our opportunities? How could we differentiate from the
others? Is innovation the key to success? How could we have a better performance? In order to
gain a sustainable competitive advantage, firms are supposed to follow a differentiation strategy.
Additionally, market orientation is the factor that connects the organization with the market
preferences (Medina & Rufın, 2008). However, this is not the only factor that affects the
organizations. Many researchers agree that innovation is the mediator for better organizational
performance. They, moreover, realize that innovation is the link between market orientation and
organizational performance (Jin K. Han et al., 1998). How is it possible to identify and understand
this kind of organizational assets?
Motivated by the previously written aspects, this study tries to identify the main perspectives of
reclaiming advantage. Strategy drivers, market orientation, innovation and organizational
performance are the four factors that are investigated and analyzed. The rest of the paper presents
the conceptual framework (Section 2), methodology, data and findings (Section 3) and finally,
conclusion (Section 4).
2. Conceptual framework
2.1. Market Orientation: Jin K. Han et al., (1998) identify market orientation as a part of the
organizational culture. The three basic forces: i) customers’ behavior and the ability to foresee
their needs, ii) competitors’ behavior and the ability to compete them in capabilities and
technology, and iii) the interfunctional coordination. Market-driven business is always ready to
anticipate the continuously changing needs of its customers and respond to them through
innovation (Salavou et al, 2004). A market oriented culture could improve the organizational
intelligence and agility by creating an interactive relationship ( Fátima Evaneide Barbosa de
Almeida et al, 2013). Customer satisfaction should be considered as a part of the market orientation
concept (Avlonitis & Gounaris, 1999).
2.2. Dynamic Hybrid Strategy: Strategy is the main framework of the organization. Having in mind
all the changes in their environment, firms set hybrid strategies which give emphasis on
differentiation, flexibility and adaptability (Pertusa-Ortega, 2008). Combining the strategy with
market orientation, we could be led to a dynamic or behavioral hybrid strategy (Avlonitis &
Gounaris, 1999). Hybrid competitive strategies are related to better organizational performance
(Pertusa-Ortega, 2008). Therefore, a market oriented culture along side with dynamic hybrid
strategy offer a stable base for higher rates of performance (Avlonitis & Gounaris, 1999; Fátima
Evaneide Barbosa de Almeida et al., 2013). Relating the above with innovation strategy, Dynamic
Hybrid Strategy appears. In fact, it includes components of innovation strategy, competitive
strategy and differentiation strategy. Hybrid strategies might be more sensitive on customer needs.
Book of Proceedings
179
The difficulty of gaining and using business intelligence is the main problem in this relationship.
The pursuit of hybrid competitive strategies may help obtaining several sources of advantage and,
therefore, make it possible to put over higher performance levels (Pertusa-Ortega, 2008).
H1: There is a positive relationship between Market Orientation and Dynamic Hybrid Strategy.
H2: There is a positive relationship between Dynamic Hybrid Strategy and Business Performance.
2.3. Innovation Management: Organizational innovation is influenced by the market orientation.
Jimenez et al, (2008) mention that innovation is the organization’s means of response to the
market. Additionally, innovation has to be developed and executed as an internal part of the
business strategy (Gerald & Emamisaleh, 2014). In addition, they support the importance of
innovation to the organizational performance, indicating that it should be followed by the
appropriate action plan. Creating innovation capabilities and giving emphasis to every innovation
type (organizational, administrative, process, production, technical and marketing innovation) is
going to make a successful introduction of competitive advantage. Competition is the force of
innovation (Salavou et al., 2004). Innovation management is the means of creating innovation
capabilities. (Salavou et al, 2004; Han et al, 1998; Avlonitis & Gounaris, 1997; Jimenez et al,
2008). It reveals the firm’s ability to obtain advantage of the appropriate innovation management
inside the business terms.
H3: There is a positive relationship between Innovation Management and Dynamic Hybrid
Strategy.
2.4. Organizational Performance: Ferraresi et al., (2012) support that organizational performance
could be counted with both financial and non-financial criteria. These are market share, sales of
new products, the rates of return on investment and evaluation of internal factors such as
operational improvements and reducing the time of response against the changes imposed by the
market. This is more or less the main idea of organizational performance that dominates the
literature. Additionally, in this study the relationship between innovation performance and
organizational performance is tested, so that it would be clear whether it is worth to invest on an
innovation which is successful for the firm and its effect on performance.
H4: There is a positive relationship between Innovation Performance and Business performance.
Book of Proceedings
180
Figure 1: Conceptual model
3. Methodology, Data and Findings
This survey has started in Greece, in order to test the hypotheses of this study. Greek SME’s have
been asked to participate in the survey. It was designed to cover a wide range of industries.
However, it is limited to the companies that create or offer IT services. This is an ongoing research
which until now does not have enough evidence to prove its hypotheses and the wanted results. A
questionnaire has been developed following the 5-point Likerty-type scale. The variables and the
measurement items were adapted and combined from existing scales in the literature. The
questionnaire has been examined by a group of experts. It is addressed to general managers or
marketing directors. Additionally, further information was collected via personal interviews with
the general manager or the marketing director of each company, using a structured questionnaire.
All questionnaires were emailed to the companies.
The structural equations model (SEM) is going to be used in order to test the validity and the
relationship among the variables. Until now, 20 questionnaires have been given and 10 of them
have been obtained, yielding a response rate of 50 percent.
4. Conclusion
The harmonious existence of innovation strategy within the general strategy of the organization
could provide a stable base for more successful business strategy, operations and higher levels of
performance (Wong, 2013). In addition, the influx from market orientation could yield a positive
impact for gaining a sustainable competitive advantage and offering added value to the business.
Considering innovation as a link between customer satisfaction and organizational performance,
it is easily understood that strategically, organizations could be able to gain intelligence and
weaponize its strategy, creating dynamic hybrid strategy, adjusted to its needs, resources and
capabilities. Performance is a fundamental factor for a successful organization. In other words,
every innovative organization acquires competitive advantage against its competitors. If an
organization succeeds in innovation management, producing innovation capabilities and
competitive advantage, it would be able to achieve the superior organizational performance.
Book of Proceedings
181
References
Alex A. Ferraresi, Carlos O. Quandt, Silvio A. dos Santos and Jose´ R. Frega. (2012). "Knowledge
management and strategic orientation: leveraging innovativeness and performance". Journal of
Knowledge Management. 16 (5), p688-701.
Affendy Abu Hassim , Asmat-Nizam, Abdul-Talib , Abdul Rahim Abu Bakar. (2011). "The Effects of
Entrepreneurial Orientation on Firm Organisational Innovation and Market Orientation Towards Firm
Business Performance". International Conference on Sociality and Economics Development. 10 (2011),
pp 280-284.
Cayetano Medina and Ramo´n Rufı´n. (2009). "The mediating effect of innovation in the relationship
between retailers’ strategic orientations and performance". International Journal of Retail &
Distribution Management. 37 (7), p629-655.
Daniel Jimenez-Jimenez, Raquel Sanz Valle, Miguel Hernandez-Espallardo. (2008). "Fostering
innovation, the role of market orientation and organizational learning". European Journal of Innovation
Management. 11 (3), pp 389-412.
Eva M. Pertusa-Ortega. (2008). "Hybrid Competitive Strategies, Organizational Structure and Firm
Performance".
George J. Avlonitis and Spiros P. Gounaris. (1999). "Marketing orientation and its determinants: an
empirical analysis". European Joumal of Marketing. 33 (11/12), p1003-1037.
Gloria L. Ge and Daniel Z. Ding. (2005). "Market Orientation, Competitive Strategy and Firm
Performance: An Empirical Study of Chinese Firms.Journal of Global Marketing".18 (3/4), p115-142.
Gurhan Gunday, Gunduz Ulusoy, Kemal Kilic, Lutfihak Alpkan. (2011). "Effects of innovation types on
firm performance". International Journal of Production Economics. 133 (2), p662-676.
H. Salavou, G. Baltas and S. Lioukas. (2004). "Organisational innovation in SMEs: The importance of
strategic orientation and competitive structure". European Journal of Marketing.38 (9/10). p1091-1112.
Jin K. Han, Namwoon Kin and Rajendra K. Srivastva (1996). "Market orientation and organizational
performance; Is innovation the missing link".
Lokman Mia and Lanita Winata. (2014). "Manufacturing strategy and organisational performance The
role of competition and MAS information". Journal of Accounting & Organizational Change. 10 (1),
p83-115.
Masood Ul Hassan and Sadia Shaukat. (2013). "Effects of Innovation Types on Firm Performance: an
Empirical Study on Pakistan’s Manufacturing Sector". Pakistan Journal of Commerce and Social
Sciences. 7 (2), p243-262.
Parvaneh Gelard and Korosh Emamisaleh. (2014). "The Evolution of Innovation Types Towards
Production Performance". International Business Management. 8 (4), p222-228.
Stanley Kam Sing Wong. (2013). "The role of Management involvement in innovation". Management
Decision. 51 (4), pp709-729.
Book of Proceedings
182
Fátima Evaneide Barbosa de Almeida, João Veríssimo Lisboa, Mário Gomes Augusto, Paulo César de
Sousa Batista. (2013). "Organizational Capabilities, Strategic Orientation, Strategy Formulation Quality,
Strategy Implementation and Organizational Performance in Brazilian Textile Industries".
Book of Proceedings
183
Country risk evaluation methodology to support bilateral cooperation in the
field of electricity generation from renewable sources
Aikaterini Papapostolou
National Technical University of Athens
9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece, email: [email protected]
Charikleia Karakosta
National Technical University of Athens
9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece
Vangelis Marinakis
National Technical University of Athens
9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece
John Psarras
National Technical University of Athens
9, Iroon Polytechniou str., 15780, Zografou, Athens, Greece
Abstract
Renewable energy sources (RES) cooperation within the European Union, as well as with EU
neighboring countries is high on Europe’s political agenda. According to the EU Directive
2009/28/EC, one or more Member States could cooperate with one or more developing countries
in joint projects, regarding the generation of electricity from renewable sources. This paper
outlines a multicriteria methodology to evaluate country opportunities and risks for the successful
implementation of the cooperation mechanisms with third countries. The proposed evaluation
criteria are built on three points of view: (1) investment framework/country risk profile, (2) social,
and (3) energy security. The overall evaluation of countries is obtained through a multicriteria
additive value model, which is assessed using an ordinal regression approach. Five countries of
North Africa are evaluated and ranked considering the latest criteria data.
Keywords: Country risk; Renewable energy; multiple criteria; robust ordinal regression.
Book of Proceedings
184
1. Introduction
Renewable Energy Sources (RES) cooperation within the European Union (EU), as well as with
EU neighboring countries is high on Europe’s political agenda. EU Renewable Energy Directive
(DIRECTIVE 2009/28/EC) allows Member States to achieve their 2020 Renewable Energy (RE)
targets by implementing joint projects in neighboring countries and thus to import electricity from
RE sources from them in order to meet domestic demand in Member States.
In literature, several studies exist that try to identify the perception of risk in RES investments for
the North Africa countries. Hawila et al. (2014) apply a consistent methodology (Weighted Sum
Method (WSM)) across all the North African countries to assess the present state of infrastructure,
institutions and human capital factors to adopt and deploy RE technologies. Komendantova et al.
(2012; 2011) address the perception of risks in RES projects, which is considered as regulatory,
political, and force majeure. Corruption and inefficient and unpredictable bureaucracies are
matters of great concern for the project developers in North Africa. In the case of Morocco and
Egypt, political instability and regulatory barriers are issues that discourage the development of
clean development mechanism (CDM) projects, however well-organized mechanisms have
already been established to support RES development in both countries so far (Karakosta et al.,
2013; Karakosta & Psarras, 2013).
An important number of Multiple Criteria Decision Making (MCDM) methods have been used so
far in order to evaluate the feasibility of RES projects, such as the Multi-Attribute Utility, the
ELECTRE and the PROMETHEE methods, the Analytical Hierarchy Process and the TOPSIS
(Cavallaro & Ciraolo, 2005; Doukas et al., 2009; 2010; Haurant et al., 2011; Pohekar &
Ramachandran, 2004; Rosso et al., 2011). To the best of our knowledge, there are only very few
studies using UTASTAR method for RE and energy sector problems.
The aim of this study is to evaluate country risk in order to support bilateral cooperation in the
field of electricity generation from RES. To this end, a multicriteria decision support methodology
has been developed taking into account three evaluation points of view, the investment
framework/country risk profile, the social, and the energy security point of view. An additive value
model has been constructed and the UTASTAR disaggregation method has been implemented to
infer the criteria weights. The obtained ranking of alternatives has been subjected to robustness
analysis and finally the proposed methodology has been applied to North Africa countries.
Apart from this introductory section, the paper is organized as follows: the second section
elaborates the multicriteria evaluation system obtained as regards the evaluation points of view
and the criteria proposed. The third section introduces the methodological framework and in
section four the implementation of the proposed methodology is presented. Finally, the fifth
section summarizes the main points of the study and presents the future perspectives.
Book of Proceedings
185
2. Designing a multicriteria evaluation system
In order to assess the opportunities and barriers for RE development in third countries, the current
situation in the analyzed countries has to be assessed as it is likely to condition the future RES
deployment path. In this framework, a multicriteria evaluation model is proposed. It comprises
nine evaluation criteria based on three evaluation points of view to support decisions for the
implementation of RES investments (Figure 1): (i) investment framework/country risk profile, (ii)
social, and (iii) energy security.
Figure 16: Evaluation points of view and criteria
This framework will determine which countries are more suitable or likely to attract a larger share
of foreign investment in RE. The definition of criteria is presented below while the scales used to
evaluate countries are given in Table 1.
OECD country risk rating (g1): This criterion is an OECD (Organization for Economic Co-
operation and Development) country risk classification index which classifies countries into one
of eight categories (0-7) through the aggregation of two dimensions: A quantitative assessment of
country credit risk produced by Country Risk Assessment Model (CRAM) and a qualitative and
subjective assessment of the CRAM results made by country risk experts from OECD members,
integrating political risk and/or other risk factors.
Ease of doing business rank (g2): The ease of doing business rank has been developed by the World
Bank and ranks economies from 1 to 189. A high ranking on this index indicates that the regulatory
environment is more conducive to the operation of a local firm.
Global competitiveness index (g3): This index, developed by the World Economic Forum, captures
the competitiveness conditions of 148 economies, integrating environmental and social
sustainability issues. Global competitiveness index includes a weighted average of many different
components, each measuring a different aspect of competitiveness.
Book of Proceedings
186
Social Hotspot Database index for the energy sector (g4): This index assesses the existing social
risk (social hotspots) as regards a variety of sectors in terms of: community infrastructure, labor
rights and decent work, health and safety, human rights and governance. In this paper the sector
that has been examined is the electricity sector.
Energy Development Index (g5): The Energy Development Index (EDI) is a multidimensional
index built by the International Energy Agency (IEA) that includes the principal indices related to
energy development (access to electricity, access to clean energy cooking facilities, access to
energy for public services, access to energy for productive use).
Rate of electricity transmission and distribution losses (g6): This indicator captures the quality of
electrical infrastructures and networks and includes losses in transmission between sources of
supply and points of distribution and in the distribution to consumers. The indicator is developed
by the World Energy Council.
Energy demand growth (g7): This indicator captures the stress of a system (understood as its
capacity to meet the energy demand). The higher the annual energy demand growth, the greater
the risk that the existing supply and new energy infrastructure investments will not be able to
provide the required energy supply.
Age of technology fleet (g8): This is a qualitative indicator from 1 to 5 intending to capture the
need to replace the existing infrastructure (as the older the fleet is, the higher the technical failure
risk).
Share of fossil fuels in electricity production (g9): This indicator intends to capture the dependency
of the country on fossil fuels (oil, coal, natural gas) in the electricity production.
Criterion Worst Level Best Level Source
g1 7 0 OECD, 2014
g2 189 1 World Bank, 2013
g3 1 7 World Economic Forum, 2013
g4 500 0 Social Hotspot Database, 2014
g5 0 1 International Energy Agency, 2011
g6 20% 0% World Energy Council, 2011
g7 30% 0% BETTER project report, 2013
g8 1 5 BETTER project report, 2014
g9 100% 50% Trading Economics, 2011
Table 17 Criteria evaluation scales
Book of Proceedings
187
3. Methodological Framework
The aim of the proposed methodology is to assess an overall evaluation model that aggregates the
nine evaluation criteria. The multicriteria preference model is assumed to be an additive value
function of the following form:
n
i
iii gupgu1
(1)
subject to the following normalization constraints:
1* ii gu
(2) 0ii gu ni ,...,2,1
11
n
i
ip (3)
0ip ni ,...,2,1
(4)
where g = (g1, g2,… gn) is the performance vector of a country on the n criteria, gi* and gi* are the
least and most preferable levels of the criterion gi, respectively, ui, i = 1,2,…,n, are non-decreasing
real valued functions, named marginal value functions, which are normalized between 0 and 1,
and pi is the weight of ui. Thus, for a given country a, g(a) and u[g(a)] represent the multicriteria
vector of performances and the global value of country a respectively.
Due to objective difficulties to convince decision makers (DM) in externalizing trade-offs between
heterogeneous criteria, it is usually preferred by the analysts to infer the additive value functions
from global preference structures, by applying disaggregation or ordinal regression methods
(Greco et al., 2008; 2010; Jacquet-Lagrèze & Siskos, 1982; 2001). In this study the method used
is the UTASTAR method (Siskos & Yannacopoulos, 1985).
4. Implementation of the model
In this section the implementation of the UTASTAR method is presented as well as the application
of the model obtained to North Africa countries.
4.1 Implementation of the UTASTAR method and robustness analysis
According to UTASTAR ordinal regression method, a ranking of reference real or fictitious
countries is required to infer an additive model that is compatible with the ranking. A set of 17
virtual countries that differ on two or three criteria is then constructed in order to be easier for the
decision maker (DM) to compare them. The DM makes pairwise comparisons and inserts
Book of Proceedings
188
progressively each alternative country into a global preference ranking. The constructed reference
set, and the DM’s given ranking are presented in Table 2.
Reference
country
g1 g
2 g
3 g
4 g
5 g
6 g
7 g
8 g
9 DM’s
Ranking
Global
Values
A 6 80 6 200 0.6 10 10 3 75 6 0.516
B 3 120 7 200 0.6 10 10 3 75 1 0.726
C 1 150 2 200 0.6 10 10 3 75 5 0.528
D 4 100 5 100 0.4 8 10 3 75 7 0.505
E 4 100 5 200 0.5 15 10 3 75 15 0.409
F 4 100 5 200 0.8 18 10 3 75 12 0.445
G 4 100 5 200 0.6 10 5 1 75 10 0.467
H 4 100 5 200 0.6 10 3 1 100 17 0.349
I 4 100 5 200 0.6 10 5 4 90 3 0.552
J 4 100 5 200 0.6 10 10 5 100 13 0.433
K 7 100 5 100 0.6 10 6 3 75 11 0.457
L 4 50 5 200 0.2 12 10 3 75 9 0.481
M 4 50 5 100 0.2 12 12 2 75 14 0.421
N 4 50 5 300 0.2 12 12 4 70 4 0.540
O 2 50 6 300 0.3 15 12 2 80 8 0.492
P 2 150 6 300 0.1 15 5 2 70 2 0.568
Q 5 50 2 100 0.3 5 12 4 80 16 0.397
Table 2 Multicriteria evaluation of the 17 fictitious countries
The application process of UTASTAR method is terminated when a full compatibility between
the additive value model and the DM’s ranking is achieved. Because of the infinity of such
compatible value functions the most representative value function is obtained (barycenter value
function, see Hurson & Siskos, 2014). The last column in Table 2 shows the global value of each
fictitious country that is compatible with the DM’s ranking. The obtained barycenter value
function is defined by the following equation (5) and the average marginal value functions of
Figure 2 (in yellow):
u(g) = 0.231u1(g1) + 0.015u2(g2) + 0.184u3(g3) + 0.036u5(g5) + 0.096u6(g6) + 0.123u7(g7) +
0.170u8(g8) + 0.145u9(g9) (5)
Book of Proceedings
189
In order to test the existence of multiple or near compatible additive value models, robustness
analysis is conducted (see Hurson and Siskos, 2014). Figure 2 and Figure 3 present graphically the
minimum, maximum and most representative marginal value function and weights of each
criterion respectively, revealing the variation of these functions. The chart shows considerable
robustness with the exception of the first and third criterion.
Figure 2 The assessed marginal value functions of the criteria
Figure 3 The assessed weights of the criteria
In order to examine further the stability of the proposed model the ASI robustness index
(Grigoroudis and Siskos, 2010) is computed. The ASI is defined as the mean normalized standard
deviation of the estimated weights of the criteria and is calculated in the case of UTASTAR as:
n
i
m
j
m
j ijij
mn
m
ppm
nASI
1
1 1
22
1
11 (6)
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
1 48 95 142 189
max
min
average
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
1 3 5 7
max
min
average
g3
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
1 3 5 7
max
min
average
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
0 1 2 3 4 5 6 7
max
min
average
g2
g5
g1
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
0 5 10 15 20
max
min
average
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
0 10 20 30
max
min
average
g7
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
1 2 3 4 5
max
min
average
0
0,1
0,2
0,3
0,4
0,5
0,6
0,7
0,8
0,9
1
50 75 100
max
min
average
g6
g8 g9
0
0,05
0,1
0,15
0,2
0,25
0,3
pi
min
avg
max
g6 g7 g8 g9 g1 g2 g3 g4 g5
Book of Proceedings
190
where m is the number of weighting instances of the system, n the number of the criteria and pij
the weight of the i-th criterion for the j-th instance. The ASI is equal to 0.998, a value indicating
that the nine criteria’s weights are adequately stable.
4.2 Application to North Africa countries
Taking into account the assessed most representative value model five North Africa countries are
evaluated. The score of the countries on the nine criteria, the global value and the final ranking
obtained are presented in Table 3.
Country g1 g2 g3 g4 g5 g6 g7 g8 g9 Global
Value Ranking
Morocco 3 87 4.11 175.85 0.532 6.2 6.2 4 89.70 0.578 1
Tunisia 4 51 4.06 183.23 0.490 12.3 3.2 2 98.99 0.514 2
Egypt 6 128 3.63 216.96 0.668 11.0 6.2 3 90.49 0.414 3
Algeria 3 153 3.79 215.88 0.706 19.5 12 4 99.02 0.386 4
Libya 7 187 3.73 234.41 0.923 15.9 9 3 100.00 0.200 5
Table 3 Evaluation of five North Africa countries
5. Conclusions
This paper evaluates the opportunities and risks for the implementation of cooperation mechanisms
in the field of RES electricity production based on a multicriteria evaluation model. The model
obtained is fully compatible with the DM’s ranking on a set of fictitious countries and enables each
individual to assess the current situation of a country as regards the investment, social and energy
security profile of it. After the implementation of the model to North Africa countries, the obtained
ranking shows that Morocco is considered to have the most suitable conditions among the five
North Africa countries for successful implementation of RES projects.
Considering the above, a decision support system would be a useful tool for the implementation
of the proposed methodology supporting potential users/ DMs in evaluating/ benchmarking energy
country risk. Further research could also address the integration of new multicriteria robustness
analysis techniques to take into account the ranking variation due to lack of sufficient preference
information.
Furthermore, it would be valuable the development of a formal protocol and adequate techniques
for the construction of reference set of countries, in order to infer the weights of the criteria.
Finally, the study would be enhanced by applying the obtained model to other regions worldwide.
Acknowledgement
The current paper was primarily based on the research conducted within the framework of the
project “BETTER - Bringing Europe and Third countries closer together through renewable
Book of Proceedings
191
Energies” (project number: IEE/11/845/SI2.616378), supported by the Intelligent Energy Europe
program.
References
BETTER, 2013. D2.5: Indicators and Methodologies to Assess Key Issues for the Implementation of the
Cooperation Mechanisms. Deliverable of the BETTER project (IEE/11/845/SI2.616378).
BETTER, 2013. D3.2.1: Demand Development Scenarios for North Africa. Deliverable of the BETTER
project (IEE/11/845/SI2.616378).
BETTER, 2014. D3.1: Inventory of RES-E in North Africa. Part A: Power Plant and Grid Inventory in
Morocco, Algeria, Tunisia, Libya and Egypt. Deliverable of the BETTER project
(IEE/11/845/SI2.616378).
Cavallaro F., Ciraolo L., 2005. A multicriteria approach to evaluate wind energy plants on an Italian island,
Energy Policy, Volume 33, Issue 2, Pages 235-244.
Coplin W., O'Leary M., 1983. Introduction to political risk analysis, Policy Studies Associates, New York,
USA.
Doukas H., Karakosta C., Psarras J., 2009. RES technology transfer within the new climate regime: a
“helicopter” view under the CDM, Renewable and Sustainable Energy Reviews, 13 (5), pp. 1138–1143.
Doukas H., Karakosta C., Psarras J., 2009. A Linguistic TOPSIS Model to Evaluate the Sustainability of
Renewable Energy Options. Energy Efficiency, Environmental Performance and Sustainability – The
International Journal of Global Energy Issues (IJGEI), 32(1/2): 102-118.
Doukas H., Karakosta C., Psarras J., 2010. Computing with Words to Assess the Sustainability of
Renewable Energy Options. Expert Systems With Applications (ESWA) 37(7): 5491-5497.
EC - European Commission, 2009. Directive 2009/28/EC of the European Parliament and of the Council
of 23 April 2009 on the promotion of the use of energy from renewable sources and amending and
subsequently repealing Directives 2001/77/EC and 2003/30/EC. Official Journal of the European Union
Belgium.
Greco S., Mousseau V., Słowiński R., 2008. Ordinal regression revisited: Multiple criteria ranking using a
set of additive value functions, European Journal of Operational Research, 191 (2), pp. 416-436.
Greco S., Słowiński R., Figueira J., Mousseau V., 2010. Robust ordinal regression, in: M. Ehrgott, S.
Greco, and J. Figueira (eds.), Trends in multiple criteria decision analysis, Springer, Berlin.
Grigoroudis E., Siskos Y., 2010. Customer satisfaction evaluation, Springer, New York, USA.
Book of Proceedings
192
Haurant P., Oberti P., Muselli M., 2011. Multicriteria selection aiding related to photovoltaic plants on
farming fields on Corsica island: A real case study using the ELECTRE outranking framework, Energy
Policy, Volume 39, Issue 2, Pages 676-688.
Hawila D., Mondal M.A.H., Kennedy S., Mezher T., 2014. Renewable energy readiness assessment for
North African countries, Renewable and Sustainable Energy Reviews, Volume 33, Pages 128-140
Hurson Ch., Siskos Y., 2014. A synergy of multicriteria techniques to assess additive value models,
European Journal of Operational Research, Volume 238, Issue 2, Pages 540–551.
Jacquet-Lagrèze E. and Siskos J., 1982. Assessing a set of additive utility functions for multicriteria
decision making, European Journal of Operational Research, 10 (2): 151-164.
Jacquet-Lagrèze E. and Siskos J., 2001. Preference disaggregation: 20 years of MCDA experience,
European Journal of Operational Research, 130: 233-245.
Karakosta C., Marinakis V., Psarras J., 2013. RES Cooperation Opportunities between EU and MENA
Countries: The Case of Morocco. Energy Strategy Reviews, 2(1):92-99.
Karakosta C., Psarras J., 2013. Understanding CDM Potential in the Mediterranean Basin: A Country
Assessment of Egypt and Morocco. Energy Policy, 60: 827-839.
Komendantova N., Patt A., Barras L., Battaglini A., 2012. Perception of risks in renewable energy projects:
the case of concentrated solar power in North Africa, Energy Policy, 40, pp. 103-109.
Komendantova N., Patt A., Williges K., 2011. Solar power investment in North Africa: Reducing perceived
risks Review Article, Renewable and Sustainable Energy Reviews, Vol. 15, Issue 9, Pages 4829-4835.
Pohekar S.D., Ramachandran M., 2004. Application of multi-criteria decision making to sustainable energy
planning – A review, Renewable and Sustainable Energy Reviews, 8 (4), pp. 365–381.
Rosso M., Bottero M., Pomarico S., La Ferlita S., Comino E., 2014. Integrating multicriteria evaluation and
stakeholders analysis for assessing hydropower projects, Energy Policy, Volume 67, pages 870-881.
Siskos J. and Yannakopoulos D., 1985. UTASTAR, and ordinal regression method for building additive
value functions, Investigacao Operacional, Vol. 5, No. 1, pp. 39-53.
Book of Proceedings
193
Sustainable Food Security: A System Dynamics Decision-Making
Methodology
Keramydas Ch.
Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus
P.O. Box 461, 54124 Thessaloniki, Greece, email: [email protected]
Tsolakis N.
Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus
P.O. Box 461, 54124 Thessaloniki, Greece.
Vlachos D.
Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus
P.O. Box 461, 54124 Thessaloniki, Greece.
Iakovou E.
Department of Mechanical Engineering, Aristotle University of Thessaloniki, University Campus
P.O. Box 461, 54124 Thessaloniki, Greece.
Abstract
In this research we first discuss the role of Small Farms (SFs) in enhancing food security focusing
in developed countries. We then present a more generic framework for Small Farms’ policy-
making, which embraces sustainability and food security aspects. A System Dynamics
methodology is employed that captures the effect of regulatory interventions on the diffusion of
SFs’ products by consumers.
Keywords: Small Farms, Food Security, Sustainability, System Dynamics, Agrifood Supply
Chains
1. Introduction
A major concern of modern world relates to the sustainability of food systems (UN, 2013). This
concern stems from the projections that indicate a global population growth to 9.1 billion in 2050
with a corresponding increase in food demand by 70% (FAO, 2009). Nevertheless, the challenge
that the agricultural sector faces is not the capability to increase food production capacity by 70%
within the forthcoming 40 years, but to make 70% more food available for households (FAO,
2009) and to meet the diverse eating and dietary habits in the developed world (Lin et al., 2014).
Despite the fact that agrifood production and distribution systems have been extensively
investigated thus far, major problems still exist. For example, estimations suggest that nearly 870
million people around the world suffer from undernourishment or chronic hunger (FAO, 2013),
about 2 billion people suffer from micronutrient insufficiency or hidden hunger (FAO, 2012) and
Book of Proceedings
194
approximately 500 million people suffer from obesity (WHO, 2011). Hence, such records have
positioned food security to the top of the public agenda not only in developing countries, but also
in developed countries. In addition, the ignorance of specific food related incidents may be
detrimental for the industrialized nations. Indicatively, incidents like the: (i) sudden and sharp
increase in world food prices in 2007–08 due to climate change, (ii) oil shortages, (iii) increased
use of biofuels, (iv) trade embargos, and (iv) rapidly growing food demand in China and India,
have challenged the ability of the developed world to maintain adequate food supply (Hubbard
and Hubbard, 2013). To that end, Small Farms are suggested to be the key to resolve the developed
world’s food security and sustainability challenges (Eurovia, 2013). However, an integrated
framework that could assist the assessment of the SFs’ impact upon the triple helix of sustainability
(i.e. economic, environmental and social) in the developed world does not yet exist. Such a
quantitative model would be of great interest to policy-makers and enterprises in order to support
related decisions and/or regulations towards SFs’ development.
In this work, the diffusion of agricultural commodities produced in SFs is particularly
addressed. The objective of this study is two-fold: (i) to provide a policy-making support tool at
the strategic level, and (ii) to identify policies that could support the development of SFs for
ensuring food security in a sustainability context. The rest of the paper is organized as follows.
First, in Section 2 generic characteristics that highlight the significance of SFs towards sustainable
food security is provided. Following, in Section 3 a System Dynamics (SD) modeling framework
is developed for managing the adoption of SF products by consumers, while incorporating an
extension of the Bass diffusion model. The application of the proposed framework is further
illustrated on the real-world case study of Greece, and interesting policy-making interventions are
analyzed. Finally, in Section 4 conclusions and suggested areas for future research are discussed.
2. Sustainable food security and small farms: theoretical background
There is a myriad of reports highlighting the challenges of sustainable food security in the
developing countries. Nevertheless, this issue has been ineffectively tackled for the case of the
developed countries.
Economic Challenges: In the develop world, the continuously rising food prices along with the
low economic recovery-rates from the global financial crisis highlight potential food insecurity.
Specifically, in Europe citizens spend one fifth of their income on food supplies, thus further
deepening social inequality in the region (FAO, 2011). Moreover, Europe is not food sufficient
considering that it suffers from 70% protein deficiency. This means that the European rural
development policy-makers need to promote the domestic production of protein crops at the
expense of other arable crops (Noleppa and Cartsburg, 2013). Moreover, in Europe a 2% decrease
in total agricultural output has been observed during the last decade (ETH, 2014), while in the
United States a 20% decline in farm holdings is reported (USDA, 2005).
Environmental Challenges: Loss of biodiversity, combined with water scarcity from overuse,
soil erosion and depletion as well as climate change may reduce agricultural yield by at least 5-
25% by 2050 (Dimitri, Effland and Conklin, 2009). Additionally, every year 1.3 billion tons of
food are wasted globally and in the industrialized world over 40% of this wastage occurs at the
retail and consumer levels (NRDC, 2012).
Book of Proceedings
195
Social Challenges: In the developed nations live about 15.7 million people who suffer from
chronic hunger and undernourishment (FAO, 2013). The most worrying fact is that the rate of
chronic hunger and undernourishment in developed countries has risen since 1990 by 15.7% (FAO,
2013). Furthermore, obesity almost doubled between 1980 and 2008 globally and in Europe nearly
50% of both men and women are overweight (WHO, 2008). Also, statistics highlight that in Europe
1 in 3 of 11-year-old children are overweight (WHO, 2008). Another concerning fact is the ageing
population in Europe (Eurobarometer, 2012), that increases the appetite for a diet rich in
carbohydrates and animal protein (ETH, 2014).
Taking into account the aforesaid challenges, a radical transformation is required to promote
sustainable agricultural intensification (UNEP/IFAD, 2013). To that end, the need to enhance the
role of smallholder farming towards food production and natural resource stewardship is critical.
This is further highlighted by the contemporary funding schemes of the European Union (EU) to
support initiatives for the development of short food supply chains and local food systems in the
Community (EAFRD, 2013). In addition, the EU recognizes the significance of SFs in food and
nutrition security and has already approved funds to support small food businesses (European
Commission, 2013).
Europe hosts around 14 million farms with the SFs to account for 2.5% of the total used
agricultural area (EU, 2013). Nevertheless, SFs in Europe disappear; in Europe around 3 million
farms (20% of the total number of farms) have disappeared during the last eight years, mainly SFs
(Eurovia, 2013). However, SFs could stimulate local business and job creation (Diao et al., 2007).
Furthermore, SFs have been found to reduce poverty gap more intensively than other sectors
(Christiaensen, Demery and Kuhl, 2011). Additionally, SFs are reported to promote welfare
through effective nutrition intake (Faber and Wenhold, 2007). Smallholdings are also documented
to be more resource-efficient (Altieri and Koohafkan, 2008), and more productive per hectare than
large-scale plantations (Borras, Kay and Akram-Lodhi, 2007).
3. System dynamics framework
In this section, a novel modeling approach for managing the diffusion of commodities produced
in SFs is developed, merging: (i) the theory of new product diffusion adapted from the field of
marketing, and (ii) the theory of SD, which has a proven track record for tackling strategic
decision-making problems. The main goals of the model are to (a) study the adoption of SFs
products by consumers, (b) predict the SFs evolution during a given time horizon, and (c) evaluate
the impact of alternative policy interventions on SFs.
The model is based on an extension of the Bass Diffusion Equation (Bass, Trichy, and Dipak,
1994):
dN
dt= {[p+
q
mN(t)] [m-N(t)]} x(t)
where, N(t) is the cumulative number of SFs products′ consumers (adopters) in a specific
geographical region at time t, dN/dt is the rate of change for the SFs products’ consumers
(adopters) at time t, m stands for the sum of potential consumers (fraction of regional or national
Book of Proceedings
196
population), p is the coefficient of innovation, q is the coefficient of imitation, and x(t) is an general
intervention function that describes the current effect of the time-dependent external decision
variables on the probability of SFs product consumption at time t.
Figure 1 presents a simplified conceptual model of the system under study. The input
parameters include the innovation and imitation factors, p and q, that represent the corresponding
trends for the first time consumers (innovators) of SFs products, and the word-of-mouth effect, i.e.
consumers that purchase the product after being influenced by previous adopters (imitators),
respectively. Moreover, the utilized agricultural area (UAA) that refers to the land used for farming
in a given region is used as a measure of the aggregate large and small farms’ size, while the
standard gross margin (SGM), that is the difference between the value of the agricultural output
(crops or livestock) and the cost of inputs required to produce that output, is used as a measure of
the economic size of the agricultural holdings. The labor force employed in these farms is also an
input parameter, as well as the productivity factor (α) that express the ratio of small farms to large
farms productivity. Finally, the total population of a given region (P), and the target population,
i.e. the total market size/sum of potential consumers, are also critical inputs for the diffusion model.
The data regarding the aforementioned input parameters were obtained through Eurostat, FAO,
and similar systems. The outcome variables of the model, which stand as the system’s key
performance and monitoring indicators, are the total employees in small farms, and the total profit
of small farms across the region under study.
The model was applied in the case of Greece. The preliminary results (Figure 2) indicate the
positive future potential of small-scale farming, in terms of employment (total SFs employees) and
profits (total SFs profits). Indicatively, given that no intervention is applied to accelerate the SFs
products market diffusion, the total number of SFs’ employees and the total SFs’ profits, are
estimated to rise by 45%, and 30% respectively within 20 years. This auspicious prospect could
be further accelerated through appropriate interventions on the part of governments and
smallholders in order to influence consumers’ behavior regarding SFs products.
Figure 18. Conceptual model of the system under study.
Potential Small Farm Products
Consumers
Small Farm Products Consumers
Adoption as
Innovators
Adoptionas
Imitators
External Interventions
Environmental Awareness
Social Awareness
Economic Incentives
Health and Nutrition
B RWord-of-
mouthMarket
Saturation
Small Farm Products Production
Small Farm Total Employees
Small Farms Total Profit
+
+
+
+ ++
+
+ +
+
+
+
Book of Proceedings
197
Towards this direction and motivated by the EU’s recognition of the need to enhance the role
of SFs towards safeguarding sustainable development and food security across Europe, the
proposed model allows for the simulated implementation, monitoring, and evaluation of a set of
real-world policy-making interventions on the SFs products’ diffusion rate, through the general
intervention function x(t). The scope of these policy-making actions is to increase the rate of
consumption of agricultural products cultivated in SFs and consequently increase the SFs’ market
share in the developed countries. Specifically, four external factors that affect the adoption rate of
SFs products in the Greek food market were identified: (i) environmental awareness, (ii) social
awareness, (iii) economic incentives, and (iv) health and nutrition awareness. Indicative policy-
making actions towards the direction of developing SFs products’ production and consumption
include the increase of advertising expenditure in promoting the sustainability aspect of small-
scale farming (products and farming practices) (environmental awareness), the market promotion
of local products, related training and/or educational interventions, the promotion of healthy food
as a critical component of wellbeing (social awareness – health and nutrition), and the subsidy of
investments in agriculture (new farmers) (economic incentives).
Figure 2. Total employees and total profit evolution (small farms).
4. Conclusions
Food security has a multidisciplinary nature and it is an important issue not only for the developing
world, but also in the developed countries. The prevailing farming practices are unsustainable.
Small Farms could play a critical role in Food Supply Chains ensuring Food Security and
Sustainable development at the same time. The operation of food markets is quite complex and
perhaps, the (mainly social and environmental) merit of a small-scale agricultural production is
not enough to build momentum for increase the number of small farms. Thus, (national and
international) governance with selected interventions may have a key role. To this end, this paper
provides a new quantitative policy-making support tool based on System Dynamics methodology,
which has been used for the Greek food market. This is a first time effort and there is space for
improvements. Future research directions include further validation and verification of the System
Dynamics model based upon data from FAO and Eurostat, the examination of alternative diffusion
models, the development of a multi-level model incorporating all the EU-27 countries and different
agricultural products on a single-product basis (i.e. cereals, nuts, wheat etc.).
Book of Proceedings
198
Acknowledgement
This research has received funding from the European Union's Seventh Framework Programme
(FP7-REGPOT-2012-2013-1) under Grant Agreement No. 316167, Project Acronym: GREEN-
AgriChains, Project Full Title: “Innovation Capacity Building by Strengthening Expertise and
Research in the Design, Planning and Operations of Green Agrifood Supply Chains”, Project
Duration: 2012-2016. All the above reflect only the authors’ views; the European Union is not
liable for any use that may be made of the information contained herein.
References
Altieri M.A. and Koohafkan P. “Enduring farms: Climate change, smallholders and traditional
farming communities”. Environment and Development Series 6. Third World Network, Penang,
2008.
Borras S.M., Kay C. and Akram-Lodhi A.H. “Agrarian reform and rural development: Historical
overview and current issues”. In A.H. Akram-Lodhi, S.M. Borras Jr and C. Kay, Land, poverty
and livelihoods in an era of globalization. Routledge, London and New York, 2007, pp.1-40.
Christiaensen L., Demery L. and Kuhl J. “The (evolving) role of agriculture in poverty reduction
– an empirical perspective”. Journal of Development Economics. Vol. 96, 2011, pp.239-254.
Diao X., Hazell P., Resnick D. and Thurlow J. “The role of agriculture in development:
Implications for sub-Saharan Africa – IFPRI Research Report No. 153”. International Food Policy
Research Institute. Washington, DC, 2007.
Dimitri C., Effland A. and Conklin N. “The 20th Century Transformation of U.S. Agriculture and
Farm Policy”. United States Department of Agriculture. Washington, D.C., 2005.
EAFRD. “Short Food Supply Chains and Local Food Systems in the EU. A State of Play of their
Socio-Economic Characteristics”. European Agricultural Fund for Rural Development. Seville,
2013.
ETH. “Food Security in Europe in the Age of Global Climatic Change”. Available at:
<http://www.isn.ethz.ch/Digital-Library/Articles/Detail?id=179230>, Accessed on June 1st,
2014.
EU. “EU farm economics 2012”. European Commission Directorate-General for Agriculture and
Rural Development. Brussels, 2013.
Eurobarometer. “Special Eurobarometer 378 – Active Ageing”. European Commission.
Luxembourg, 2012.
European Commission. “Sustainable Food Security H2020-SFS-2015-2 – Small farms but global
markets: the role of small and family farms in food and nutrition security”. European Commission.
Brussels, 2013.
Book of Proceedings
199
Eurovia. “Land concentration, land grabbing and people’s struggles in Europe”. European
Coordination Via Campesina. Brussels, 2013.
Faber M. and Wenhold F. “Nutrition in contemporary South Africa”. Water SA. Vol. 33, No. 3,
2007, pp.393-400.
FAO. “The State of Food Insecurity in the World – Eradicating world hunger: taking stock ten
years after the World Food Summit”. Food and Agriculture Organization of the United Nations.
Rome, 2007.
FAO. “How to Feed the World in 2050. Food and Agriculture Organization of the United Nations.
Rome, 2009.
FAO. “The State of Food Insecurity in the World – How does international price volatility affect
domestic economies and food security?”. Food and Agriculture Organization of the United
Nations. Rome, 2011.
FAO. “The State of Food Insecurity in the World – Undernourishment around the world”. Food
and Agriculture Organization of the United Nations. Rome, 2012.
FAO. “The State of Food Insecurity in the World – The multiple dimensions of food security”.
Food and Agriculture Organization of the United Nations. Rome, 2013.
Foodsecure. “Foodsecure briefing for key European commission stakeholders”. FOODSECURE.
Brussels, 2013.
Hubbard L.J. and Hubbard C. “Food security in the United Kingdom: External supply risks”. Food
Policy. Vol. 43, 2013, pp.142-147.
Lin B.-H., Ver Ploeg M., Kasteridis P. and Yen S.T. “The roles of food prices and food access in
determining food purchases of low-income households”. Journal of Policy Modeling. In Press,
2013.
Noleppa S. and Cartsburg M. “Agricultural self-sufficiency of the European Union: Statistical
evidence”. Agripol – network for policy advice GbR. Berlin, 2013.
NRDC. “Wasted: How America Is Losing Up to 40 Percent of Its Food from Farm to Fork to
Landfill”. Natural Resources Defense Council. New York, 2012.
UN. “World Economic and Social Survey 2013: Sustainable Development Challenges”.
Department of Economic and Social Affairs of the United Nations Secretariat. New York, 2013.
UNEP/IFAD. “Smallholders, food security, and the environment. United Nations Environment
Programme/International Fund for Agricultural Development. Kenya/Rome, 2013.
Book of Proceedings
200
UNWFP. “Emergency Food Security Assessment Handbook”. United Nations World Food
Programme. Rome, 2009.
WHO. “World health statistics 2008”. World Health Organization. Geneva, 2008.
WHO. “Global status report on noncommunicable diseases 2010”. World Health Organization.
Geneva, 2011.
Book of Proceedings
201
Towards the implementation of optimal train loading plan in the Athens –
Thessaloniki freight services
A. Ballis, Assoc. Professor
Department of Transportation Planning and Engineering, National Technical University of
Athens, 5, Iroon Polytechniou, GR-15773, Athens, Greece, email: [email protected]
F. Karapetis, Senior Software Eng.
TRAINOSE, Karolou 1 Athens 104 37, Greece
Th. Ballis Civil/Transportation Eng.
National Technical University of Athens, Greece
Abstract
An effective train loading plan contributes positively to the profitability of the railway services, to
train safety, to energy consumption and to the efficiency of rail terminal operations. The main goal
of an optimized train loading is the proper assignment of loading units to the wagons of a train so
that the utilization of the train is maximized while taking under consideration the maximum axle
load restrictions imposed by the design or the condition of the railway infrastructure, operating
conditions and safety regulations. The problem typically is expressed in two ways: (a) given a
predefined commodity load, which is the minimum number of wagons required to perform the
transportation task or (b) given a standard set of wagons (e.g. in the case of a shuttle train) which
is the maximum commodity weight or volume that can be transported. The current work describes
the way that the train loading plan has been analyzed, solved and integrated in the information
system that supports the new railway service of TRAINOSE for container transport in the Athens
– Thessaloniki line. This new service, named iCS, was launched in December 2013 and since then
operates on a daily basis. The work includes the literature review on the train loading plan, the
pragmatic aspects of the wagon loading problem and the heuristic implemented in the information
system of iCS service. It also includes the results of the validation process of the above heuristic
against an accurate (but more demanding in computational time) mixed integer optimization
model. The presentation concludes with the proposed solution for the iCS wagon loading problem.
Keywords: Train loading plan, wagon loading heuristic, mixed-integer programming.
Book of Proceedings
202
1. Introduction
An effective train loading plan contributes positively to the profitability of the railway services, to
train safety, to energy consumption and to the efficiency of rail terminal operations. The main goal
of an optimized train loading plan is the proper allocation of loading units to the wagons so that
the utilization of the train is maximized while satisfying a number of restrictions. For the narrow
view (optimization of train utilization) typical restrictions are the length of the wagon, the
maximum wagon axle load due to railway infrastructure conditions or due to safety regulations,
etc. For the wider view of the problem (optimization of the whole system namely, train, terminal
and trucks) typical restrictions are the performance of terminal handling equipment, the acceptable
average truck waiting time etc.
In the current work, the algorithm that performs the container train loading plan for TRAINOSE’s
intermodal Cargo Shuttle (iCS) is presented and validated. The remainder of the paper includes:
the literature review (Section 2) where studies addressing the narrow and the wider view of train
loading problem are presented. Section 3 that presents the pragmatic aspects of a wagon loading
process as well as the algorithm/heuristic developed and used by iCS. Next in Section 4 the process
for the evaluation of the above heuristic is analyzed. The last part, Section 5, hosts the conclusions.
2. State of the Art
The problem of train loading is a subcategory of the known "bin packing" problem. In this problem,
objects of different volumes (containers) must be packed into a finite number of bins (wagons) in
a way that minimizes the number of bins used. Solving techniques based on mathematical models
using integer linear programming or metaheuristics are the most well-known for this class of
problems. Examples of the constraints taken into account are (narrow view) the maximum weight
attached to wagons, the maximum number of containers per wagon, the number of wagons
attached to a train and its total weight as well as (wider view) terminal equipment resources,
maximum truck waiting time allowed, etc.
One of the first works concerning the narrow view of the problem is the one of Feo and Gonzáles-
Velarde (1995) who treated a problem where trailers had to be assigned to wagons. Their models
and solutions were based on the assumption that no more than two trailers can fit on one wagon.
Nearly a decade later, Corry and Kozan (2006) optimize the load planning with respect to the
service time of the train and the weight distribution along the train. Only one type of container and
no weight restrictions for the wagons have been taken into account. In a subsequent paper Corry
and Kozan (2008) aimed at minimizing the train length and the train service time. Neither weight
restrictions for the wagons nor for the whole train have been considered. The model is formulated
(but not solved) as an integer linear program and real-world problem instances have been treated
by use of heuristics (local search). A recent (2011) work on the subject was this of Aggoun et al
(2011) that incorporated into the problem the aspects of business constraints, like handling of
dangerous goods and incompatibilities between families of containers.
Book of Proceedings
203
As far as the wider view of the train loading problem, Powell and Carvalho (1998) aimed to
optimize the circulation of intermodal wagons. The same year, in Bostel and Dejax (1998) the
positioning of containers on incoming and outgoing trains is optimized with the aim of reducing
transport distances of containers that have to be handled by cranes. Bruns et al. (2010) proposed
three different integer linear programming formulations in which the real weight restrictions
related to wagon configurations are considered. The resulting train loading plan is calculated so
the terminal’s operating costs are minimized. Anghinolfi et al. (2012) studied a case with multiple
trains of different destinations and took into account the minimization of distances between the
stocking area and the train. In 2012, Basetas investigated the spectrum of parameters involved in
the problem and performed a comparison among the allocation methods concerning their
applicability to the train loading problem.
3. Pragmatic aspects in container wagon loading
The problem of train loading plan (in its narrow view) is typically expressed in two ways: (a) given
a predefined commodity load, which is the minimum number of wagons required to perform the
transportation task or (b) given a standard set of wagons (e.g. in the case of a shuttle train) which
is the maximum commodity weight or volume that can be transported.
In any case and in order to take into account pragmatic aspects of train loading, the understanding
of the mechanics of wagon loading must be acquired. Figure 1 presents the (typically uneven)
distribution of forces on the bogies of the wagon. The effect of uneven allocation of forces is
explained through an example of “good wagon utilization” versus “bad wagon utilization” shown
in Figure 2. In this example a 20 tons per axle limit has been assumed, resulting to maximum of
40 tons per bogie. On the left side of Figure 2 the loaded containers exhausts the loading capacity
of the left bogie (F1 = 39 tn ~ 40 tn) while the loading capacity of the right bogie remains at 18 tn.
This leads to lower wagon utilization by 21 tn. On the contrary, the right side of Figure 2 both
forces F3 and F4 are reaching the loading limits of 40 tn, leading to a 79 tn wagon weight. The
identification of suitable container combinations (among the set of containers to be transported by
the train) can maximize the overall transported commodity weight and therefore the profitability
of the transportation service.
Book of Proceedings
204
Figure 1: Distribution of the weight of a loaded container into wagon bodies
Figure 2: Examples of good and bad wagon utilization
Book of Proceedings
205
The Current Train Loading System of TRAINOSE
In December 2013 TRAINOSE launched an intermodal Cargo Shuttle (iCS) which connects
Athens to Thessaloniki on a daily basis. Customers have to enter their orders in the web site of iCS
(Figure 3) which must respond immediately to inform them if the order can be accepted or not. In
order to do this, the program must “simulate” the train loading process in a way that maximizes
the utilization factor of each wagon and provide the answer (to accept or refuse the customer’s
order) instantly. It must be noted that the above train loading problem is re-solved each time a new
order enters the system and for this reason the program code must be lightweight to keep the
computational effort low. For this reason the software has incorporated a heuristic based on the
Best Fit allocation method (see Figure 4). According to Best Fir rule, each new container is
allocated onto the wagon which, after the container loading, has the least remaining capacity. The
iCS heuristic is used to load 45, 40 and 20 feet containers on a 60 feet wagon used for the iCS
container transport service. In terms of length, the following cases are investigated:
Case 202020 (three 20 feet containers loaded on the same wagon)
Case 2040 (one 20 feet container along with a 40 feet container on the same wagon)
Case 45 (only one 45 feet container on the wagon as the remaining length of 15 feet cannot
accommodate any other container)
Case x2020 (two 20 feet containers loaded on the center of the wagon. This is technically
feasible as the wagons used allow for a container placement starting 10 feet away from the
edge of the wagon. The mark {x} indicates a 10 feet space)
Case x20x20 (ten feet space, one 20 feet container, ten feet space, 20 feet container)
As it is shown in Section 5, the results of the heuristic are not always optimal but the computational
effort is always very low.
Figure 3: The web based interface of the program
Figure 4: Illustration of Best/Worst/First/Next fit allocation [Source: Basetas (2012)]
First fit: container loaded on the first available slot (leftmost or rightmost side)
Next fit: container loaded on the next available slot relatively to the reach stacker’s position
Best fit: container loaded on the slot which minimizes the wagon’s remaining capacity
Worst fit: container loaded on the slot which maximizes the wagon’s remaining capacity
Book of Proceedings
206
4. Evaluating the efficiency of the loading algorithm
The train loading plan, in its narrow view, can be solved to optimality if modeled as a mixed integer
optimization problem. The disadvantage of this approach is the high computational power that
demands as well as for the need for an (expensive commercial) solver.
In the current work, the accuracy of the currently used heuristic was measured by comparing its
results with these of the mixed integer programming model using the solver of NEOS (Network-
Enabled Optimization System) server (see, Czyzyk et.al. (1998), Dolan (2001) & Gropp et.al.
(1997)). The model takes into account wagon loading weight restrictions, the overall train weight
and the train length restriction. One aspect that differentiates this model from the work of other
researches is that the sequence of servicing the customer orders is retained. Customer orders are
served according to the first-in sequence (e.g. an order which was accepted cannot be replaced by
a posterior order, even if this increases the overall train utilization).
In order to measure the efficiency of the heuristic a series of tests were executed. Each scenario
assumes a train with 20 wagons (having 60 feet useful platform length) to be loaded by 35
container orders. Each order concerns a container transport with specific length and weight
characteristics. The container length is generated randomly from an empirical {45, 40 and 20 feet
container} distribution. The container weight is defined randomly from an empirical {container
weight distributions per container type} (see Tournaki (2014)).
Chart 1 presents the results after the completion of 800 scenarios, each one solved with iCS
heuristic and with the mixed integer model. In 257 cases (32%) the heuristic produced the same
(optimal) solution with the optimization model. In 424 cases (53%) the heuristic used one extra
wagon and in 105 cases (13%) used two extra wagons compared to the mixed-integer model.
Finally, for the last 14 cases (2%) the iCS heuristic used three more wagons in order to allocate all
containers onto wagons.
The best solution that combines low computational effort and optimal solution is a suitable
combination of both solvers: e.g. in the case of a 20 wagon train, the iCS heuristic can be used to
allocate containers to the first 16 wagons (20 – 3 worst difference – 1 safety margin) while the
mixed integer model will be used to finalize the loading of the latest customer orders (last 4
wagons) rearranging all the available containers to the appropriate wagon positions.
Next the wagons are sorted from the heaviest to the lightest to satisfy train breaking rules that
require that heavy loaded wagons must be placed near to the locomotive of the train.
Book of Proceedings
207
Chart 1: Comparison of the best fit heuristic against mixed integer optimization
5. Conclusions
The current work presents the wagon loading heuristic of the information system of iCS service
that performs the “simulation” of train loading each time a customer enters a new order in the
system. The performance of this algorithm was evaluated against the solutions produced by an
integer programming model. The analysis concluded that the current algorithm although fast,
produces results which in most cases are near optimal. Precisely, the tests concluded that in 32%
of the cases the iCS heuristic provides optimal results while in 53% of the cases the iCS heuristic
uses one more wagon to accomplish the allocation of all containers. For the 13% of the cases the
gap of the result between the heuristic and the optimization model is two wagons and in some rare
cases (2%) the optimization model uses three wagons less in comparison to the heuristic.
In order to improve the results while retaining small execution time, a combined approach could
probably be applied: The first container orders will be allocated to wagons by using the iCS
heuristic (in order to benefit from the low computational effort required) while the few latest
container orders will be allocated by using the mixed-integer optimization model that provides
the optimal solution (rearranging all the available containers to the appropriate wagon positions).
Next the wagons are sorted from the heaviest to the lightest to satisfy train breaking rules.
References
Feo TA, Gonzáles-Velarde JL. “The intermodal trailer assignment problem”. Transportation
Science 29(4), 1995, pp.330–341.
Corry P, Kozan E. “An assignment model for dynamic load planning of intermodal trains.
Comput. Oper Res 33, 2006, pp.:1–17.
257
424
105
14
0
50
100
150
200
250
300
350
400
450
0 1 2 3
Sce
nar
ios
Additional wagons required by the heuristic
0
1
2
3
Book of Proceedings
208
Corry P, Kozan E. “Optimised loading patterns for intermodal trains”. OR Spectr 30(4), 2008,
pp:721–750.
Aggoun A., Rhiat A. & Grassien J.P. Online Assignments of Containers to Trains Using
Constraint Programming. Proceedings of the 5th International Workshop on Multi-disciplinary
Trends in Artificial Intelligence Hyderabad, India, December 7-9, 2011, pp. 395-405.
Powell, B. & Carvalho, A. “Real-time optimization of containers and flatcars for intermodal
operations”. Transportation Science, 32, 1998, pp:110-26.
Bostel N, Dejax P. “Models and algorithms for the container allocation problem on trains in a
rapid transshipment yard”. Transp Sci, 32(4), 1998, pp.370–379
Bruns F., Knust S. “Optimized load planning of trains in intermodal transportation”. OR
Spectrum, 2010, Published online.
Anghinolfi D., Foti L., Maratea M., Paolucci M., Siri S. “Optimal loading plan for multiple trains
in container terminals”. 5th International Workshop on Freight Transportation and Logistics.
2012, Mykonos, Greece.
Basetas E. “Railway freight transport: Algorithms for the loading of containers onto railway
freight wagons”, Diploma Thesis, NTUA, 2012.
Czyzyk J., Mesnier M. P., and Moré J. J. “The NEOS Server”. IEEE Journal on Computational
Science and Engineering 5(3), 1998, pp.68-75. This paper discusses the design and
implementation of the NEOS Server.
Dolan E. “The NEOS Server 4.0 Administrative Guide”. Technical Memorandum ANL/MCS-TM-
250, Mathematics and Computer Science Division, Argonne National Laboratory, 2001. (This
technical report, which discusses the implementation of the server and its use in detail, is available
for download in PDF format).
Gropp W. and Moré J. J. “Optimization Environments and the NEOS Server”. Approximation
Theory and Optimization, M. D. Buhmann and A. Iserles, eds., Cambridge University Press, 1997,
pp. 167-182. (This paper discusses the NEOS Server as a problem-solving environment that
simplifies the formulation of optimization problems and the access to computational resources).
Tournaki E. “The carbon footprint of rail intermodal freight transport: Case study of Athens-
Thessaloniki line”, Diploma Thesis, NTUA, 2014.
Book of Proceedings
209
An exact method for the inventory routing problem
Evangelia Chrysochoou
University of Thessaly / Department of Mechanical Engineering Pedion Areos, 38334, Volos
Greece
Prof. Athanasios Ziliaskopoulos
University of Thessaly / Department of Mechanical Engineering Pedion Areos, 38334, Volos
Greece
Email of corresponding author: [email protected]
Abstract
Vendor inventory management is a concept which is adapted highly nowadays where the decision
maker in the process is the supplier. The combination of the inventory management with the
vehicle routing problem constitutes one of the latest trends of logistics and supply chain
management and constitutes the backbone of vendor managed inventory systems. As new
emerging technologies are introduced in the context of freight transportation systems, research
requires the development of new models and algorithms that can incorporate their advantages. In
this context, this paper aims to discuss all significant elements of inventory routing problem. New
valid inequalities are proposed to stronger the formulation of the transported quantities and
enhance the Maximum Level (ML) policy. This approach was motivated by the fact that, nowadays
where infrastructures were manufactured for much higher consumption rates of goods, retailers
are opposed to the Order – Up – to level (OU) policy and look for more economic and competitive
inventory plans. A branch and cut algorithm was developed to solve the problem exactly. In order
to evaluate the performance of the algorithm the benchmark instances set for the single vehicle
case created by Arhetti et al. (2007) was used. Computational results have shown that this approach
improves the optimal solution on an average at least 20%.
Keywords: Vendor Managed Inventory, Inventory Routing, Transshipment, Branch and Cut
1. Introduction
Vendor Managed Inventory (VMI) systems seem to be one of the most tractable business model
nowadays in global logistics and supply chain operations. This is increasingly the case for
electronics and automotive parts manufactured in China and assembled in the European Union
countries. Most of these parts are assembled in five (5) major plants in Central Europe, operating
with Just – In – Time production procedures, using the VMI principles. The general concept
behind this model is that replenishments and distribution making process is centralized at the
supplier level. It is characterized as a win – win situation for both supplier and manufacturers, or
in general retailers due to the fact that it provides the ability to the supplier to combine and
coordinate the demand and shipments of a network of retailers (or more generally stock holding
entities, such as manufacturers, wholesalers, retailers or 3rd party logistics providers). On the other
hand these retailers secure the shortage of their inventories without allocating resources to control
and manage them. Backbone of the VMI system is the solution of inventory routing problem (IRP)
Book of Proceedings
210
which is one of the most interesting extensions of routing problems. IRP combines the decision
process of inventory management and distribution – transportation of goods. The decision maker
in such a model has to make three decisions: the amount to be transported, the frequency of
shipments as well as the distribution plan. However, the IRP in practice becomes meaningful when
customers’ demand is considered to be stochastic instead of assuming a fixed usage rate. The basic
difference behind the SIRP and the deterministic IRP is the level of realism and the difficulty of
solving instances given the data in a probabilistic sense. In a two stage stochastic program a long
term anticipatory decision must be made prior to the full information of the random parameter of
the problem and short terms decisions are available as recourse actions once the uncertainty has
been revealed. The overall aim is to make “here and now” a decision which minimizes the total
expected cost associated with both the long term and the short term decisions (Carøe and Tind.
,1998). IRP was introduced 30 years ago by the seminal paper of Bell et al., (1983) which studied
the case with stochastic demand accounting only for transportation costs. They proposed a linear
programming model to solve the deterministic version of the problem. To the best of our
knowledge there are two very recent literature reviews on the subject. We refer to the work of
Andersson et al. (2010) which was focused mostly on industrial aspects and Coelho et al. (2014)
which provides the most up to date overview of the problems and methodologies of the VMI
problem. Bertazzi, Palettas and Speranza (2002) introduced a practical VMI policy, called
deterministic order – up – to – level (OU) policy for the IRP. Based on the proposed policy Arhetti
et al. (2007) developed the fist exact algorithm using a branch and cut scheme for the single
vehicle. Based on their work very recently Coelho and Laporte (2013) and Adulyasak et al. (2014)
have solved multivehicle version of IRP in a branch and cut fashion under OU and maximum level
(ML) policies. Solyali and Sural (2011) also based on the work of Arhetti et al. (2007) proposed a
strong formulation for the inventory replenishment part of the IRP. In this paper new valid
inequalities are introduced to enhance the computational process of the optimal transported
quantities under the ML policy. This approach was motivated by the fact that in the context of a
deterministic model all parameters are known at the beginning of the process; thus a vendor can
take advantage of the fact that the he knows the total demand of each stock keeping venue in
advance and can transport quantities in an early stage in order to fulfill the future known demand.
However, the amounts that he is able to transport are bounded by the amounts that are made
available to him at each stage. This seems to be an important issue for major multinationals that
ship parts from China to Europe to be assembled in a number of locations in Central Europe, but
also keep inventory either in 3rd party facilities or at the production sites. Transshipment is in fact
a recourse action they use in practice in case that there is shortage at a particular venue. These
important realizations gave us the motivation to introduce new inequalities in order to enhance the
ML policy. The key deference of ML policy in contrast with the OU policy is that the supplier is
free to decide about any quantity to be transported to the inventories of his retailers (in fact stock
keeping venues) bounded only by their capacity or maximum level defined by them. On the other
hand OU policy restricts the amount to be such that fills the inventory to its capacity. However
nowadays where infrastructures were manufactured for much higher consumption rates of goods
retailers are opposed to the OU policy and look for more economic and competitive inventory
plans. A convenient approach to address these particularities is Coelho and Laporte’s (2013)
proposed new tactical policy, called optimized target level that yields lower cost and inventory
levels than the OU policy. Reviewing their approach in comparison to the strong formulation of
Solyali and Sural (2011) this paper was motivated to introduce new bound in order to determine
optimal quantities to be transported. Therefore, we introduce a modification for the mixed integer
Book of Proceedings
211
programming model of the IRP. To the best of our knowledge this assumption was not proposed
before. The remainder of this paper is organized as followed. In §2, we give the formal description
of the deterministic IPR model and the branch and cut. Computational results are provided in §3.
Significant remarks as well as conclusions are given in§4.
2. Inventory routing problem deterministic model
We consider an inventory routing problem where a supplier denoted by node 1 is distributed to N-
1 retailers over a finite discrete time T, using a single vehicle of capacity C. Traditionally the
problem is defined on an undirected graph G=(V,E) where {1} is the vertex representing the
supplier and vertices V’ = {2,3,…,N-1} represent the set of stock keeping venues (will be called
retailers from thereafter as it is commonly called in the literature). A = {(i, j): i ≠ j, i, jϵV} is the set of
arcs . Inventory holding cost occurs for both supplier and set of retailers and is denoted as ℎ𝑖 ∶ 𝑖𝜖𝑉
per period and each vertex has and inventory capacity𝐶𝑖 ∶ 𝑖𝜖𝑉. The length of the discrete planning
horizon is H where 𝑡𝜖𝑇 = {1,… , 𝐻}. At the beginning of the planning horizon the decision maker
knows that (1) each period the quantities 𝑟𝑡 is made available to the supplier in order to fulfill the
request of his retailers; (2)the initial inventory levels of both supplier and t retailers are known {
𝐼10, 𝐼𝑖
0, 𝑖𝜖𝑉′} (3) of each retailer at each period is denoted with 𝑑𝑖𝑡 , 𝑖𝜖𝑉′. A single vehicle can perform
the route one at each period with capacity C, and a routing cost 𝑐𝑖,𝑗is associated with arc(i, j)ϵA.
Throughout the paper we assume that since the supplier has the information of the demand of his
retailers in advance he can transport the quantities 𝑞𝑖𝑡 to meet the demand of period t and subsequent
periods as well. However the available quantities 𝑟𝑡 shall be added to the total available quantities
at period t can be used for deliveries to retailer in the same period t and subsequent periods. The
objective function is defined in a way to minimize the total transportation and inventory cost of
the whole planning horizon while meeting the demand of each retailer.
minimize∑∑ℎ𝑖𝐼𝑖𝑡
𝑖𝜖𝑉𝑡𝜖𝑇
+∑∑∑𝑐𝑖𝑗𝑥𝑖,𝑗𝑡
𝑖𝜖𝑉𝑖<𝑗
(1)
𝑖𝜖𝑉𝑡𝜖𝑇
Subject to the following constrains:
𝐼1𝑡 ≥ 0 ∀𝑡 ∈ 𝑇 (2)
𝐼1𝑡 = 𝐼1
𝑡−1 + 𝑟𝑡 −∑𝑞𝑖𝑡
𝑖∈𝑉′
∀𝑡 ∈ 𝑇 (3)
𝐼𝑖𝑡 ≥ 0 ∀𝑡 ∈ 𝑇 , ∀𝑖 ∈ 𝑉′ (4)
𝐼𝑖𝑡 = 𝐼𝑖
𝑡−1 + 𝑞𝑖𝑡 − 𝑑𝑖
𝑡 ∀𝑡 ∈ 𝑇, ∀𝑖 ∈ 𝑉′ (5)
𝐼𝑖𝑡 ≤ 𝐶𝑖 ∀𝑡 ∈ 𝑇 , ∀𝑖 ∈ 𝑉′ (6)
∑𝑞𝑖𝑡
𝑖∈𝑉′
≤ 𝐶 ∀𝑡 ∈ 𝑇 (7)
∑𝑞𝑖𝑡 =
𝑡∈𝑇
∑𝑑𝑖𝑡
𝑡∈𝑇
− 𝐼𝑖0, ∀𝑖 ∈ 𝑉′ (8)
Book of Proceedings
212
𝑞𝑖𝑡 ≤ 𝑦𝑖
𝑡∑𝑑𝑖𝑗
𝐻
𝑗=𝑡
, ∀𝑖 ∈ 𝑉′, ∀𝑡 ∈ 𝑇 (9)
∑𝑞𝑖𝑗
𝑡
𝑗=1
≤ 𝐼10 +∑𝑟𝑗
𝑡
𝑗=1
, ∀𝑡 ∈ 𝑇, ∀𝑖 ∈ 𝑉′ (10)
∑∑𝑥1𝑗𝑡
𝑗∈𝑉′𝑡∈𝑇
≤ 𝐻 (11)
∑𝑥1𝑗𝑡
𝑗∈𝑉′
≤ 𝑦1𝑡 ∀𝑡 ∈ 𝑇 (12)
∑𝑥𝑖𝑗𝑡
𝑗∈𝑉′
= ∑ 𝑥𝑗𝑖𝑡 ∀𝑡 ∈ 𝑇 ,
𝑗∈𝑉′
∀𝑖 ∈ 𝑉′ (13)
∑𝑥𝑖𝑗𝑡
𝑗∈𝑉′
+∑ 𝑥𝑗𝑖𝑡 = 2𝑦𝑖
𝑡 ∀𝑡 ∈ 𝑇 ,
𝑗∈𝑉′
∀𝑖 ∈ 𝑉′ (14)
𝑥𝑖,𝑗𝑡 ≤ 𝑦𝑖
𝑡∀𝑡 ∈ 𝑇, ∀𝑖, 𝑗 ∈ 𝑉′ (15)
𝑥𝑖,𝑗𝑡 ≤ 𝑦𝑗
𝑡∀𝑡 ∈ 𝑇, ∀𝑖, 𝑗 ∈ 𝑉′ (16)
𝐶(1 − 𝑥𝑖,𝑗𝑡 ) + 𝑢𝑖
𝑡 ≥ 𝑢𝑗𝑡 + 𝑞𝑗
𝑡
∀𝑖, 𝑗 ∈ 𝑉′: 𝑖 ≠ 𝑗, 𝑡 ∈ 𝑇 (17)
𝑢𝑖𝑡 ≤ 𝑞𝑖
𝑡 ∀𝑖, 𝑗 ∈ 𝑉′, 𝑡 ∈ 𝑇 (18)
𝑢𝑖𝑡 ≤ 𝑦𝑖
𝑡 ∗ 𝐶 ∀𝑖 ∈ 𝑉′, 𝑡 ∈ 𝑇 (19)
𝑦𝑖𝑡 ≤ 𝑦1
𝑡∀𝑡 ∈ 𝑇, ∀𝑖 ∈ 𝑉′ (20)
𝑞𝑖𝑡 , 𝑢𝑖
𝑡 ≥ 0∀𝑖 ∈ 𝑉′, 𝑡 ∈ 𝑇 (21)
𝑥𝑖,𝑗𝑡 ∈ {0,1} ∀𝑖, 𝑗 ∈ 𝑉′: 𝑖 ≠ 𝑗, 𝑡 ∈ 𝑇 (22)
𝑦𝑖𝑡 ∈ {0,1} ∀𝑖 ∈ 𝑉, 𝑡 ∈ 𝑇 (23)
Constraints (2) and (3) are related to the inventory level at the supplier’s site. The first one
expresses the fact that inventory level at the supplier level cannot be negative in any period, thus
avoiding a stock out situation. The second one defines the inventory level at the supplier at the end
of period t by the inventory level at the end of period t-1, minus the total quantities to be transported
at period t, plus the quantities 𝑟𝑡that are made available at time t. Constraint (4) secures the stock
out avoidance of each retailer as well. Constraint (5) defines the inventory level at each retailer at
the end of period t by the inventory level at the end of period t-1, plus the quantities that is made
available at period t – the demand at period t as well. Constraint (6) secures that the inventory level
of each retailer cannot exceed its capacity. Constraint (7) – (10) defined the quantities delivered.
These set of constrains are opposed to the OU policy instead they aim to secure the ML policy.
More precisely constraint (7) secures that for each period the quantities to distribute cannot exceed
the capacity of the vehicle. Constraint (8) declares that the total quantities to be transported to each
retailer are equal to the total demand over the whole planning horizon minus the starting inventory
Book of Proceedings
213
level. Constraint (9) expresses the fact that the quantities to be transported to each retailer at period
t can be less or equal to the demand requested at period t and subsequent periods when the retailer
is served at period t. Constraint (10) ensures that the transported quantities at period t cannot
exceed the suppliers staring inventory level plus the product made available since period t.
Constraints (11) – (20) serves the routing counterpart of the problem. More specifically, constraint
(11) secures that the total number of routes cannot exceed the number of periods of the planning
horizon, however it is not necessary to perform a route for each period. Constraint (12) ensures
that if a route is performed at time t it will start from the supplier and will visit only one retailer.
Constraints (13) and (14) secure the flow of the route among intermediate retailers. Constraints
(15) and (16) define the relationship of the two indexed of the three indexed variables of the routing
constrains and stated that when a retailer is served at time t he will be an origin or a destination of
a valid path. Constraints (17) – (19) is the well known sub tour elimination constrains based on
the Miller-Tucker-Zemlin (MTZ) constraint formulation also suggested by Anken et. al. (2012);
this is achieved by introducing extra variables 𝑢𝑖𝑡 that express the quantities that are in the vehicle
until retailer i. Constraint (20) secures that if a route is performed at period t , then there will be
intermediate points in the route. Constraints (21) – (23) enforce integrality and non - negativity
conditions. The IRP is NP – hard since it contains the VRP as a special case. If the problem size
is relatively small the formulation can be solved by the framework of a branch and cut algorithm
as follows: Initially at a generic node of the search tree the relaxed linear program defined by the
(1) – (16) and (20) to (23) is solved. Next a search of violated sub tour elimination constraints (17)
– (19) is made and sequentially those constraints are generated and introduced to the current
problem which is then re - optimized. The process is repeated until a feasible or dominated solution
is reached, or until there are no more cuts to added and then branching on fractional variables is
performed.
3. Computational results
The algorithm described above was coded in C++ using IBM Concert Technology and CPEX 12.4
with 2 threads. All computations were executed in an Intel Atom 1.83 GHz and 2 GB RAM
personal laptop with maximum time of 2 hours. To evaluate the performance of the algorithm, we
have used the benchmark instances set for the single vehicle case created by Arhetti et al. (2007).
Those instances was used to evaluate the performance of the proposed valid inequalities for the
ML policy in coherent to the OU policy. The small instances up to 20 customers were used for
both high and low level of inventory holding cost. The small number of experiments is indicative
in order to present proposed approach potential solutions that yield almost 20% less IRP cost. The
computational results are shown in table 1 -2. Table 1 provides optimal solution of each of the 5
instances with 5, 10, 15 and 20 retailers. Table 1 contains the results of instances with time horizon
H = 3 and high inventory cost ( ℎ𝑖 ∈ [0.1 , 0.5] 𝑎𝑛𝑑 ℎ1 = 0.3) and results with low inventory cost (ℎ𝑖 ∈[0.01 , 0.05] 𝑎𝑛𝑑 ℎ1 = 0.03). Column 1 shows the corresponding name of the data set, columns 2 – 3
contain the CPU time (in sec) and the optimal value of the objective function as it was found by
Arhetti et al (2007) . Columns 4 – 5 contain the CPU time (in sec) and the optimal value of the
objective function of our model, and columns 6 – 7 contain the difference of the optimal solutions
and the percentage of it as well. Analogues remain columns contain the results on low inventory
cost.
Book of Proceedings
214
High Inventory cost , Horizon = 3 Low Inventory cost , Horizon = 3
Arhetti et.al. Chrysochoou&
Ziliaskopoulos
z*
Arhetti et.al. Chrysochoou
&
Ziliaskopoulos z*
Instances CPU z* CPU z* Diff %Diff CPU z* CPU z* Diff %Diff
abs1n5.dat 0 2149,8 1 1868 281,83 13% 0 1281,7 1 1210,1 71,58 6%
abs2n5.dat 0 1959,1 1 1583,7 375,39 19% 0 1176,6 1 967,76 208,9 18%
abs3n5.dat 0 3265,4 1 2533,3 732,14 22% 0 2020,7 1 1633,4 387,3 19%
abs4n5.dat 0 2034,4 1 1677,8 356,65 18% 0 1449,4 1 1245,89 203,5 14%
abs5n5.dat 0 2362,2 1 1819,4 542,74 23% 0 1165,4 1 959,4 206 18%
abs1n10.dat 0 4970,6 13 3678,9 1291,7 26% 0 2167,4 13 2126,44 40,93 2%
abs2n10.dat 0 4803,2 11 3842,4 960,78 20% 0 2510,1 11 2142,79 367,3 15%
abs3n10.dat 0 4289,8 3 3425,8 864,03 20% 0 2099,7 3 1802,02 297,7 14%
abs4n10.dat 0 4347,1 5 3324 1023 24% 0 2188 5 1702,31 485,7 22%
abs5n10.dat 0 5041,6 6 3835,2 1206,5 24% 0 2178,2 6 1740,55 437,6 20%
abs1n15.dat 0 5713,8 6 4585 1128,9 20% 0 2236,5 6 1980,18 256,4 12%
abs2n15.dat 1 5821 520 4593,4 1227,6 21% 1 2506,2 630 2100,03 406,2 16%
abs3n15.dat 4 6711,3 37 5222 1489,3 22% 1 2841,1 37 2344,44 496,6 18%
abs4n15.dat 1 5227,6 57 4083,3 1144,2 22% 1 2430,1 110 1980,16 449,9 19%
abs5n15.dat 3 5210,9 49 3952,2 1258,7 24% 2 2453,5 46 1914,1 539,4 22%
abs1n20.dat 10 7353,8 42 5585,4 1768,4 24% 12 2793,3 43 2124,09 669,2 24%
abs2n20.dat 8 7385 605 5821,8 1563,2 21% 6 2799,9 605 2341,76 458,1 16%
abs3n20.dat 5 7904 133 6006,8 1897,2 24% 8 3101,6 143 2431,14 670,5 22%
abs4n20.dat 4 7050,9 138 5570,2 1480,7 21% 4 3239,3 132 2580,13 659,2 20%
abs5n20.dat 11 8405,8 92 6976,8 1429 17% 7 3331 92 2731,41 599,6 18%
Table 19 Computational Results on instances with time horizon H = 3 and high and low
inventory cost
In the set of instances with high inventory cost the average percentage is 21. 3% and yields within
the interval (13.1 – 26) %. However in the second case with low inventory cost the average
percentage of improvement on the optimal solution found is 16.7% and yields within the interval
(1.9 – 24.1) %. This is due to the fact that the transportation cost is higher. Thus our approach can
perform significant saving in the cases where the inventory cost is high and competitive to the
transportation cost.
Book of Proceedings
215
4. Conclusions
In this paper the IRP problem was analyzed which constitutes the backbone of the well known
VMI systems. New valid inequalities were introduced in order to enhance the performance of the
ML policy in contrast to the OU policy which is used in most recent research papers. This approach
was motivated by the fact that nowadays retailers are opposed to the OU to level policy and seek
for more economic and competitive inventory plans. In the context of a deterministic model all
parameters are known at the beginning of the process; thus a vendor can take advantage of the fact
that the he knows the total demand of each stock keeping venue in advance and can transport
quantities in an early stage in order to fulfill the future known demand. However, the amounts that
he is able to transport are bounded by the amounts that are made available to him at each stage. A
branch and cut algorithm was developed to solve the problem exactly. In order to o evaluate the
performance of the algorithm the benchmark instances set for the single vehicle case created by
Arhetti et al. (2007) was used. Computational results have shown that this approach improves the
optimal solution on an average at least 20%.
Acknowledgement
This research has been co-financed by the European Union (European Social Fund-NSF) & Greek
national funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework (NSRF)-Research Funding Program: Heracleitus II.
Investing in knowledge society through the European Social Fund.
References
Adulyasak,Y., J.-F. Cordeau, R.Jans. “Formulations and Branch-and-Cut Algorithms for
Multivehicle Production and Inventory Routing Problems.” INFORMS Journal on
Computing Vol. 26 (1), 2014,pp. 103 – 120.
Aksen D., O. Kaya, F. Salman, and Y. Akça. “Selective and periodic inventory routing problem
for waste vegetable oil collection.” Optimization Letters, Vol. 6(6), 2012, pp. 1063–1080.
Andersson H., A. Hoff, M. Christiansen, G. Hasle, and A. Løkketangen. “Industrial aspects and
literature survey: Combined inventory management and routing.” Computers & Operations
Research, Vol. 37(9), 2010,pp.1515–1536.
Archetti C., L. Bertazzi, G. Laporte, and M. G. Speranza). “A branch-and-cut algorithm for a
vendor-managed inventory-routing problem.” Transportation Science, Vol. 41(3), 2007,pp.
382–391.
Bell W. J., L. M. Dalberto, M. L. Fisher, A. J. Greenfield, R. Jaikumar, P. Kedia, R. G. Mack, and
P. J. Prutzman. “Improving the distribution of industrial gases with an on-line computerized
routing and scheduling optimizer.” Interfaces, Vol. 13(6), 1983, pp.4–23.
Bertazzi L., G. Paletta, and M. G. Speranza,. “Deterministic order-up-to level policies in an
inventory routing problem.” Transportation Science, Vol. 36(1), 2002, pp.119 – 132.
Carøe C. C., and J.Tind,. “L- shaped decomposition of two – stage stochastic programs with integer
recourse.” Mathematical Programming, Vol. 83, (1-3), 1998, pp. 451 -464.
Coelho L. C. and G. Laporte. “The exact solution of several classes of inventory routing
problems.” Computers & Operations Research, Vol. 40(2), 2013, pp.558–565.
Coelho L. C. and G. Laporte. “An Optimized Target Level Inventory Replenishment Policy for the
VMI Systems.” 2013, Technical Report CIRRELT 2013-05, Montreal, Canada.
Book of Proceedings
216
Coelho L. C., J.-F. Cordeau, and G. Laporte. “Thirty years of Inventory Routing.” Transportation
Science, Vol. 48(1), 2014, pp.1-19.
Solyalı O.and H. Süral. “A branch-and-cut algorithm using a strong formulation and an a priori
tour based heuristic for an inventory-routing problem.” Transportation Science, Vol. 45(3),
2011, pp.335–345.
Book of Proceedings
217
Open Governmental data sources in Europe: A comparative evaluation of
semantic and technical characteristics
Olga Vasileiou
National Technical lUniversity of Athens
Heroon Polytechniou 9, 15780, Zografou, Greece
Charilaos Georgis
London School of Economics and Political Science
Michael Petychakis
Decision Support Systems Laboratory, NTUA
Heroon Polytechniou 9, 15780, Zografou, Greece
Spiros Mouzakitis
Decision Support Systems Laboratory, NTUA
Heroon Polytechniou 9, 15780, Zografou, Greece, email: [email protected]
Dimitris Askounis
Decision Support Systems Laboratory, NTUA
Heroon Polytechniou 9, 15780, Zografou, Greece
Abstract
This paper aims to summarize a comparative study of end-user services and technical
characteristics of current Open Governmental data sources in Europe. The analysis has been
performed in the context of the ENGAGE FP7 e-Infrastructures Project. The research began with
a snapshot of the current situation of open data sources in Europe. Thereafter, we managed to
collect, categorize, statistically analyze and comparatively assess the open government data
throughout the European Union. The results of our study show that there is still no uniform policy
regarding the provision of public sector information across data sources in the countries of the
European Union. The quality of the government data sources varies significantly depending on the
country and the data provider. In general, the majority of the datasets is not completely open, as it
has been published under restricted or non-specified licenses. Nevertheless, in the recent years
there is an increasing effort in the adoption of open licenses; especially in the newly launched
national open data portals. The ENGAGE project will constitute the initial test-bed of the results
of the work performed.
Keywords: Open Data, Infrastructures, governmental data
Book of Proceedings
218
1. Introduction
Public sector information (PSI) is the single largest source of information in Europe. It is produced
and collected by public bodies and includes digital maps, meteorological, legal, traffic, financial,
economic and other data. Most of this raw data could be re-used to new products and services,
such as car navigation systems, weather forecasts, financial and insurance services. Since the last
decade there has been an ever-increasing number of open data initiatives and portals. The European
Commission (EC) and national governments all over the world have established of national open
data portals in order to increase public access to high value, machine readable datasets generated
by public agencies and organizations. In 2003, the EU adopted the Directive on the re-use of
public sector information (PSI Directive) that introduced a common legislative framework
regulating how public sector bodies should make their information available for re-use in order to
remove barriers such as discriminatory practices, monopoly markets and a lack of transparency.
During the last years, a vast number of PSI portals and open data communities that develop new
ideas and apps have emerged. For the first time in decades the importance of government openness
has received attention. Previously, government openness was regarded as a passive provision of
information to citizens, while nowadays more proactive approaches of data handling in open
formats are preferred (Zuiderwijk, Helbig, Gil-García, & Janssen, 2014; Zuiderwijk & Janssen,
2014). More and more governmental organizations are releasing their data, as can be seen for
instance in the United States of America and the United Kingdom. Open government data are
expected to bring many advantages. The literature shows that in general potential advantages of
big open data can be, for instance, political and social, economic, technical and operational (M.
Janssen, Charalabidis, & Zuiderwijk, 2012). Open data provide the potential to unlock business
innovation and financial performance (European_Commission, 2011a; K. Janssen, 2011; Jetzek,
Avital, & Bjørn-Andersen, 2012; Kundra, 2012; Yang & Kankanhalli, 2013). As confirmed by
various studies, proactive release of public and private data may create considerable benefits for
businesses, researchers, citizens and other stakeholders.
The proliferation of such Open Government Data initiatives and particularly Open Government
Data portals during the recent years, however, has raised significant questions in terms of the
variety in the interoperability, standards, services, openness and vision of each portal. Both
systematic analysis of these portals and comprehensive analysis of the capabilities and potential
of these initiatives are hardly available or generally missing from the recent research literature. In
order to address this gap, current paper presents our initial findings from the analysis and potential
of the European Open Data portals.
2. Methodology
More specifically, we focused on the countries of the European Union, where we intended to
identify irregularities and challenges with regard to the provision of Open Data across the studied
countries. The research began with a snapshot of the current situation of open data sources in
Book of Proceedings
219
Europe. Thereafter, we managed to collect, categorize, statistically analyze and comparatively
assess the open government data throughout the European Union.
Within the current study, a representative number of diverse and distributed open government data
sources from all the countries of the European Union were analyzed. Data was collected by means
of an online research for organizations and public offices of each country that provide open public
data in electronic form, as well as data aggregators. The collected datasets were classified and
analyzed by genre and country in view of enabling a qualitative comparison between the EU
member states.
Our research mainly focused on a central government level, meaning that for each of the 27 EU
countries we reviewed every working ministry website with publicly available datasets. Our
research was also extended to other central government-related websites, such as national and
regional open data portals, public services, national statistical offices, central banks, national
geodata-related websites as well as each country’s official police, fire service and army website.
Furthermore, besides national public datasets, we also included official European Union portals
and websites.
In addition to this research, which targeted at covering an in-breadth analysis, we also conducted
in-depth research for 3 representative countries of the European Union that have exhibited a
proliferation of open data initiatives during the last years. For that purpose, we chose the United
Kingdom (being a pioneer about open data amongst European countries), France (being a
representative country of mainland Western Europe) and Greece (being a country of Eastern
Europe).
With regard to this research:
The United Kingdom demonstrates great differentiation among local administration
structures. Unlike other European countries, there is a great number of local administrative
sections, a tradition holding since the Middle-Ages . In terms of this research, counties,
boroughs and unitary authorities, were regarded as equivalent to the municipalities in the
rest of Europe. In summary, we proceeded to the investigation of 98 counties, 61 boroughs,
15 unitary authorities and 60 urban areas (cities/towns).
In France, we investigated 95 departments’ and 11 urban areas’ (cities/towns) data
sources).
In Greece, 13 Regions and 325 Municipalities were investigated, as predefined by the
Kallikrates program of the Greek Ministry of Internal Affairs (Site 4).
We subsequently proceeded to the collection and categorization of the datasets from each data
source researched. The process was initiated on March 2012 and was concluded on January 2013.
In total, 3,466 data portals were analyzed.
Book of Proceedings
220
3. Results
In this section, the results of the analysis per key attributes are presented.
Licenses
One of the most important issues of public data is the definition of the license related to it and
whether it is open or not. The results showed that the vast majority of the datasets (3,045 or 87.88%
of all of the datasets) are published without a clearly defined or open license (License Not Specified
or Restricted) while in the case of 64 of them (1.85%) the original providers held the license. On
the contrary, only 356 datasets (10.27%) were published with an open license, with most of them
under either the UK Open Government License or the Creative Commons Attribution License. It
is evident that one of the most crucial steps into making data truly open – apply an open license
that will support the openness of data- is now lacking in Europe, with the bright exception of the
emerging national open data portals.
Interface and Data Languages
Another point of interest in an ever-diverse and multilingual set of countries such as the European
Union is the existence of multilingual support in the data that each provider publishes. More
specifically, it was deemed important to separate the case of languages that the user interface,
through which the data is provided, is available in from the case of the languages of the data itself.
The results of the analysis showed that nearly half of the web interfaces (47.39%) support only the
native language of the country they belong to. Additionally, a large percentage of the interfaces
(38.18%) is accessible by an additional language, (mainly in English), while 7.07% of them
support 3 languages and 7.36% 4 languages or more. The percentages are quite lower with regard
to the language of the actual provided datasets. The vast amount (77.80%) of the datasets is
available only in their native language. As a result, only 22.19% are available in 2 languages or
more, contrary to the UIs where 52.50% of them provide multilingual support. Thus, it is now
obvious that the separation of UI and data languages was the only way to showcase the differences
between them and to prove that while the former has now reached an acceptable level of
multilingual support, the latter remains lacking in that aspect. The most common language used is
English.
Data Provision
This attribute refers to utilities that serve the purpose of providing the data / information to users
and applications, either in a human readable format or a machine-processable format. For instance,
this attribute indicates whether the data is available for online view only, via a downloadable file
or both, as well as the existence of value-added services (charts, maps, APIs). Our analysis
Book of Proceedings
221
revealed in absolute numbers the different ways the data is provided in EU websites. According to
the analysis, 2542 datasets are provided in downloadable form (73.34%), 1751 datasets are
provided through an online view (50.52%), 328 datasets through a map service (9.46%), 293
datasets through charting capabilities (8.54%) and only 37 through an API (1.07%). The above
analysis was repeated specifically for the national open data portals in order to highlight the
difference in data provision between them and the rest of the public data providers. According to
this analysis, in 5 out of the 14 at the time available open government data portals, the user can
access datasets through an API; a percentage (35.71%) much higher than the average 1.08%. This
fact clearly demonstrates the technical maturity of the new open data portals in contrast to the
legacy websites of ministries, municipalities and public agencies.
Data Formats
The available data representation formats of the published information are one of the key features
of open government portals, as this defines the inherent properties of the datasets, their usability
and interoperability. In our analysis we portrayed the most prominent data formats encountered in
all of the 27 European countries: The largest percentage of the datasets is stored in the PDF (38%)
and in the HTML format (28%). In addition, 14% of the datasets are available through RSS feeds,
8% in the XLS/XLSX formats, 7% in the DOC/DOCX formats, 4% in the CSV format and 1% in
the KML format. According to the results we clearly conclude that the majority of the data sources
do not provide datasets in machine processable-formats that can be directly consumed through
applications, thus limiting their usability significantly.
The analysis was repeated across the EU countries’ so as to reach a conclusion about the status of
each country in each category under investigation.
The countries which have made the greatest progress are the Netherlands (56% of the data sources
contain machine processable datasets), Bulgaria (51.61%), Spain (51.45%), United Kingdom
(48.95%) and Hungary (41.15%). On the contrary, the countries that are lagging the most are
Slovenia (4.18%), Luxembourg (7.46%), Malta (11.95%), Slovakia (15.85%), Germany (23.71%)
and Cyprus (24%).
Our analysis also illustrates the findings particularly for local administration data sources in the
UK, France and Greece where notably few differences were found compared to their national
average. In the UK, 43.56% of the data sources contain machine processable datasets (compared
to 48.95% of the UK average) and 56.44% contain non machine processable datasets. Meanwhile,
in France 31.06% of the data sources provide machine processable datasets (compared to 32.67%
of the French average) while 68.94% do not. Finally, in Greece 25.67% of the datasets are machine
processable (compared to 27.33% of the Greek average) and 74.33% are not. Similarly to open
licenses and multilingual support, it is apparent that local administration data infrastructure in each
country is lacking compared to that of the central administration.
Book of Proceedings
222
4. Conclusions
The aim of this paper was to investigate and provide an insight into the current public data
infrastructures in the European Union. For this purpose, our research included both central and
local government data sources as well as official EU portals and websites in order to ensure that
the data sources investigated complement each other and provide the full picture of the current
public data landscape. The results of our study show that there is still no uniform policy regarding
the provision of public sector information across data sources in the countries of the European
Union. The quality of the government data sources varies significantly depending on the country
and the data provider. In general, the majority of the datasets is not completely open, as it has been
published under restricted or non-specified licenses. Nevertheless, in the recent years there is an
increasing effort in the adoption of open licenses, especially in the newly launched national open
data portals.
In terms of multilingual support, only 22% of the actual datasets are available in more than one
language, whereas in the case of user-interfaces of the data portals (static website text), 52% of
them support multiple languages. The discrepancy between the two cases is expected, given the
fact that the task of translating the rapidly growing volume of information published by each data
provider in more languages other than the original is challenging. Hence, there is a notable
difficulty faced by researchers or citizens to access and utilize foreign datasets. Moreover, most
data portals provide the ability to search data only through browsing of categories and simple text
search, rarely supporting semantic search – with the bright exception of open government data
portals. Thus, the low percentage of SPARQL and CKAN searches as well as the small number of
cases where data is provided through an API, clearly indicates that currently there is low usage of
Linked Data and Semantic Web technologies. Furthermore, most datasets are published in non-
machine processable formats, rendering their technical re- use demanding. This is evident in the
case of ministry and other public administration websites, where a simple publication of the data
is sufficient, as opposed to the national portals where the availability of technically re-usable
formats and semantic interoperability is also a concern.
Despite these shortcomings, it should be noted that the quality of open government infrastructures
is steadily improving. Particularly, throughout the EU there is an ever-growing trend of countries,
cities and regions towards launching official open data portals where data is published under
universally open standards. The United Kingdom was found to be the leading country in that trend,
but also France, Austria, Italy, Spain, Germany, Belgium, Portugal, Estonia and the Netherlands
have launched their own national and regional open data portals. It is expected that even more
countries are going to adopt open government policies bringing together considerable
advancements in the following years.
Book of Proceedings
223
Acknowledgement
This work has been partly funded by the European Commission through the Project ENGAGE (An
Infrastructure for Open, Linked Governmental Data Provision towards Research Communities and
Citizens).
References
Book
ENGAGE, 2012. Deliverable 7.7.1 Analysis Report of Public Sector Data and Knowledge Sources
United Nations Data, UN statistical databases, http://data.un.org/
Papers
J. C. Bertot, P. T. Jaeger, S. Munson and, T. Glaisyer, Social media technology and government
transparency, Computer, vol. 43, no 11, pp. 53-59, 2010.
A. Burton, D. Groenewegen, C. Love, A. Treloar and, R. Wilkinson, Making research data
available in Australia, Intelligent Systems, IEEE, vol. 27, no. 3, pp. 40-43, 2012.
Carte de France [online], Available at: http://www.cartesfrance.fr/carte-france-departement/carte-
france-departements.html
A. Cordella and F. Iannacci, Information systems in the public sector: The e-Government
enactment framework, The Journal of Strategic Information Systems, vol. 19, no. 1, pp. 52–66,
2010.
L. Ding, T. Lebo, J. S. Erickson, D. DiFranzo, G. T. Williams, X. Li, J. Michaelis, A. Graves, J.
G. Zheng, Z. Shangguan, J. Flores and, J. A. H. Deborah L. McGuinness, TWC LOGD: A portal
for linked open government data ecosystems, Web Semantics: Science, Services and Agents on
the World Wide Web, vol. 9, no. 3, p. 325–333, 2011.
European Commission. (2012, October) Directive 2003/9 8/EC of parliament and council on the
re-use of public sector information. European Commision. [Online]. Available:
http://ec.europa.eu/information_society/policy/psi/docs/pdfs/directive/psi_dire ctive_en.pdf
M. B. Gurstein, Open data: Empowering the empowered or effective data use for everyone?, First
Monday, vol. 16, no. 2, pp. 2-7, 2011.
B. Hogge. (2010, May) Transparency accountability initiative, Open data study, 2010. [Online].
Available:http://www.soros.org/initiatives/information/focus/communication/articles_publicat
ions/publications/open-data-study-20100519
Book of Proceedings
224
A branch and price solution algorithm for the tail assignment problem
George Kozanidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece, email: [email protected]
Elina Gioti
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Abstract
We consider the tail assignment problem, i.e., the problem of assigning a set of passenger flights,
which must be carried out by a commercial fleet, to specific aircraft. Each flight has a fixed
departure time, as well as an origin and a destination airport, which, combined, determine the
duration of the associated trip. The aim is to schedule all the flights, while also minimizing the
number of aircraft utilized for this purpose. Motivated from theory that has been developed in the
past, we develop an integer programming formulation and a branch and price solution algorithm
for this problem.
Keywords: aircraft scheduling, integer programming, column generation, branch & price.
1. Introduction
We consider the tail assignment problem, i.e., the problem of assigning a set of passenger flights,
which must be carried out by a commercial fleet, to specific aircraft. Each flight has a fixed
departure time, as well as an origin and a destination airport, which, combined, determine the
duration of the associated trip. The aim is to schedule all the flights, while also minimizing the
number of aircraft utilized for this purpose. Motivated from theory that has been developed in the
past (Grönkvist, 2005), we develop an integer programming formulation and a branch and price
solution algorithm for this problem.
The proposed methodology utilizes a master problem that tries to schedule the maximum
possible number of flights using a set of aircraft-routes, and a column generation (colgen) sub-
problem that generates cost-effective aircraft-routes which are fed into the master problem. Due
to the huge number of alternative aircraft-routes, the master problem minimizes the number of
aircraft utilized to operate the flights, using only a small subset of these routes. At each iteration,
the column generation sub-problem uses dual information obtained from the optimal solution to
the master problem’s linear relaxation, in order to generate the most cost-effective (the one with
the minimum reduced cost) aircraft-route, out of those that have not been considered yet. This
route is then added to the master problem. The optimal solution of the column generation sub-
problem is obtained with an efficient network optimization solution algorithm, which outperforms
Book of Proceedings
225
existing commercial optimization software packages that can be utilized for the same purpose
alternatively.
The procedure continues similarly, until the optimal solution to the linear relaxation of the
master problem is obtained. This happens when no other aircraft-route with negative reduced cost
can be identified. In order to eliminate the non-integralities, the algorithm creates next a branch
and price tree by branching on the fractional decision variables of this solution. Additional decision
variables representing aircraft-routes are generated during this phase, due to the integrality
restrictions that are gradually introduced.
The remainder of this work is structured as follows. In Section 2, we introduce the two
optimization models that the proposed methodology utilizes. In Section 3, we develop the proposed
solution algorithm, and finally, in Section 4 we summarize the present work and we point to
promising directions for future research.
2. Problem Formulation
In this section, we present the model formulation of the master problem and that of its column
generation counterpart. Both these models utilize the following two common sets:
I : set of aircraft,
S : set of flights.
Additional notation specific to each of the two formulations is defined in each corresponding part.
2.1 Master problem formulation
For the formulation of the master problem, we introduce the following mathematical notation:
Sets:
Ri : set of routes of aircraft i,
Parameters:
f : cost for each aircraft utilized,
h : cost for each flight that remains uncovered,
aijs : binary parameter that takes the value 1 if route j of aircraft i covers flight s, and 0
otherwise, iI, jRi, sS,
Decision Variables:
xij : binary decision variable that takes the value 1 if route j of aircraft i is scheduled, and 0
otherwise, iI, jRi,
ys : binary decision variable that takes the value 1 if flight s remains uncovered, and 0
otherwise, sS.
Utilizing this notation, the master problem is formulated as follows:
Min +i
ij s
i I j R s S
fx hy
(19)
s.t. 1, i
ij
j R
x i I
(20)
Book of Proceedings
226
1, i
s ijs ij
i I j R
y a x s S
(21)
, binary, , ,ij sx y i j s (22)
The objective function (1) minimizes the total cost, which comprises of the aircraft utilization
cost and the cost of the uncovered flights. Cost coefficient h is always much larger than cost
coefficient f, imposing the relative priority between the two objectives. Constraint set (2) ensures
that at most one route is scheduled for each aircraft. In conjunction with the objective coefficient
of variables xij, a cost equal to f is imposed this way, for each aircraft utilized. We call these
constraints the aircraft-rows. Constraint set (3) states that each shipment, s, is either covered by
exactly one aircraft-route, or remains uncovered (variable ys is equal to 1 instead), in which case
the corresponding penalty h is imposed in the objective. We call these constraints the flight-rows.
Finally, constraint set (4) restricts the decision variables of the problem to binary values.
Clearly, any two distinct flights covered by the same aircraft-route must be temporally non-
overlapping. Additionally, economic efficiency reasons dictate that an aircraft is not allowed to
travel empty. This implies that for any pair of consecutive flights covered by the same aircraft, the
arrival airport of the preceding one must coincide with the departure airport of the succeeding one.
These, as well as several other rules that the generated aircraft-routes must abide by, are
incorporated into the colgen sub-problem formulation which is presented next.
2.2 Column generation sub-problem formulation
The routes that are candidate to enter the master problem are ranked in terms of their reduced-cost
with respect to the optimal solution of the current master LP relaxation. The aim of the colgen sub-
problem is to identify the aircraft-route with the minimum reduced cost. If this reduced-cost is
negative, this is an indication that the associated aircraft-route has the potential to improve this
solution; therefore, it is added to the master problem, and the new optimal master LP dual solution
is updated. If not, this implies that no other aircraft-route can improve the optimal solution of the
current master LP relaxation; therefore, the column generation procedure terminates.
Let dls, als, dts and ats be the departure airport, the arrival airport, the departure time, and the
arrival time of flight s, respectively. For each flight sS, we define two sets, as explained next. Ns
is the set of flights which are next-compatible with shipment s, while Ps is the set of flights which
are previous-compatible with flight s. A flight s’ is next-compatible with flight s if als = dls’ and
ats < dts’. A flight s’ is previous-compatible with flight s, if flight s is next-compatible with flight
s’. Let also li be the airport at which aircraft i is currently located. We also define two additional
sets. For each aircraft iI, Fi is the set of flights which are next-compatible with aircraft i, while
for each flight sS, Vs is the set of aircraft which are previous-compatible with flight s. A flight s
is next-compatible with aircraft i if li = dls. An aircraft i is previous-compatible with flight s if s is
next-compatible with i.
For the formulation of the colgen sub-problem, we consider a network N = {A,V}. The set of
vertices, V, includes one node for each aircraft, one node for each flight, as well as one node E,
which acts as the terminal node. This latter node is fictitious because the duty of each aircraft stops
upon completion of the last flight. The set of arcs, A, comprises of the edges which connect aircraft
Book of Proceedings
227
nodes with nodes that correspond to next-compatible flights for the associated aircraft, the edges
which connect pairs of nodes that correspond to compatible flights, as well as one edge for each
flight that connects the node that corresponds to this flight with the terminal node. The aim of the
colgen sub-problem is to identify the longest (minimum negative-distance, to be precise) path in
this network that begins in one of the aircraft nodes, visits at least one flight node, and ends in the
terminal node. The length of any path is equal to –f+ci s
s C
d
, where i is the index of the aircraft
associated with the node this path begins from, C S is the set of flight nodes that this path visits,
and ci/ds is the dual variable of the corresponding aircraft/flight row in the current master LP
optimal solution. Since the cost of any aircraft route in the master problem is equal to f, this is
equivalent to finding the aircraft-route with the minimum reduced-cost. With these in mind, we
introduce the following additional mathematical notation for the colgen sub-problem:
Sets:
Fi : set of flights which are next-compatible with aircraft i, iI,
Vs : set of aircraft which are previous-compatible with flight s, sS,
Ns : set of flights which are next-compatible with flight s, sS,
Ps : set of flights which are previous-compatible with flight s, sS,
Parameters:
ci : dual value of aircraft row i in current master LP optimal solution, iI,
ds : dual value of flight row s in current master LP optimal solution, sS,
Decision Variables:
zi : binary decision variable that takes the value 1 if the generated route utilizes aircraft i, and
0 otherwise, iI,
xis : binary decision variable that takes the value 1 if the generated route includes a direct travel
from aircraft-node i to flight-node s, and 0 otherwise, iI, sFi,
xst : binary decision variable that takes the value 1 if the generated route includes a direct travel
from flight-node s to node t, and 0 otherwise, where t is either a flight node, or the terminal node,
sS, t { }sN E ,
ys : binary decision variable that takes the value 1 if the generated route covers flight s, and 0
otherwise, sS.
Utilizing the above notation, the colgen sub-problem is formulated as follows:
Min i i s s
i I s S
f c z d y
(23)
s.t. 1i
i I
z
(24)
, i
i is
s F
z x i I
(25)
{ }
, s s s
is rs st
i V r P t N E
x x x s S
(26)
, s s
s is rs
i V r P
y x x s S
(27)
Book of Proceedings
228
, , , , binary, , , ,i is rs st sz x x x y i s r t (28)
The objective function (5) minimizes the reduced-cost of the aircraft-route that will be
identified, which is equal to the fixed cost, f, of each aircraft-route, minus the dual of the aircraft
this route pertains to, minus the sum of the duals of the flights that this route covers. Constraint
(6) ensures that this route utilizes exactly one aircraft. Constraint set (7) states that the selected
aircraft should visit a node that corresponds to one of its next-compatible flights first. Constraint
set (8) ensures the flow balance at each flight-node. Incoming flow can originate either at an
aircraft-node or at a flight-node, while outgoing flow can be directed either to a flight-node or to
the terminal node. Constraint (9) states that a flight is covered if and only if there is incoming flow
in the corresponding flight-node. Finally, constraint set (10) imposes integrality on the decision
variables.
3. Solution Methodology
3.1 Solving the master LP relaxation using column generation
Each node of the branch and price tree is associated with a distinct master problem and its
companion colgen sub-problem, which are based upon the two fundamental formulations (1)-(4)
and (5)-(10), respectively. Two distinct master or colgen problems differ from each other only with
respect to the additional constraints that have been added as a result of branching. The optimal
solution of each master LP relaxation is obtained with column generation, according to the logic-
flow shown in Figure 1.
Figure 1: Column generation logic-flow.
3.2 Solving the colgen sub-problem
Instead of using commercial optimization software for solving each colgen sub-problem, it is far
more efficient to utilize a modification of the shortest path algorithm of Desrochers and Soumis
(1988) in order to find the aircraft-route with the minimum reduced-cost, taking advantage of the
Solve restricted master LP problem
Update dual values in objective function of colgen sub-problem
Solve colgen sub-problem to find vehicle-route with minimum reduced-cost
Is the reduced-cost of the identified vehicle-route negative?
Add this vehicle-route
to the master problem
YES
NO
DONE
Book of Proceedings
229
fact that the associated network is acyclic. This algorithm scans the network nodes in topological
order finding possible path extensions for each node through its next-compatible nodes, and
updates their corresponding distances accordingly. For each node, the algorithm stores a label
denoting the best path distance of this node from the source, which is updated accordingly each
time that an improved path length is discovered. Naturally, since the complexity of the algorithm
is linear in the number of network arcs, its performance is significantly superior to that of
commercial optimization software packages that can be alternatively utilized for solving the colgen
sub-problem, which is an integer program. Moreover, the additional constraints that are added as
a result of branching can be incorporated rather easily into this algorithm, making it possible to
find the optimal integer solution of the problem without resorting to generic optimization software
for the solution of colgen.
3.3 Branching
When the optimal solution to the currently explored master LP relaxation is fractional, the
algorithm performs branching in order to continue the search for the optimal integer solution. As
in the case of a typical branch and bound solution algorithm, new sub-problems are created through
the addition of constraints that eliminate fractional solutions. A typical design involves the
selection of one such fractional decision variable for branching, and the partition of the solution
space by setting this variable equal to 0 and 1. Rather than appending the associated branching
constraints to the master problem, we incorporate them directly into the existing formulation
instead. The benefit of doing this is that it retains the same number of master problem constraints,
and thus we do not have to deal with extra dual variables.
4. Summary and Future Work
In this work, we have presented a solution algorithm for the tail assignment problem, i.e., the
problem of assigning a set of passenger flights to specific aircraft. The proposed methodology
utilizes a branch and price tree, at each node of which a linear problem comprising a suitable
modification of the original formulation is solved with column generation. When the problem is
too large and cannot be handled efficiently, the user must inevitably compromise for a near-optimal
solution instead of the exactly optimal one. The most common technique for achieving this is the
incorporation of tolerances on the optimal objective. When such tolerances are present, the
algorithm does not backtrack to nodes created earlier in the search tree, unless these tolerances are
violated. In any other case, the algorithm continues its dive deeper, which makes it easier to obtain
faster an integer solution. Through the choice of the tolerance values and the relaxation bounds on
the optimal objective, the user can control how close this solution will be to the truly optimal one,
and may select to interrupt the execution of the algorithm before its termination if the quality of
the best integer solution that has been found so far is acceptable. Future research should be directed
towards the development of similar enhancements that can be incorporated into the design of the
proposed solution algorithm, in order to enable the more efficient treatment of large scale
problems.
Book of Proceedings
230
References
Desrochers, M., and F. Soumis. “A Generalized Permanent Labelling Algorithm for the Shortest
Path Problem with Time Windows”. INFOR, Vol. 26(3), 1988, pp. 191–212.
M. Grönkvist. “The Tail Assignment Problem”. PhD Dissertation, Göteborg University,
Department of Computer Science and Engineering, 2005.
Book of Proceedings
231
A multi-stage column generation solution approach for the bidline aircrew
scheduling problem
Panagiotis Andrianesis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
George Kozanidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece, email: [email protected]
Abstract
We consider the bidline scheduling problem that typically arises in the commercial airline industry,
i.e., the problem of generating anonymous duty lines, which will be subsequently matched to
specific aircrew members, according to their seniority and preferences. Each duty line consists of
duty and rest periods beginning and ending at the same crew-base, typically called pairings. It must
abide by certain safety and collective agreement rules, and the credited hours (i.e., the hours for
which aircrew members will be paid) it contains must fall within a certain interval. The generated
bidlines must collectively cover a given set of pairings; at the same time, they should also satisfy,
to the greatest extent possible, a given set of quality criteria. Such criteria include purity (the degree
to which the bidline contains routes of the same type), regularity (the degree to which the
duties/offs are repeated in a specific pattern), hour balance (the degree to which the number of
credited hours approaches a desired target), as well as several other criteria related to particular
characteristics (e.g., number of single duties/offs, etc.).
For this problem, we develop an integer programming formulation, and a multi-stage branch and
price solution methodology. The first stage aims at generating a large number of high quality duty
lines that satisfy a certain quality threshold. The second stage aims at generating additional duty
lines so as to cover the flights that remain uncovered at the end of the first stage, without any
special concern for their quality. Finally, the last stage aims at further improving the quality of the
combined solution obtained by the first two stages.
The proposed solution methodology utilizes a master problem that tries to optimize some
appropriate measure of performance using a given set of duty lines, and a column generation sub-
problem that generates cost-effective duty lines which are fed into the master problem. We
describe the steps of the proposed solution algorithm, focusing on novel formulations for the
master problem. We supplement our analysis with numerical results demonstrating the
performance of the algorithm on a medium-sized airline company.
Keywords: aircrew scheduling, bidline, column generation.
Book of Proceedings
232
1. Introduction
The airline crew scheduling problem is one of the most important problems in the airline industry,
due to the fact that the involved cost is the second largest following the fuel cost. The objective of
the problem is to determine schedules (typically over a monthly period) for the crewmembers,
given the flight schedule of one or more aircraft types.
Traditionally, the crew scheduling problem is solved in two steps. Firstly, the crew pairing
problem is solved, which creates sequences of duties and rest periods that start and end at the
crewmember’s base (pairings). Secondly, the crew assignment problem is solved to assign
crewmembers to given pairings. Recently, integrated approaches have appeared in the literature
(Saddoune et al. 2012).
The main approaches that airlines employ for solving the crew assignment problem can be
classified in three categories:
(a) Bidline scheduling, where anonymous monthly schedules are produced, and the
crewmembers bid on them (Boubaker et al., 2010).
(b) Rostering, where personalized schedules are produced that take into account pre-
assignments and preferences.
(c) Preferential bidding, which is similar to rostering, but accounts for seniority, too.
In this paper, we consider the bidline scheduling problem, which is a typical problem
encountered by many US airline companies. The remainder of the paper is structured as follows.
In Section 2, we formulate the bidline problem, and in Section 3, we present the proposed
methodology. Lastly, in Section 4, we present numerical results for a medium-sized airline
company, and provide directions for further research.
2. The bidline problem
Given a set of pairings produced by the solution of the crew pairing problem, the bidline scheduling
problem aims at finding bidlines, which must collectively cover this set of pairings and have the
following characteristics:
(a) The credited hours that the bidline contains must fall within a certain interval (minimum
and maximum credited hours).
(b) The bidline must abide by certain safety and collective agreement rules, e.g., required rest
periods between pairings, minimum number of days off in a monthly or weekly period, etc.
(c) The bidline should satisfy, to the greatest extent possible, a given set of quality criteria,
such as:
(1) “Purity”, i.e., the degree to which the bidline contains pairings of the same type, e.g.,
pairings that contain morning daily flights.
(2) “Regularity”, i.e., the degree to which the duties/offs are repeated in a specific pattern,
e.g., duties from Monday to Thursday, and offs from Friday to Sunday.
(3) Hour balance, i.e., the degree to which the number of the credited hours approaches a
desired target.
(4) Avoid single duties/offs, and/or include a certain number of consecutive duties/offs.
The bidline problem is formulated as a set partitioning problem, as follows:
Book of Proceedings
233
minimize c Mb b pb py x (1)
subject to: ,a 1p b b pby x p (2)
{0,1}by b (3)
0px p (4)
where cb is the cost of bidline b, by is a binary variable, which takes the value 1 if bidline b
participates in the solution, and 0 otherwise, M is a large number which represents the cost of not
covering a pairing p, px is a slack variable which allows for a pairing not being covered subject
to the large cost M, and ,a p b
is a coefficient which indicates if bidline b covers pairing p, i.e., ,a p b
= 1 if bidline b covers pairing p, and 0 otherwise.
3. Methodology
For the bidline problem, as formulated in the previous section, we develop an integer programming
formulation, and a multi-stage branch and price solution methodology. The first stage aims at
generating a large number of high quality duty lines that satisfy a certain quality threshold. The
second stage aims at generating additional duty lines so as to cover the flights that remain
uncovered at the end of the first stage, without any special concern for their quality. Finally, the
last stage aims at further improving the quality of the combined solution obtained by the first two
stages.
The proposed solution methodology utilizes a master problem that tries to optimize some
appropriate measure of performance using a given set of duty lines, and a column generation
pricing sub-problem that generates cost-effective duty lines, which are fed into the master problem.
The column generation methodology is presented in the following figure.
Figure 1. Column generation procedure
The above procedure solves the linear relaxation of the master problem. To obtain an integer
solution, branching is applied (see Barnhart et al., 1998), including special branching rules and
tree-searching strategies. In what follows, we discuss the formulations of the master problem. The
pricing sub-problem is not presented, due to space consideration.
At the first stage, the objective is to generate a large number of high quality duty lines that
satisfy a certain quality threshold; measured in terms of cost, this can be expressed as generating
Solve Restricted Master Problem
Bidline with
negative reduced
cost?
Update and solve
pricing subproblem
Duals
Add new column
YES
NO
OPTIMAL
Book of Proceedings
234
bidlines with cost c Cb (C is the threshold). This objective is achieved by modifying the
objective function in (1) as follows:
minimize c Cb bby (5)
Note that (5) does not include the second term of (1), i.e. M = 0, and that the coefficient c Cb
is negative. Hence, (5) can be viewed as a maximization of C cb bby , where C cb
represents a measure of the “merit” of bidline b. In other words, the objective is to maximize the
cumulative merit of the bidlines. By controlling the value of C compared to the quality penalties
that determine the cost of a bidline, we can ensure a minimum quality threshold for the generated
bidlines.
At the second stage, the objective is to cover as many pairings as possible. To meet this
objective, we solve the original problem (1)-(4) with c = 0b . Since this stage may sacrifice quality
in order to achieve maximum coverage, we choose to fix a certain number of high quality bidlines
obtained at the first stage.
Finally, at the last stage, we revert cb to its actual value, in order to fine tune the quality of the
solution.
4. Numerical Results and Further Research
We applied the proposed methodology on actual instances of a medium-sized US airline company
(~100 cockpit crewmembers, ~700 pairings; sub-problem: ~ 1000 discrete variables, ~10,000
constraints). We used CPLEX 12.4 to solve both the master problem and the pricing sub-problem,
and AIMS proprietary software for modeling and supporting the bidline problem on an Intel Core2
Duo @ 2.4 GHz, with 3 GB RAM, and no parallel processing. The results are shown in the
following Table.
Month April 2014 May 2014 June 2014 July 2014
No. of Pairings 776 684 693 746
Pre-processing 17 min 15 min 16 min 17 min
1st stage 30 min 24 min 20 min 28 min
2nd stage 15 min 13 min 24 min 13 min
3rd stage 48 min 33 min 10 min 28 min
Total time 110 min 85 min 70 min 86 min
Table 1. Numerical Results for a Medium-Sized US Airline Company
In all four instances, all pairings were covered, and with suitable user-acceptable tolerances the
optimality gap was less than 5%. An initial integer solution, with all pairings covered was obtained
within an hour, at the end of the second stage.
Book of Proceedings
235
In our future research, we intend to link the three stages and pass information of the later stages
to the earlier ones. Some preliminary results indicated about 20% reduction in computational time
and about 10% improvement in the quality of the first integer solution this way. We also recognize
that there is significant potential in improving the methodology for producing the bidlines (sub-
problem), as well as improving the tree-search strategies and branching rules.
References
Barnhart, C., Johnson, E.L., Nemhauser, G.L., Savelsbergh, M.W.P., and P.H Vance. “Branch-
and-price: Column generation for solving huge integer programs”. Operations Research, Vol. 46
No. 3, 1998, pp.316–329.
Boubaker, K., Desaulniers, G., and I. Elhallaoui. “Bidline scheduling with equity by heuristic
dynamic constraint aggregation”. Transportations Research Part B, Vol. 44, 2010, pp. 50-61.
Saddoune M., Desaulniers G., Elhallaoui, I., and F. Soumis. “Integrated Airline Crew Pairing and
Crew Assignment by Dynamic Constraint Aggregation”. Transportation Science, Vol. 46, No. 1,
2012, pp. 39-55.
Acknowledgement
The work of the authors is sponsored by the research grant “Development of operations research
tools for the optimization of functional procedures of commercial airlines”, which is funded by
AIMS Corporation.
Book of Proceedings
236
A Calibration Tool for Macroscopic Traffic Flow Models
Anastasia Spiliopoulou*
Dynamic Systems and Simulation Laboratory
Technical University of Crete, 73100 Chania, Greece
Ioannis Papamichail
Dynamic Systems and Simulation Laboratory
Technical University of Crete, 73100 Chania, Greece
Markos Papageorgiou
Dynamic Systems and Simulation Laboratory
Technical University of Crete, 73100 Chania, Greece
John Chrysoulakis
Department of Civil and Infrastructure Engineering
TEI of Athens,12210 Egaleo-Athens, Greece
Abstract
This paper presents a software tool that has been recently developed for the calibration and
validation of macroscopic traffic flow models using real traffic data and appropriate optimization
methods. The software has a user-friendly graphical interface which makes the calibration
procedure an easy task. Apart from the description of the software tool, an application example is
presented as well.
Keywords: Traffic flow models, calibration, validation, calibration tool.
*Corresponding author, email: [email protected]
Book of Proceedings
237
1. Introduction
During the last decades, several mathematical road traffic flow models have been proposed (see
Hoogendoorn and Bovy (2001) for an overview on macroscopic traffic flow models). These
models can be used for planning of new or upgraded road infrastructures, for development and
testing of traffic estimation algorithms, for designing and testing of traffic control strategies, as
well as, for other traffic engineering tasks. The models include a number of physical or non-
physical parameters whose values may differ for different freeway sites. Thus, before employing
a traffic flow model in practice, it is important to first calibrate it against real traffic data.
The calibration procedure aims to appropriately specify the model parameter values, so that the
representation of the network and traffic flow characteristics is as accurate as the model structure
allows. The most common approach is to minimize the discrepancy between the model’s
estimations and the real traffic data, by use of appropriate optimization tools. The nonlinear, non-
convex least-squares optimization problem of parameter estimation is known to have multiple
local minima and hence only derivative-free optimization algorithms should be utilized (see
Kontorinaki et al. (2014) for an overview on suitable optimization methods).
Within the literature there are only few works on the calibration of macroscopic traffic flow
models. The main reasons being: first, it is quite difficult to have access to real traffic data, and
second, there is no available tool that can be easily employed to solve the parameter estimation
problem. Within this work, a software tool has been developed for the calibration and validation
of macroscopic traffic flow models with a user-friendly graphical interface.
In the following sections, first the software tool is shortly described followed by a short application
example using real traffic data from a freeway stretch in Athens, Greece. Finally the last section
concludes with the main remarks of this study.
2. CALISTO graphical user interface
CALISTO (CALIbrationS Tool) is a software tool that enables the calibration and validation of
macroscopic traffic flow models for various freeway sites using real traffic data. Figure 1 presents
the application window of the software which contains the following basic elements:
Freeway network description: this feature includes all the required information needed so
that a freeway site is described adequately, such as the number of freeway links, the number
of freeway on-ramps and off-ramps and their location, the number of detector stations and
their location etc. See Figure 2 for an example of the “Freeway network description” editor.
Simulated traffic data: this feature contains information about the simulated data, such as
the simulation step, the traffic measurements interval and the simulation duration, as well
as the specification of the traffic data input file.
Book of Proceedings
238
Other settings: it consists of some extra simulation features regarding the utilized
performance index and the simulation outputs.
Figure 11 CALISTO application window.
Book of Proceedings
239
Figure 12 Example of “Freeway network description” editor.
Book of Proceedings
240
Figure 13 Example of “METANET parameters” editor.
Selection of the traffic flow model: the user may select one of the available macroscopic
traffic flow models. Two models are available in the current version of the software. In
particular, the first-order Cell Transmission Model (CTM) (Daganzo, 1995a, 1995b) and
the second-order model METANET (Messmer and Papageorgiou, 1990). Both models are
discrete-time state-space models and they are the most commonly used models for the
freeway traffic flow representation. The structure of the program is modular enabling the
addition of more discrete-time state-space models in the future. Depending on the model
that is selected, some model parameter values need to be specified, by clicking on the
“Model parameters” button (see Figure 1 and Figure 3). Note, that in case of calibration
these values are actually the initial model parameter values while in case of validation these
are the values of the resulted model.
Selection of the optimization algorithm: the user may select one of the available
optimization methods. Three methods are available in the current version of the software,
namely, the Nelder-Mead method, a genetic algorithm and the cross-entropy method. All
three optimization methods are derivative-free methods and are suitable for the calibration
of macroscopic traffic flow models (see Spiliopoulou et al. (2014) for an illustration).
Again, the structure of the program is modular enabling the addition of more optimization
algorithms in the future. Depending on the algorithm that is selected, some parameters need
to be specified, by clicking on the “Algorithm parameters” button (see Figure 1).
Book of Proceedings
241
Selection of the operation: two operations are available, either “Calibration” or
“Validation”. The calibration aims at estimating the optimal model parameter values so
that the model may represent the traffic conditions of a particular freeway site with the
highest achievable accuracy. The validation, is usually carried out after the model
calibration, and aims to test the validity of the produced model, thus the resulting model is
applied to the same freeway site using different traffic data than the data used for its
calibration.
Execution: the selected operation is executed, by clicking on the “Run” button (see Figure
1), taking into account all the introduced information.
The output of the calibration procedure includes graphs of the calibration progress, the optimal
model parameter values, the performance index (PI) value and plots of the real traffic data and the
corresponding model estimations at various network locations; while the output of the validation
procedure includes the obtained PI value and plots of the of the real traffic data and the
corresponding model estimations.
3. Application example
This section presents the application of the developed software tool for the calibration of the
METANET model using real traffic data from a part of the Attiki Odos freeway (34th to 28th km,
direction from the Airport to Elefsina) in Athens, Greece. This freeway stretch includes three on-
ramps and three off-ramps, as shown in Figure 4. In order to model the network by use of
METANET, the freeway stretch is represented through 9 nodes (N0−N8) and 8 links (L1−L8),
where each node corresponds to a bifurcation or a junction or any location marking a change of
the network geometry; whereas the homogeneous road stretches between these locations are
represented by links. Each network link is subdivided in model sections of equal length; see for
example link L1 which is divided in 3 sections, with the vertical short lines denoting the section
borders. Figure 4 displays the length, number of sections and number of lanes for each link; the
exact location of the on-ramps and off-ramps; as well as the location of 19 available detector
stations which are depicted by bullets. The METANET model was calibrated using real traffic data
from the morning peak hours, 6 – 12 am, of 16/06/2009.
In this calibration exercise, the Nelder-Mead algorithm is employed to solve the parameter
estimation problem, aiming at minimizing the RMSE (Root Mean Squared Error) of the real-speed
measurements and the model’s estimation of speed.
Book of Proceedings
242
Figure 14 Representation of the considered freeway stretch.
Figure 15 Convergence of the Nelder-Mead algorithm over iterations.
Book of Proceedings
243
Figure 16 Optimal model parameter values.
Book of Proceedings
244
Figure 17 Time-series of the real speed measurements and the model’s estimations of speed at various detector locations.
Figure 5 presents the PI value over iterations. It is seen that the algorithm converges after around
140 iterations achieving a PI value equal to 10.12. At the end of the calibration procedure, a
window with the optimal model parameters appears, as shown in Figure 6. The user may utilize
these values to validate the resulted model using real traffic data from other dates, different than
the data used for its calibration. Moreover, Figure 7 presents the time-series of the real speed
measurements and the corresponding model’s estimations at various network locations. It is
observed that the resulted model is able to reproduce the real traffic conditions of this particular
freeway stretch with sufficient accuracy.
As presented above, this tool enables the calibration of various traffic flow models for different
freeway sites. The software is very easy to use and is expected to be very useful to both researchers
and practitioners.
Book of Proceedings
245
4. Conclusions
This paper presents a software tool that has been recently developed for the calibration and
validation of macroscopic traffic flow models and has a user-friendly graphical interface. An
application example illustrates the easiness of use and the effectiveness of the software enabling
the calibration and validation of various traffic flow models at different freeway sites.
Acknowledgements
This research was co-financed by the European Union (European Social Fund - ESF) and by
national funds through the Operational Program "Education and Lifelong Learning" of the
National Strategic Reference Framework (NSRF) - Research Funded Project: ARCHIMEDES III.
Investing in society’s knowledge through the European Social Fund. The authors would like to
thank ATTIKES DIADROMES S.A. for providing the utilised traffic data from Attiki Odos
motorway in Athens, Greece.
References
Hoogendoorn, S. P., and P. H. Bovy, 2001. State-of-the-art of vehicular traffic flow modelling.
Proceedings of the Institution of Mechanical Engineers, Part I. Journal of Systems and Control
Engineering, Vol. 215, No. 4, pp. 283−303.
Daganzo, C. F., 1995a. A Finite Difference Approximation of the Kinematic Wave Model of
Traffic Flow. Transportation Research Part Β, Vol. 29, No. 4, pp. 261−276.
Daganzo, C. F., 1995b. The Cell Transmission Model, Part II: Network Traffic. Transportation
Research Part Β, Vol. 29, No. 2, pp. 79−93.
Messmer, A., and M. Papageorgiou, 1990. METANET: A macroscopic simulation program for
motorway networks. Traffic Engineering & Control, Vol. 31, No. 8-9, pp.466−470.
Book of Proceedings
246
Kontorinaki, M., Spiliopoulou, A., Papamichail, I., Papageorgiou, M., Tyrinopoulos, Y.,
Chrysoulakis, J., 2014. Overview of nonlinear programming methods suitable for calibration of
traffic flow models. Operational Research: An International Journal. In press.
Spiliopoulou, A., Papamichail, I., Papageorgiou, M., Tyrinopoulos, Y., Chrysoulakis, J., 2014.
Macroscopic traffic flow calibration using different optimization algorithms. Proc. of the
International Symposium of Transport Simulation, Ajaccio, Corsica, France, June 1−4.
Book of Proceedings
247
Air Traffic Management: The free flight concept.
Coletsos John
National Technical University of Athens, School of Applied Mathematical and Physical Sciences
Zografou campus, 15780, Athens, Greece, email: [email protected]
Ntakolia Charis
National Technical University of Athens, School of Applied Mathematical and Physical Sciences
Zografou campus, 15780, Athens, Greece
Abstract
The insufficient air routes combined with the adverse weather and congestion to air sectors lead to
economic, environmental and safety problems to political aviation in Europe. This situation creates
negative aspects to airlines and airports, as well. Furthermore, according to recent studies over
40,000 daily flights are predicted for 2020, and therefore the current ATM system will not be able
to handle this volume of traffic in an efficient manner. A new and promising approach of solving
these problems in the future consists of transforming the ATM system from an ‘airport-centered’
to an ‘airplane-centered’ system so it can: (i) prioritize the airline preferences, (ii) support the free
flight concept, (iii) distribute fairly ground – holding and air delays among the flights, (iv)
minimize the volume of work of ATCs as an observer, (v) relax the existing distance limits
between airplane since the human factor has been annihilated, and therefore, (vi) increase the air
sectors’ capacity avoiding congestions and (vii) increase safety and efficiency. Our attempt will
be to develop a mathematical model for a support system for the free flight concept. We divide the
problem into two sub – problems (upper and lower level) in order to decrease the computational
efforts and the complexity of the air traffic flow management problem and to allow flexibility
between the decision maker levels enforcing in the same time free flight scenario
Keywords: Air-Traffic Management, Integer Programming, Operations Research.
Book of Proceedings
248
1. Introduction
Information technology has already had a major influence on air travel. And with the number of
global travelers expected to double by 2030 according to the Federal Aviation Administration
(FAA), it will continue to lead the way for the industry. However, every day over 100,000 flights
take off at airports across the world. Some are short hops to nearby destinations; some flights cross
the oceans, but all have to fly in the same sky. Until recently, air traffic has been managed by
routing airplane into narrow, predetermined routes – much like highways in the sky – originally
developed to meet the domestic airspace requirements of countries and often defined by the
location of ground-based navigational aids. The above can cause congestion, ground holding and
airborne delays or even worse cancellation. Indeed, Congestion phenomena are persistent and arise
almost on a daily basis as a consequence of bad weather conditions which cause sudden capacity
reductions. The Air Transport Association has estimated that system delays drove an estimated
$5.9 billion in direct operating costs for United States airlines in 2005. As a result, air traffic flow
management (ATFM) has become increasingly crucial.
Free flight is a developing air traffic control method that uses no centralized control (e.g. air traffic
controllers). Instead, parts of airspace are reserved dynamically and automatically in a distributed
way using computer communication to ensure the required separation between aircraft. Free flight
is a new concept being developed to take the place of the current air traffic management methods
through the use of technology. True free flight eliminates the need for Air Traffic Control (ATC)
operators by giving the responsibility to the pilot in command. This gives the pilot the ability to
change trajectory in mid-flight. With the aid of computer systems and/or ATC, pilots will be able
to make more flight path decisions independently. As in most complex systems, distributed yet
cooperative decision making is believed to be more efficient than the centralized control
characterized by the current mode of air traffic management
Therefore, research should be turn to free flight in order to solve these issues and increase
network’s capacity, safety and efficiency. Research should also focus on European airspace, since
en-route airspace, and not only airports, is highly congested. This is due to both the airway system,
built up by a fixed track system connecting airports, and to the existing air navigation and air traffic
control rules. Nowadays, the minimum safe separation between aircraft is assured only by means
of altitude and/or longitudinal separations. This type of structure represents a bottle-neck for air
traffic flow with the increase of flight volume. Though some measures have been taken to reduce
traffic congestion, much more is needed before air traffic can once again flow smoothly and
efficiently. We are interested in the development of all decision methodologies which can make
maximum use of airspace without violating safety constraints.
2. Mathematical model overview
Our philosophy is based on Dell’Olmo and Lulli’s (2002) model, where we divide the problem
into two sub – problems (upper and lower level) in order to decrease the computational efforts and
the complexity of the air traffic flow management problem and to allow flexibility between the
decision maker levels enforcing in the same time free flight scenario (Fig.1).
Book of Proceedings
249
Figure 1 - Mathematical model's structure.
In the upper level, the inputs, such as airspace structure, schedule timetables and the airplane’s
characteristics are integrated in our model. In this way, we represent a fixed topology network,
and it is formulated as a special dynamic multicommodity network flow problem with side
constraints. The upper level’s output is the airplane sequences for each way point and it is the input
of the lower level. The lower level represents some free flight aspects in a constrained scenario, as
it is analytically mentioned in (Dell'Olmo and Lulli 2002). It is formulated as an optimization
model. The output of this level is the 4D – trajectories for each airplane. The 4D trajectory of an
aircraft consists of the three spatial dimensions plus time as a fourth dimension.
The novelty of this research is the enhancement to the above model the ability: (i) to segregate the
airborne delay cost from the ground holding cost, (ii) to impose cancellation policies, (iii) to take
into account the arrival and departure capacity of each airport, (iv) to ensure connectivity between
airports for continued flights and (v) to minimize the cost of flight due to airborne delays, ground
holding delays, flight speed and cancellation. The proposed architecture supports the free flight
concept by guaranteeing freedom of movement to all the airplanes.
3. Free flight mathematical model (Lower Level)
In this section, based on the (Dell'Olmo and Lulli 2002), the lower level is presented as a decision
maker that represents the flight operation of the airplane (mainly the navigation operation)
according to the new possible routes, guaranteeing at the same time freedom of movement to all
airplanes in the network (free flight scenario) with respect to conflict measures and safety
assurance. After the process, a 4 – D trajectory is obtained for each airplane.
Sets: 𝒦 is the set of airplanes, 𝒯 = {1,… , 𝑡𝑁} the subset of time periods, 𝒱𝑘 the set of speeds for
airplane 𝑘 and 𝒵𝑘 the set of admissible flight levels for airplane 𝑘.
Input data: 𝑡𝑠𝑘 is the starting time of airplane 𝑘, 𝑐𝑧,𝑣
𝑘 the fuel consumption cost at level 𝑧 (𝑧 ∈ 𝒵𝑘)
and speed 𝑣 (𝑣 ∈ 𝒱𝑘) for airplane 𝑘, 𝑐𝑡𝑟𝑛𝑘 the fuel consumption cost due to turn for airplane 𝑘, 𝑢𝑘
the initial flight level of airplane 𝑘, 𝑆𝑙𝑜𝑛𝑚𝑖𝑛 the minimum longitudinal separation, 𝑆𝑙𝑎𝑡
𝑚𝑖𝑛 the minimum
lateral separation, 𝐷𝑚𝑎𝑥 the maximum route deviation, 𝐵 the airway breadth and 𝜏 the length of
Book of Proceedings
250
the time interval. We consider also 𝜂−𝑘(𝑡) , 𝜂+
𝑘(𝑡) the negative and the positive deviation variable,
respectively, 𝛿−𝑘,𝑙(𝑡) , 𝛿+
𝑘,𝑙(𝑡) the negative and the positive ordinate deviation variable of airplane
𝑘 and 𝑙, respectively and 𝜃−𝑘,𝑙(𝑡) , 𝜃+
𝑘,𝑙(𝑡) ∶ the negative and the positive abscissa deviation variable
of airplane 𝑘 and 𝑙, respectively. In addition we define the decision variables 𝑥𝑘(𝑡) the abscissa of
airplane 𝑘 at time 𝑡 and 𝑦𝑘(𝑡) the ordinate of airplane 𝑘 at time 𝑡.
𝜑𝑧,𝑣𝑘 (𝑡) = {
1 , 𝑖𝑓 𝑎𝑖𝑟𝑝𝑙𝑎𝑛𝑒 𝑘 𝑖𝑠 𝑎𝑡 𝑓𝑙𝑖𝑔ℎ𝑡 𝑙𝑒𝑣𝑒𝑙 𝑧 𝑤𝑖𝑡ℎ 𝑣𝑒𝑙𝑜𝑐𝑖𝑡𝑦 𝑣 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡0, 𝑜𝑡ℎ𝑒𝑟𝑤𝑖𝑠𝑒
∀𝑘 ∈ 𝒦, ∀𝑧 ∈ 𝒵𝑘, 𝑣 ∈ 𝒱𝑘,
𝜎𝑧𝑘,𝑙(𝑡) = {
1 , 𝑖𝑓 𝑎𝑖𝑟𝑝𝑙𝑎𝑛𝑒𝑠 𝑘, 𝑙 ℎ𝑎𝑣𝑒 𝑡𝑜 𝑟𝑒𝑠𝑝𝑒𝑐𝑡 𝑙𝑎𝑡𝑒𝑟𝑎𝑙 𝑠𝑒𝑝𝑎𝑟𝑎𝑡𝑖𝑜𝑛 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡 0, 𝑖𝑓 𝑎𝑖𝑟𝑝𝑙𝑎𝑛𝑒𝑠 𝑘, 𝑙 ℎ𝑎𝑣𝑒 𝑡𝑜 𝑟𝑒𝑠𝑝𝑒𝑐𝑡 𝑙𝑜𝑛𝑔𝑖𝑡𝑢𝑑𝑖𝑛𝑎𝑙 𝑠𝑒𝑝𝑎𝑟𝑎𝑡𝑖𝑜𝑛 𝑎𝑡 𝑡𝑖𝑚𝑒 𝑡
∀𝑘, 𝑙 ∈ 𝒦, ∀𝑧 ∈ 𝒵𝑘 ∩ 𝒵𝑙 ,
Objective function: The objective function minimizes the total cost of the traffic through the
considered airway. It includes the fuel consumption cost term ∑ 𝑐𝑧,𝑣𝑘
𝑧∈𝒵𝑘,𝑣∈𝒱𝑘 𝜑𝑧,𝑣𝑘 (𝑡) and the
increased fuel consumption cost term due to turn 𝑐𝑡𝑟𝑛𝑘 |𝑦𝑘(𝑡 + 1) − 𝑦𝑘(𝑡)|. In order to linearize
this term, we introduce two non-negative variables 𝜂−𝑘(𝑡) , 𝜂+
𝑘(𝑡), for the positive and the negative
deviation of ordinate of 𝑦𝑘(𝑡) of airplane 𝑘 at time 𝑡. Therefore, we set 𝑦𝑘(𝑡 + 1) − 𝑦𝑘(𝑡) =
𝜂+𝑘(𝑡) − 𝜂−
𝑘(𝑡), ∀𝑘 ∈ 𝒦, ∀𝑡 ∈ [𝑡𝑠𝑘 , 𝑡𝑁 − 1]. In case of positive lateral deviation we have 𝜂+
𝑘(𝑡) >
0 and 𝜂−𝑘(𝑡) = 0 . In case of negative lateral deviation: 𝜂+
𝑘(𝑡) = 0 and 𝜂−𝑘(𝑡) > 0 . So, we
substitute |𝑦𝑘(𝑡 + 1) − 𝑦𝑘(𝑡)| = 𝜂+𝑘(𝑡) + 𝜂−
𝑘(𝑡) and the term becomes 𝑐𝑡𝑟𝑛𝑘 (𝜂+
𝑘(𝑡) + 𝜂−𝑘(𝑡)). So
the objective function is the following:
𝑚𝑖𝑛 ∑ { ∑ [𝑐𝑧,𝑣𝑘 𝜑𝑧,𝑣
𝑘 (𝑡)]
𝑧∈𝒵𝑘,𝑣∈𝒱𝑘
+ 𝑐𝑡𝑟𝑛𝑘 (𝜂+
𝑘(𝑡) + 𝜂−𝑘(𝑡))}
𝑘∈𝒦,𝑡∈𝒯
Subject to the following constraints
We have initial conditions ∑ 𝜑𝑢𝑘,𝑣𝑘 (𝑡𝑠
𝑘)𝑣∈𝒱𝑘 = 1 , ∀𝑘 ∈ 𝒦 (10), ∑ 𝜑𝑧,𝑣𝑘 (𝑡𝑠
𝑘)𝑧(≠𝑢𝑘)∈𝒵𝑘,𝑣∈𝒱𝑘 =
0, ∀𝑘 ∈ 𝒦 (11), 𝑥𝑘(𝑡𝑠𝑘) = 0, ∀𝑘 ∈ 𝒦 (12) and 𝑦𝑘(𝑡𝑠
𝑘) = 0, ∀𝑘 ∈ 𝒦 (13). We introduce
representation constraints in order to impose that only one of the variables 𝜑𝑧,𝑣𝑘 (𝑡) is equal to one,
for each airplane 𝑘 and for each time period 𝑡 ≥ 𝑡𝑠𝑘, so ∑ 𝜑𝑧,𝑣
𝑘 (𝑡)𝑧∈𝒵𝑘,𝑣∈𝒱𝑘 = 1, ∀𝑘 ∈ 𝒦, ∀𝑡 ≥
𝑡𝑠𝑘 (14). The continuity constraints link the positions of each airplane at consecutive time periods.
∑ ∑ 𝜑𝑤,𝑣𝑘 (𝑡 + 1)
𝑣∈𝒱𝑘
− ∑ 𝜑𝑧,𝑣𝑘 (𝑡)
𝑣∈𝒱𝑘
≥ 0 ,
𝑧+1
𝑤=𝑧−1
∀𝑘 ∈ 𝒦, ∀𝑧 ∈ 𝒵𝑘, ∀𝑡 ∈ [𝑡𝑠𝑘 , 𝑡𝑁 − 1] (15)
𝑥𝑘(𝑡 + 1) − 𝑥𝑘(𝑡) = 𝜏 ∑ 𝑣𝜑𝑧,𝑣𝑘 (𝑡)
𝑧∈𝒵𝑘,𝑣∈𝒱𝑘
, ∀𝑘 ∈ 𝒦, ∀𝑡 ∈ [𝑡𝑠𝑘 , 𝑡𝑁 − 1] (16)
Book of Proceedings
251
|𝑦𝑘(𝑡 + 1) − 𝑦𝑘(𝑡)| ≤ 𝐷𝑚𝑎𝑥 𝜏 ∑ 𝑣𝜑𝑧,𝑣𝑘 (𝑡)
𝑧∈𝒵𝑘,𝑣∈𝒱𝑘
, ∀𝑘 ∈ 𝒦, ∀𝑡 ∈ [𝑡𝑠𝑘 , 𝑡𝑁 − 1] (17)
𝑙𝑖𝑛𝑒𝑎𝑟𝑖𝑧𝑎𝑡𝑖𝑜𝑛⇒
𝑙𝑖𝑛𝑒𝑎𝑟𝑖𝑧𝑎𝑡𝑖𝑜𝑛⇒ 𝜂+
𝑘(𝑡) + 𝜂−𝑘(𝑡) ≤ 𝐷𝑚𝑎𝑥 𝜏 ∑ 𝑣𝜑𝑧,𝑣
𝑘 (𝑡)
𝑧∈𝒵𝑘,𝑣∈𝒱𝑘
, ∀𝑘 ∈ 𝒦, ∀𝑡 ∈ [𝑡𝑠𝑘 , 𝑡𝑁 − 1] (17𝑙)
The separation constraints must be satisfied in order to assure the minimal separation among
airplanes. Note that for each couple of airplanes which are at the same flight level, at least one of
either longitudinal or lateral separation constraints must be assured.
|𝑦𝑘(𝑡) − 𝑦𝑙(𝑡)| ≥ 𝑆𝑙𝑎𝑡𝑚𝑖𝑛 [∑ 𝜑𝑧,𝑣
𝑘 (𝑡)
𝑣∈𝒱𝑘
+ ∑ 𝜑𝑧,𝑣𝑙 (𝑡)
𝑣∈𝒱𝑙
− 2 + 𝜎𝑧𝑘,𝑙(𝑡)]
∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘 , 𝑡𝑠
𝑙)
, (18)
|𝑥𝑘(𝑡) − 𝑥𝑙(𝑡)| ≥ 𝑆𝑙𝑜𝑛
𝑚𝑖𝑛 [∑ 𝜑𝑧,𝑣𝑘 (𝑡)
𝑣∈𝒱𝑘
+ ∑ 𝜑𝑧,𝑣𝑙 (𝑡)
𝑣∈𝒱𝑙
− 1 − 𝜎𝑧𝑘,𝑙(𝑡)] ,
∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘, 𝑡𝑠
𝑙)
(19)
In order to linearize these constraints, we introduce two non-negative variables 𝛿−𝑘,𝑙(𝑡) , 𝛿+
𝑘,𝑙(𝑡),
the positive and the negative deviation, respectively, of ordinate of 𝑦𝑘(𝑡) of airplane 𝑘 and 𝑦𝑙(𝑡)
of airplane 𝑙 at time 𝑡. Therefore, we set: 𝑦𝑘(𝑡) − 𝑦𝑙(𝑡) = 𝛿−𝑘,𝑙(𝑡) − 𝛿+
𝑘,𝑙(𝑡) , ∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈
𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘 , 𝑡𝑠
𝑙). In case of positive lateral deviation we have 𝛿+𝑘,𝑙(𝑡) > 0 and
𝛿−𝑘,𝑙(𝑡) = 0. On the other hand in case of negative lateral deviation 𝛿+
𝑘,𝑙(𝑡) = 0 and 𝛿−𝑘,𝑙(𝑡) > 0.
So, we substitute |𝑦𝑘(𝑡) − 𝑦𝑙(𝑡)| = 𝛿+𝑘,𝑙(𝑡) + 𝛿−
𝑘,𝑙(𝑡) in the constraint (18). Similarly, for the
constraint (19), we introduce two non-negative variables 𝜃−𝑘,𝑙(𝑡) , 𝜃+
𝑘,𝑙(𝑡), the positive and the
negative deviation, respectively, of abscissa of 𝑥𝑘(𝑡) of airplane 𝑘 and 𝑥𝑙(𝑡) of airplane 𝑙 at time
𝑡. Therefore, we set: 𝑥𝑘(𝑡) − 𝑥𝑙(𝑡) = 𝜃−𝑘,𝑙(𝑡) − 𝜃+
𝑘,𝑙(𝑡) , ∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥
𝑚𝑎𝑥(𝑡𝑠𝑘, 𝑡𝑠
𝑙). In case of positive lateral deviation we have 𝜃+𝑘,𝑙(𝑡) > 0 and 𝜃(𝑡) = 0 . In case of
negative lateral deviation 𝜃+𝑘,𝑙(𝑡) = 0 and 𝜃−
𝑘,𝑙(𝑡) > 0. So, we substitute |𝑥𝑘(𝑡) − 𝑥𝑙(𝑡)| =
𝜃+𝑘,𝑙(𝑡) + 𝜃−
𝑘,𝑙(𝑡) in the constraint (19)
𝛿+𝑘,𝑙(𝑡) + 𝛿−
𝑘,𝑙(𝑡) ≥ 𝑆𝑙𝑎𝑡𝑚𝑖𝑛 [∑ 𝜑𝑧,𝑣
𝑘 (𝑡)
𝑣∈𝒱𝑘
+ ∑ 𝜑𝑧,𝑣𝑙 (𝑡)
𝑣∈𝒱𝑙
− 2 + 𝜎𝑧𝑘,𝑙(𝑡)]
∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘, 𝑡𝑠
𝑙)
, (20)
𝜃+𝑘,𝑙(𝑡) + 𝜃−
𝑘,𝑙(𝑡) ≥ 𝑆𝑙𝑜𝑛𝑚𝑖𝑛 [∑ 𝜑𝑧,𝑣
𝑘 (𝑡)
𝑣∈𝒱𝑘
+ ∑ 𝜑𝑧,𝑣𝑙 (𝑡)
𝑣∈𝒱𝑙
− 1 − 𝜎𝑧𝑘,𝑙(𝑡)] ,
∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘 , 𝑡𝑠
𝑙)
(21)
In the sequence there are track constraints 𝑥𝑘(𝑡) ≥ 0 , ∀𝑘 ∈ 𝒦, ∀𝑡 > 𝑡𝑠𝑘 (22), −𝐵 ≤ 𝑦𝑘(𝑡) ≤
𝐵 , ∀𝑘 ∈ 𝒦, ∀𝑡 > 𝑡𝑠𝑘 (23), non negativity constraints: 𝜂−
𝑘(𝑡), 𝜂+𝑘(𝑡) ≥ 0 , ∀𝑘 ∈ 𝒦 , ∀𝑡 ∈
Book of Proceedings
252
[𝑡𝑠𝑘 , 𝑡𝑁 − 1] (24) and 𝛿−
𝑘,𝑙(𝑡), 𝛿+𝑘,𝑙(𝑡), 𝜃−
𝑘,𝑙(𝑡), 𝜃+𝑘,𝑙(𝑡) ≥ 0 , ∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥
𝑚𝑎𝑥(𝑡𝑠𝑘, 𝑡𝑠
𝑙) (25). Also, we have the binary constraints 𝜑𝑧,𝑣𝑘 (𝑡) ∈ {0,1}, ∀𝑘 ∈ 𝒦, ∀𝑡 ∈ [𝑡𝑠
𝑘 , 𝑡𝑁 −
1], ∀𝑧 ∈ 𝒵𝑘 , ∀𝑣 ∈ 𝒱𝑘 (26),
𝜎𝑧,𝑣𝑘,𝑙(𝑡) ∈ {0,1}, ∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, ∀𝑡 ∈ 𝒯, ∀𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙) , ∀𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠
𝑘, 𝑡𝑠𝑙) (27) and
linearity constraints 𝑦𝑘(𝑡 + 1) = 𝑦𝑘(𝑡) + 𝜂+𝑘(𝑡) − 𝜂−
𝑘(𝑡), ∀𝑘 ∈ 𝒦, ∀𝑡 ∈ [𝑡𝑠𝑘 , 𝑡𝑁 − 1] (28)
𝑦𝑘(𝑡) − 𝑦𝑙(𝑡) = 𝛿−𝑘,𝑙(𝑡) − 𝛿+
𝑘,𝑙(𝑡) , ∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘 , 𝑡𝑠
𝑙) (29)
𝑥𝑘(𝑡) − 𝑥𝑙(𝑡) = 𝜃−𝑘,𝑙(𝑡) − 𝜃+
𝑘,𝑙(𝑡) , ∀𝑘, 𝑙(𝑘 ≠ 𝑙) ∈ 𝒦, 𝑧 ∈ (𝒵𝑘 ∩ 𝒵𝑙), 𝑡 ≥ 𝑚𝑎𝑥(𝑡𝑠𝑘, 𝑡𝑠
𝑙) (30)
This mathematical model, is a contribution to the free flight concept, and must be considered as
part of an integrated decision system, where a batch of models cooperates under one or more global
strategies.
4. Conclusions
Our research’s goal is to develop a dynamic integer optimization mathematical model with two
level hierarchical architecture that support free flight in a constrained scenario that will: (i)
Reassure safety between airplanes, (ii) Minimize airborne and ground holding delays, (iii)
Cancellation costs, (iv) Minimize speed deviations, (v) Distribute fairly the delay among the flight
path, (vi) Take into account airport’s arrival and departure capacity and arc’s capacity as well.
References
Agustín A., Alonso-Ayuso A., Escudero L.F., Pizarro C., 2012. On air traffic flow management
with rerouting. Part I: Deterministic case, European Journal of Operational Research, Vol. 219,
pp. 156–166.
Betsimas D., Stock S., 1998. The air traffic flow management problem with en route capacities,
Operations Research, Vol. 46, pp. 406-422.
Betsimas D., Lulli G., Odoni A., 2008. The air traffic flow management problem: an integer
optimization approach. 13th International Conference, IPCO. 2008, pp. 36-46.
Betsimas D., Lulli G., Odoni A., 2011. An integer optimization approach to large-scale air traffic
optimization flow management, Operations Research, Vol. 59, pp. 211-227.
D'aspremont A., Sohier D., Nilim A., El Ghaoui L., Duong V., 2006. Optimal path planning for
air traffic flow management under stochastic weather and capacity constraints, Proceedings of 4th
IEEE International Conference on Research, Innovation and Vision for the Future, RIVF'06. 2006,
pp. 1-6.
Book of Proceedings
253
Dell'Olmo P, Lulli G., 2002. A new hierarchical architecture for air traffic management :
Optimization of airway capacity in a free flight scenario. European Journal of Operational
Research, Vol. 144, pp. 179-193.
Grignon L., 2002. Analyses of Delay in an Air Traffic System with Weather Uncertainty. PhD
thesis. s.l., University of Washington.
Leal de Matos P., Powell P.L., 2002. Decision support for flight rerouting in Europe, Decision
Support Systems, Vol. 34, pp. 397-412.
Lulli G., Odoni A., 2007. The european air traffic flow management problem, Transportation
Science, Vol. 41, No. 4, pp. 431-443.
Richetta O., Odoni A.R., 1994. Dynamic solution to the ground-holding policy problem in air
traffic control, Transportation Research, Vol. 28A, No. 3, pp. 167-185.
Sheridan, T.B. 2006. Next Generation air transportation systems: human–automation interaction
and organizational risks, [http://www.resilienceengineering.] s.l, Paper Presented at the Second
Symposium on Resilience Engineering.
Soomer M.J., Franx G.J., 2008. Scheduling aircraft landings using airline's preferences, European
Journal of Operational Research, Vol. 190, No. 1, pp.277-291.
Vranas P.B., Bertsimas D.J., Odoni A.R., 1994. Dynamic ground-holding policies for a network
of airports, Transportation Science, Vol. 28, pp. 275-291.
Waslander S.L., Raffard R.L., Tomlin C.J., 2008. Market-based air traffic flow control with
competing airlines. Journal of Guidance Control and Dynamics, Vol. 31, No. 1, pp. 148-161.
Book of Proceedings
254
Comparison of pricing mechanisms in markets with non-convexities
Panagiotis Andrianesis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
George Liberopoulos
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece, email: [email protected]
Abstract
We consider markets that are characterized by non-convexities or indivisibilities, due to the
presence of avoidable costs and minimum supply requirements. The motivation for our work has
been the area of electricity markets, which allow the submission of multi-part bids and take into
account the technical characteristics of the generation units. Such market designs, when operated
under marginal pricing, may lead to market outcomes where truthful bidding results in losses for
some participants. To deal with this highly undesirable prospect, some approaches provide make-
whole payments, or uplifts, as they are often called, whereas others modify the market-clearing
prices to ensure sufficient revenues to the suppliers. In this work, we present and compare revenue-
adequate pricing approaches. These include the Semi-Lagrangean Relaxation and the so-called
“Primal-Dual” approaches for generating efficient revenue-adequate prices. We supplement these
schemes with a newly proposed scheme, which we refer to as Minimum Zero-Sum Uplift (MZU).
To facilitate the comparisons, we apply these schemes on a stylized example that appears in the
literature.
Keywords: non-convexities, electricity market, revenue-adequate pricing.
Book of Proceedings
255
1. Introduction
Electricity markets in which generation units are allowed to submit multi-part bids and which take
into account the technical characteristics of these units are characterized by non-convexities. Such
market designs, when operated under marginal pricing, may result in market outcomes where
truthful bidding results in losses for some participants. To deal with this undesirable prospect,
some approaches provide external payments, or uplifts, as they are often called, to ensure sufficient
revenues to the market participants (O’Neill et al., 2005; Hogan and Ring, 2003; Bjørndal and
Jörnsten, 2008; Gribik et al., 2007; Andrianesis et al. 2013), whereas others ensure sufficient
revenues without the provision of external uplifts (Motto and Galiana, 2002; Galiana et al., 2003;
Araoz and Jörnsten 2011; Ruiz et al., 2012; Van Vyve, 2011).
In this paper, we focus on the latter approaches, which are pure revenue-adequate in that the
prices that they generate guarantee that no supplier incurs losses without the need for additional
external/internal uplifts. We also discuss a new mechanism, referred to as “Minimum-Zero Sum
Uplift”. The remainder of the paper is structured as follows. Section 2 presents the market model
we use for our study and various pricing approaches. Section 3 illustrates the application of the
approaches on a numerical example, and discusses some interesting findings.
2. Market Model and Pricing Approaches
We consider a single-commodity, single-period stylized Unit Commitment and Economic
Dispatch (UCED) problem, where supplier i submits a bid for its marginal cost bi and its startup
cost fi, to an auctioneer. The auctioneer solves a bid/cost minimization problem to obtain the
optimal commitment and dispatch, represented for supplier i by variables zi and qi respectively,
that satisfy a deterministic and inelastic demand d. Supplier i is subject to technical maximum and
minimum constraints denoted by parameters ki for the capacity and mi for the minimum output.
The formulation of the Mixed Integer Linear Programming problem is presented below.
,
mini i
i i i iiz qL b q z f (1)
subject to: ii
q d (2)
i i iq z k i (3)
i i iq z m i (4)
0iq i (5)
{0,1}iz i (6)
Problem (1)-(6) is characterized by non-convexities due to the presence of the fixed costs and
the minimum output requirements. We mark with an asterisk the optimal solution, and we denote
by λ* the marginal cost price, which is equal to the dual variable associated with constraint (2), if
the commitment variables are fixed to their optimal value, so that problem (1)-(6) is transformed
into a Linear Programming problem. In what follows, we present the basic elements of the
Minimum Zero-Sum Uplift, the Semi-Lagrangean Relaxation and the Primal-Dual approaches.
Book of Proceedings
256
2.1 Minimum Zero-Sum Uplift (MZU)
The MZU scheme is based on the idea of maintaining the optimal solution and increasing the
commodity price so that eventually all suppliers who would incur losses under marginal pricing,
break even. The profitable suppliers are allowed to keep the profits that they would make under
marginal cost pricing but are not allowed to gain any additional profits beyond that. This can be
achieved if the extra commodity payments that the profitable suppliers receive as a result of the
price increase are transferred as side-payments to the non-profitable suppliers, in addition to the
extra commodity payments that the latter suppliers also receive as a result of the price increase.
The smallest price at which the non-profitable suppliers break even is such that the total additional
payments that they receive are just enough (hence the term “minimum zero-sum”) to cover their
losses. The MZU price λ is given as follows:
* * *
*min 0, i i i ii
b q z f
d
(7)
2.2 Semi-Lagrangean Relaxation
The Semi-Lagrangean Relaxation (SLR) approach computes a uniform price that produces the
same solution as the original UCED problem while ensuring that no supplier incurs losses. The
formulation of the SLR problem is presented below.
,
min ( )i i
SLR i i i i ii iz qL b q z f q d (8)
subject to: ii
q d (9)
and primal constraints (3) – (6).
The SLR approach consists of solving the dual problem:
*max ( )SLRL
(8)
To find λ, Araoz and Jörnsten (2011) suggested using an iterative algorithm that increases λ in
each iteration and solves the relaxed problem until the objective function reaches the optimal value
of the objective function of the original UCED problem.
2.3 Primal – Dual Approach
Ruiz et al. (2012) proposed a so-called primal-dual (PD) approach for deriving efficient uniform
revenue-adequate prices. This approach consists of: (a) relaxing the integrality constraints of the
MILP problem so that it becomes a (primal) LP problem, (b) deriving the dual LP problem
associated with the primal LP problem, (c) formulating a new LP problem that seeks to minimize
the duality gap of the primal and dual LP problems, subject to both primal and dual constraints,
and (d) adding the integrality constraints back to the problem as well as additional constraints to
ensure that no participant incurs losses. This procedure yields a new Mixed Integer Non-Linear
Programming (MINLP) problem, which is not presented due to space considerations.
Book of Proceedings
257
3. Numerical Results and Discussion
Α common test-bed for evaluating different pricing schemes that deal with non-convexities has
been an example introduced by Scarf (1994). In this paper, we use a modification of this example,
introduced by Gribik et al. (2007). We modeled the pricing approaches using GAMS 24.1.2 and
solved the SLR and MZU schemes with the CPLEX 12.5.1 solver and the PD scheme with
BARON, on an Intel Core i5 at 2.67GHz, with 6GB RAM. Diagram 1 shows the price vs. the
demand level for the aforementioned pricing schemes for a demand granularity of 0.5 units. Note
that all schemes except PD actually use the optimal UCED solution. PD is the only scheme that
allows for different allocations. Diagram 2 shows the percent increase of the total cost under PD
compared to the optimal (minimum) total cost.
Diagram 1. Price vs demand under PD, SLR, MZU schemes (modified Scarf Example)
Diagram 2. Cost increase (%) under the PD scheme vs demand (modified Scarf example)
Diagram 1 indicates that the prices under all pricing schemes are not monotonically increasing
in demand. This is the main effect of the non-convexities. Diagram 2 indicates that the PD scheme
may result in inefficient commitment and dispatch quantities; the cost increase reaches up to about
7%. This effect is due to the fact that the PD scheme exchanges price for cost efficiency, by re-
allocating the quantities, so that the average costs are actually lower than the ones of the optimal
allocation.
Diagram 1 also shows that the SLR scheme exhibits price spikes. The SLR prices yield
competitive prices that are high enough to make the market participants willing to generate the
amounts of electricity scheduled by auctioneer. To achieve this, the SLR scheme may result in
prices that are higher than the ones required to cover the losses.
6
8
10
12
14
16
18
20
0.5
2.5
4.5
6.5
8.5
10
.51
2.5
14
.51
6.5
18
.52
0.5
22
.52
4.5
26
.52
8.5
30
.53
2.5
34
.53
6.5
38
.54
0.5
42
.54
4.5
46
.54
8.5
50
.55
2.5
54
.55
6.5
58
.56
0.5
62
.56
4.5
66
.56
8.5
70
.57
2.5
74
.57
6.5
78
.58
0.5
82
.58
4.5
86
.58
8.5
90
.59
2.5
94
.59
6.5
98
.51
00
.51
02
.51
04
.51
06
.51
08
.51
10
.51
12
.51
14
.51
16
.51
18
.51
20
.51
22
.51
24
.51
26
.51
28
.51
30
.51
32
.51
34
.51
36
.51
38
.51
40
.51
42
.51
44
.51
46
.51
48
.51
50
.51
52
.51
54
.51
56
.51
58
.51
60
.5
Price
DemandPD SLR MZU
0
2
4
6
8
0.5
3.5
6.5
9.5
12
.5
15
.5
18
.5
21
.5
24
.5
27
.5
30
.5
33
.5
36
.5
39
.5
42
.5
45
.5
48
.5
51
.5
54
.5
57
.5
60
.5
63
.5
66
.5
69
.5
72
.5
75
.5
78
.5
81
.5
84
.5
87
.5
90
.5
93
.5
96
.5
99
.5
10
2.5
10
5.5
10
8.5
11
1.5
11
4.5
11
7.5
12
0.5
12
3.5
12
6.5
12
9.5
13
2.5
13
5.5
13
8.5
14
1.5
14
4.5
14
7.5
15
0.5
15
3.5
15
6.5
15
9.5
%
Demand% Cost Increase [PD]
Book of Proceedings
258
Lastly, we note that the prices of the PD and MZU schemes are comparable. The MZU scheme
allows for internal transfers between the suppliers, and the uplifts are zero-sum. Hence, the
profitable suppliers may transfer part of their revenues to the non-profitable ones, which in general
keeps prices low. The PD scheme may yield lower prices than the MZU price, exchanging price
for cost efficiency. In all cases where the PD price is lower than the MZU price, we observe that
the dispatching is less efficient than the optimal one. This is the tradeoff for seeking price
efficiency.
References
Andrianesis, P., Liberopoulos, G., Kozanidis, G., and A. Papalexopoulos. “Recovery mechanisms
in day-ahead electricity markets with non-convexities – Part I: Design and evaluation
methodology”. IEEE Transactions on Power Systems, Vol. 28, No. 2, 2013, pp. 960-968.
Araoz, V. and K. Jörnsten, “Semi-Lagrangean approach for price discovery in markets with non-
convexities”. European Journal of Operational Research, Vol. 214, No. 2, 2011, pp. 411-417.
Bjørndal, M. and K. Jörnsten, “Equilibrium prices supported by dual price functions in markets
with non-convexities”. European Journal of Operational Research, Vol. 190, 2008, pp. 768-789.
Galiana, F.D., Motto, A. L., and F. Bouffard, “Reconciling social welfare, agent profits, and
consumer payments in electricity pools”. IEEE Transactions on Power Systems, Vol. 18, No. 2,
2003, pp. 452-459.
Gribik, P.R., Hogan, W.W. and S.L. Pope. (2007, Dec.). “Market-clearing electricity prices and
energy uplift”. Working Paper, John F. Kennedy School of Government, Harvard University.
Available: http://www.hks.harvard.edu/hepg/Gribik_Hogan_Pope_ Price _Uplift_123107.pdf
Hogan W. W. and B. J. Ring. “On minimum-uplift pricing for electricity markets”. 2003,
unpublished. Available: http://www.hks.harvard.edu/fs/whogan/.
Motto, A. L. and F. D. Galiana. “Equilibrium of auction markets with unit commitment: The need
for augmented pricing”. IEEE Transactions on Power Systems, Vol. 17, No. 3, 2002, pp. 798-805.
O’Neill, R.P., Sotkiewicz, P.M., Hobbs, B.F., Rothkopf, M.H. and W. R. Stewart Jr. “Efficient
market-clearing prices in markets with nonconvexities”. European Journal of Operational
Research, Vol. 164, 2005, pp. 269-285.
Ruiz, C., Conejo, A. J., and S. A. Gabriel. “Pricing non-convexities in an electricity pool”. IEEE
Transactions on Power Systems, Vol. 27, No. 3, 2012, pp. 1334-1342.
Scarf, H.E.. “The allocation of resources in the presence of indivisibilities”. Journal of Economic
Perspectives, Vol. 8, No. 4, 1994, pp. 111-128.
Book of Proceedings
259
Van Vyve, M., “Linear prices for non-convex electricity markets: models and algorithms”, CORE
Discussion Paper 2011/50, Université Catholique de Louvain, Louvain-la-Neuve, Belgium, Oct.
2011.
Book of Proceedings
260
Development of Optimization Models for Addressing Various Decision and
Information Related Issues in Supply Chain Planning
George Liberopoulos
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece, email: [email protected]
Dimitrios G Pandelis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece, email: [email protected]
George Kozanidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece, email: [email protected]
George K.D. Saharidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion
Areos, 38334 Volos, Greece, email: [email protected]
Abstract
We consider a model of a two-stage serial supply chain that processes a single part type. Each
stage has an infinite-capacity raw-parts (RP) buffer, a finite-capacity production facility (PF)
with deterministic production lead time (PLT), and an infinite-capacity finished-parts (FP)
buffer. Stage 2 receives orders from end customers and places orders to stage 1. Stage 1 receives
orders from stage 2 and places orders to an initial supplier with inexhaustible supply of initial
raw parts. Upon receipt of an order, a stage immediately ships the order quantity from its FP
buffer to its customer. The order arrives after a deterministic order lead time (OLT). If there are
not enough parts in the FP buffer to meet the order, an expensive external inexhaustible-supply
subcontractor (S) immediately complements the missing parts of the order. Each stage has
revenue from the parts it sells and incurs inventory holding costs in its RP and FP buffers, as
well as fixed and variable production and order costs. In case it cannot meet all the demand, it
either pays the cost of complementing the order to the subcontractor, or it passes this cost to its
customer. For this model, we formulate several variants of a finite-horizon production-and-order
planning problem. The variants differ with respect to the level of collaboration and information
Book of Proceedings
261
sharing between the two stages. First, we distinguish between the cases where the decisions are
made in a centralized/decentralized way. In the letter case, we further distinguish between the
cases where the decisions are made sequentially/simultaneously and use local/global
information. In a follow up work, we plan to numerically experiment with these variants in order
to quantify the effect of the problem parameters, the type of collaboration, and the level of
information sharing on order and production variability and supply chain profitability.
Keywords: supply chain planning; centralized vs. decentralized decision making; local vs. global
information.
Nomenclature
Facilities
iR : stage- i raw-parts (RP) buffer, 1,2i ; 3R : customer demand source;
iP : stage- i production facility (PF), 1,2i ;
iF : stage- i finished-parts (FP) buffer, 1,2i ; 0F : (inexhaustible-supply) initial raw-parts
buffer;
iS : (inexhaustible-supply) stage- i subcontractor, 1,2i ;
Indices
i : stage index, 1,2i ;
t : period index, 1, ,t T ;
Decision variables ( 1,2i , 1, ,t T );
,i tP : quantity produced by iP in period t ;
,i tX : indicator (binary) variable of ,i tP equal to 0 if
, 0i tP , and 1 if , 0i tP ;
,i tR : inventory in iR at the end of period t ;
,i tF : inventory in iF at the end of period t ;
,i tD : quantity of order placed by iR to 1iF at the end of period t ;
,i tY : indicator (binary) variable of ,i tD equal to 0 if
, 0i tD , and 1 if , 0i tD ;
,i tS : quantity of order placed by iF to iS at the end of period t ;
Book of Proceedings
262
Parameters ( 1,2i , 1, ,t T );
max
,i tP : production capacity of PF iP in period t ;
p
iL : production lead time (number of periods) of PF iP ;
d
iL : order lead time (number of periods) from 1iF to
iR ;
3,tD : (external) final customer orders placed by 3R to
2F at the end of period t ;
iI : interest rate used by stage i to compute inventory holding cost rates;
M : a very large number;
Costs ( 1,2i )
ip : (variable) unit production cost at iP ;
ix : fixed setup cost at iP ;
ir : unit inventory holding cost per period in iR ;
if : unit inventory holding cost per period in iF ;
iI : interest rate used by stage i to compute inventory holding cost rates;
id : (variable) unit order cost from iR to 1iF , 1,2,3i ;
is : (variable) unit order cost from iF to iS ;
iy : fixed order cost from iR to 1iF ;
1. Introduction
The work presented in this paper is part of a project supported by grant MIS 379526
“ODYSSEUS: A holistic approach for managing variability in contemporary global supply chain
networks,” which is co-financed by the EU-ESF and Greek national funds through NSRF –
Operational Program “Education and Lifelong Learning” – “THALES: Reinforcement of the
Interdisciplinary and/or Inter-Institutional Research and Innovation”. The main goal of
ODYSSEUS is to study the phenomenon of supply chain demand variability, identify the
physical points of its creation, analyze its causes, and evaluate its negative impact on supply
chain performance. One of the requirements of ODYSSEUS is to develop quantitative models to
support decisions related to demand variability and in particular the “bullwhip effect” (the
phenomenon that demand variability increases as one moves upstream in the supply chain). The
Book of Proceedings
263
literature on the bullwhip effect is vast. Much of it involves the development and analysis of
stochastic dynamic models of supply chains. Representative examples are Chen et al. (2000a,b),
Cachon and Lariviere (2001), Lee et al. (1997a,b), Alwan et al. (2003), and Zhang (2004).
In this paper, we formulate a deterministic dynamic capacitated lot-sizing planning problem
(Buschkühl, et al. 2010) and variants of it for a simple two-stage serial supply chain model, in
order to study the bullwhip effect. Such problems are simple and fit the practical MRP-
framework (Tempelmeier, 1997). They are also solvable with readily available mathematical
programming software and heuristic approaches (Tempelmeier and Destroff, 1996). In a follow
up work, we plan to use these variants to quantify the effect of the problem parameters, the type
of collaboration, and the level of information sharing on order and production variability and
supply chain profitability. In this respect, our models are related to Saharidis et al. (2006, 2009).
2. Basic Supply Chain Model
Diagram 1 shows a graphical representation of the basic model described in the Abstract.
Triangles represent buffers, and circles represent production facilities. Solid black arrows
indicate the material flow and dashed grey arrows indicate the order flow. The decision variables
of the model are shown in blue color, while its parameters are shown in red color.
Diagram 1. Basic supply chain model.
We make the following assumptions regarding the variable cost rates:
1i i id d p , 1,2i (29)
i i ir I d , 1,2i (30)
( )i i i if I d p , 1,2i (31)
1i is d , 2,3i (32)
Inequalities (29) and (32) are necessary for ensuring the profitability and competitiveness of
stage i , respectively. Expressions (30) and (31) are the usual inventory holding cost
assumptions.
1R 1F1P2R 2F2P
3R0F
1S 2SStage 1 Stage 21,tS
max
,
1
1
,
1,
,t
t tP
X
P
1,tR 2,tR1,tF 2,tF
2,tS
max
,
2
2
,
2,
,t
t tP
X
P
2
dL1
dL 2
pL1
pL
1, 1,,t tY D 2, 2,,t tY D 3,tD
Book of Proceedings
264
We consider a finite-horizon planning problem for the basic model. The horizon is divided into T
discrete time periods, and decisions are made at the end of each period. The final customer
orders of each period are known in advance. The PLTs and OLTs are constant.
In each period, the order of events and decisions is as follows. For 1,2i : 1) iR receives
, 1 dii t L
D
parts from 1iF ; 2)
iP starts processing ,i tP parts which it takes from
iR . 3) iF receives
,pii t L
P
parts from iP . 4)
iF also receives ,i tS parts from
iS . 5) iR orders
,i tD parts from 1iF ,
and 1iF immediately sends these parts to
iR .
Next, we formulate several variants of the finite-horizon planning problem. The variants differ in
terms of the level of collaboration and information sharing between the two stages.
3. Variants of the Planning Problem
A) Centralized Decision Making: The two stages maximize their total profits jointly and
simultaneously subject to customer order requirements and other constraints. This problem can
be formulated as the following MILP problem:
2
1 1, , , , , , , ,
1 1
maxT
i i t i i t i i t i i t i i t i i t i i t i i t
t i
d D y Y d D x X p P s S rR f F
(33)
Subject to , , 1 ,, 1 di
i t i t i ti t LR R D P
, 1,2i
1, ,t T (34)
, , 1 , 1,,pi
i t i t i t i ti t LF F P S D
, 1,2i
1, ,t T (35)
max
, , ,i t i t i tP P X , 1,2i
1, ,t T (36)
, ,i t i tD M Y , 1,2i
1, ,t T (37)
, , , , ,, , , , 0i t i t i t i t i tR F P S D , 1,2i
1, ,t T (38)
, ,, {0,1}i t i tX Y , 1,2i
1, ,t T (39)
B) Decentralized Sequential Decision Making: Stage 2: Leader; Stage 1: Follower. The stages
maximize their individual profits separately and sequentially, starting with stage 2.
Book of Proceedings
265
B.1) Local Information: Stage 1 pays the cost of S1. Stage 2 solves a local-information self-profit
maximization problem and decides, among others, the values of D2,t. Stage 1 takes these values
as given and solves its own local-information self-profit maximization problem.
Stage-2 problem:
3 3, 2 2, 2 2, 2 2, 2 2, 2 2, 2 2, 2 2,
1
maxT
t t t t t t t t
t
d D y Y d D x X p P s S r R f F
(40)
Subject to (34)-(39) for 2i only.
Stage 1 problem:
2 2, 1 1, 1 1, 1 1, 1 1, 1 1, 1 1, 1 1,
1
maxT
t t t t t t t t
t
d D y Y d D x X p P s S r R f F
(41)
Subject to (34)-(39) for 1i only.
B1.2) Global information: Stage 2 pays the cost of S1. It solves a global-information self-profit
maximization problem and decides, among others, the values of D2,t and S1,t. Stage 1 takes these
values as given and solves its own local-information self-profit maximization problem.
Stage-2 problem:
3 3, 2 2, 2 2, 1 2 1, 2 2, 2 2, 2 2, 2 2, 2 2,
1
max ( )T
t t t t t t t t t
t
d D y Y d D s d S x X p P s S r R f F
(42)
Subject to (34)-(39).
Stage-1 problem:
2 2, 1, 1 1, 1 1, 1 1, 1 1, 1 1, 1 1,
1
max ( )T
t t t t t t t t
t
d D S y Y d D x X p P r R f F
(43)
Subject to (34)-(39) for 1i only.
C) Decentralized Sequential Decision Making: Stage 1: Leader; Stage 2: Follower. The stages
maximize their individual profits separately and sequentially, starting with stage 1. Stage 1
decides Y2,t, D2,t, and ΣS2,t and pays y2,tY2,t. Stage 2 plans only its production and detailed supply
from S2, given that ΣS2,t has been decided by stage 1.
Stage-1 problem:
2 2, 1 1, 1 1, 2 2, 1 1, 1 1, 1 1, 1 1, 1 1,
1
maxT
t t t t t t t t t
t
d D y Y d D y Y x X p P s S r R f F
(44)
Subject to (34)-(39) and
2, 3, 2,
1 1
T T
t t t
t t
D D S
(45)
Book of Proceedings
266
Stage-2 problem:
3 3, 2 2, 2 2, 2 2, 2 2, 2 2, 2 2,
1
maxT
t t t t t t t
t
d D d D x X p P s S r R f F
(46)
Subject to (34)-(36) and (38)-(39)
D) Decentralized simultaneous Decision Making: The stages maximize their individual profits
separately and simultaneously. Stage 2 solves the same problem as in variant B.1 and decides,
among others, the values of D2,t. Stage 1 solves a local-information self-profit maximization
problem and decides the selling price 2d that allows it to achieve a desired profit margin β. The
two problems comprise the components of an equilibrium problem.
Stage-2 problem:
3 3, 2 2, 2 2, 2 2, 2 2, 2 2, 2 2, 2 2,
1
maxT
t t t t t t t t
t
d D y Y d D x X p P s S r R f F
(47)
Subject to (34)-(39) for 2i only
Stage-1 problem:
2min d
(48)
Subject to (34)-(39) for 1i only, and
2 2, 1 1, 1 1, 1 1, 1 1, 1 1, 1 1, 1 1,
1
(1 )T
t t t t t t t t
t
d D y Y d D x X p P s S r R f F
(49)
2 0d
(50)
References
Alwan, L.C., J.J. Liu, and D. Q. Yao,. (2003). “Stochastic characterization of upstream demand
processes in a supply chain”. IIE Transactions, Vol. 35, No. 3, pp. 207-219.
Buschkühl, L., F. Sahling, S. Helber, and H. Tempelmeier (2010). “Dynamic capacitated lot-
sizing problems: a classification and review of solution approaches”. OR Spectrum, Vol. 32, pp.
231-161.
Cachon, G.P. and M. Lariviere, M. (2001). “Contracting to assure supply: how to share demand
forecasts in a supply chain”. Management Science, Vol. 47, No. 5, pp. 629-646.
Book of Proceedings
267
Chen, F., Z. Drezner, J.K. Ryan, and D. Simchi-Levi (2000a). “Quantifying the bullwhip effect
in a simple supply chain: The impact of forecasting, lead times, and information”. Management
Science Vol. 46, No. 3, pp. 436-443.
Chen, F., J.K. Ryan, and D. Simchi-Levi (2000b). “The impact of exponential smoothing
forecasts on the bullwhip effect”. Naval Research Logistics, Vol. 47, No. 4, pp. 269-286.
Lee, H., P. Padmanabhan, and S. Whang (1997a). “The bullwhip effect in supply chains”. Sloan
Management Review, Vol. 38, No. 3, pp. 93-102.
Lee, H., P. Padmanabhan, and S. Whang (1997b). “Information distortion in a supply chain: The
bullwhip effect”. Management Science, Vol. 43, No. 4, pp. 546-558.
Saharidis, G., Y. Dallery, and F. Karaesmen (2006). “Centralized versus decentralized
production planning”. RAIRO Operations Research, Vol. 40, pp. 113-128.
Saharidis, G., V. Kouikoglou, and Y. Dallery (2009). “Centralized and decentralized control
policies for a two-stage stochastic supply chain with subcontracting”. International Journal of
Production Economics, Vol. 117, No. 1, pp. 117-126.
Tempelmeier H. (1997). “Resource-constrained materials requirements planning – MRP rc”.
Production Planning and Control, Vol. 8, No 5. pp. 451-461.
Tempelmeier H. and M. Derstroff (1996). “A Lagrangean-based heuristic for dynamic multilevel
multiitem constrained lotsizing with setup times”. Management Science, Vol. 42, No. 5 pp. 738-
757.
Zhang, Χ. (2004). “The impact of forecasting methods on the bullwhip effect”. International
Journal of Production Economics, Vol. 88, No. 1, pp. 15-27.
Book of Proceedings
268
Measuring employee satisfaction in a Greek academic environment
Nancy Bouranta
Department of Business Administration of Food and Agricultural Enterprises, University of
Patras, 2 G. Seferi St, Agrinio 30 100 Greece, email: [email protected]
Christian Hurson
IAE, Université de Rouen, 3, Avenue Pasteur, 76186 Rouen Cedex, France.
Yannis Siskos
Department of Informatics, University of Piraeus, 80, Karaoli & Dimitriou Street, 18534 Piraeus,
Greece.
Abstract
The cycle of organizational success proposed by Schlesinger and Heskett suggests that satisfied
employees deliver high service quality. Employee satisfaction or dissatisfaction hinges on whether
there is a productive, fulfilling relationship between staff and management. This paper focuses on
administrative staff job satisfaction measurement in the academic environment. The survey
conducted at a Greek Business University used the MUSA system of multi-criteria satisfaction
analysis. Looking at the partial satisfaction, weight, demanding and impact indexes, it is observed
that the criterion of co-workers is a leverage opportunity for the university authorities.
Keywords: Job satisfaction, Multicriteria decision analysis, MUSA system.
Book of Proceedings
269
1. Introduction
It has been well established in the literature that job satisfaction exerts an important influence on
customer satisfaction, as satisfied employees tend to be more productive, provide better services
to customers, and hence, can significantly enhance business profitability. Motivated by the
importance of employee satisfaction, the current paper focuses on administrative staff job
satisfaction measurement. The survey was conducted at the same Business University, using the
same multi-criteria method.
The following part of the paper presents the theoretical background. In the next section,
the research methodology is reported. A discussion of the findings and the managerial implications
comprises the following section. The paper ends with survey conclusions.
2. Literature Review
Job satisfaction has been one of the most germane issues for researchers, because it contains useful
information for predicting employee work-related behaviours and attitudes (Zimmerman and
Darnold, 2009). A significant trend toward overwork has also been observed. Specifically, job
satisfaction is associated with increased productivity, customer satisfaction, less absenteeism,
lower turnover and life satisfaction (Chen et al., 2006).
Job satisfaction has also been described as a person’s overall affective reaction to a set of
work and work-related factors. According to this perception, job satisfaction contains a number of
job characteristics that need to be obtained within a broad measure of employee beliefs and
attitudes about the job. Some of these facets can be divided among five main groups of job
characteristics and work environment as follows: organisational image, organisational vision,
superiors, co-workers and conditions of work (Eskildsen et al., 2010). These characteristics may
not be of equal importance to every individual (Boles et al., 2003). This means that the overall
rating for job satisfaction is not a simple average of the employees’ satisfaction levels for the
different facets of a job, but will be a more complex assessment (D’ Addio et al., 2003). In some
research, multi-faceted questions are used as stand-alone questions regarding job satisfaction; in
other cases, they are used in addition to single-item questions (Robbins and Judge, 2011).
3. Methodological frame
3.1 Objectives of research and Satisfaction Criteria
The objective of the present research is to identify the level of reported job satisfaction among the
participating employees by using Multicriteria Satisfaction Analysis, as well as to formulate
proposals for improvement or modification of administrative practices. A pilot questionnaire was
created to reflect the policy of the Greek universities. Employees completed the pilot questionnaire
and indicated any ambiguity or other difficulty they experienced in responding to the questions, as
Book of Proceedings
270
well as offering suggestions. Based on this feedback, some questions were eliminated, others were
modified, and additional items were developed.
The first measure of global job satisfaction is derived from responses by employees to the
statement: “Generally speaking, I am very satisfied with my job”. Participants were asked to select
a number from 1 to 5, where 1 = strongly disagree and 5 = strongly agree; 3 is interpreted as a
neutral response. For the purposes of this analysis, it is assumed that the higher the number
selected, the greater the level of job satisfaction. A second measure contained a set of criteria (5)
and sub-criteria (19) about employee job satisfaction (Figure 1). The items of the job satisfaction
scale were adapted from Brown and Peterson (1993). All the items were in the form of statements
on Likert–type scale.
3.2 The MUSA system
The Multicriteria Satisfaction Analysis (MUSA) system of Grigoroudis and Siskos (2010) has
been used in order to measure customer or employee satisfaction, assuming that their global
satisfaction depends on a set of criteria representing service characteristic dimensions. Thus, the
global satisfaction is denoted as a variable Y and the set of criteria is denoted as a vector Χ = (Χ1,
Χ2,…,Χn).
MUSA system uses a preference disaggregation logic. In the traditional aggregation
approach, the criteria aggregation model is known a priori, while the global preference is unknown.
On the contrary, the philosophy of disaggregation involves the inference of preference models
from given global preferences. This preference disaggregation methodology is implemented
through an ordinal regression based approach in the field of multicriteria analysis used for the
assessment of a set of a marginal satisfaction functions in such a way, that the global satisfaction
becomes as consistent as possible with participants’ marginal judgements.
According to the survey, each participant is asked to express his/her own judgements,
namely his/her global satisfaction and his/her satisfaction with regard to a set of discrete criteria.
Based on these assumptions, the problem is approached as a problem of qualitative regression and
solved via special linear programming formulations where the sum of deviations between global
satisfaction evaluation explicitly expressed by employees and the one resulting from their
multicriteria satisfaction evaluations is minimized.
The main results from the aforementioned preference disaggregation approach are focused
on global and partial explanatory analysis. Global explanatory analysis lays emphasis on
customers’ global satisfaction and its primary dimensions, while partial explanatory analysis
focuses on each criterion and their relevant parameters separately. Satisfaction analysis results
consist of:
Global satisfaction index: it shows in a range of 0-100% the level of global satisfaction of the
customers; it may be considered as the basic average performance indicator for the
organisation.
Book of Proceedings
271
Global demanding index: it shows in a range of -100%-100% the demanding level of customers
according to the following: demanding index 100%: extremely demanding customers,
demanding index 0%: "normal" customers, demanding index -100%: non-demanding
customers.
Criteria/sub-criteria satisfaction indices: they show in a range of 0-100% the level of partial
satisfaction of the customers according to the specific criterion/sub-criterion, similarly to the
global satisfaction index.
Weights of criteria/sub-criteria: they show the relative importance within a set of criteria or
sub-criteria.
Demanding indices: they show in a range of -100%-100% the demanding level of customers
according to the specific criterion/sub-criterion, similarly to the global demanding index.
The above methodology has been successfully implemented in many customer satisfaction surveys.
Recently, Gosse and Hurson (2014) applied MUSA methodology to measure job satisfaction of
recent employees in a major French organisation.
Conditions
of work
Leadership Rewards Co-workers Job security
Physical
work
environment
Leadership
style
Pay Relationship
with
coworkers
Job security
before crisis
Working
hours
Interesting for
employees’
needs
Benefit
package
Sense of
social
belonging
Job security
after crisis
Workload Opportunity
for initiatives
Free
days
Relationship
with
supervisor
Learning
opportunities
Job content
Employee
appraisal
Opportunity
of promotion
Communication
Book of Proceedings
272
Figure 1. Hierarchical structure of employee’s job satisfaction criteria
3.3. Sampling
The questionnaire was sent via e-mail to all the participants, along with an explanation of the
purpose of this academic study. A sample of 78 questionnaires from the contacted employees was
collected, of which six were excluded because they provided answers that were uniformly positive
or negative (skewed responses). The 72 usable questionnaires constitute 49.6% of the total
employee population. The demographic information showed that 58.3% of the respondents were
females and 41.7% male. The age groupings were 18-34 years (11.1%), 35-44 years (41.6%), 45-
54 years (38.9%), <55 years (8.4%). The majority of the respondent had a bachelor’s degree
(76.4%) or a high school degree (23.6%). Tenure length groupings were 2-10 years (22.2%), 11-
22 years (56.5%), 23+ (21.3%).
4. Survey Results
Seven out of ten employees declared they were from moderately (31.94%) to very satisfied
(44.44%) while only 17.9% of them declared they were very unsatisfied. Only a few (6.94%)
adopted a neutral attitude by declaring they were neither satisfied/nor unsatisfied. Globally
administrative staffers seemed to be satisfied with their jobs (global satisfaction index equal to
88.05%). The satisfaction levels of the criteria – regarding co-workers and work conditions- are
very high, as they exceed 80%. The criteria of leadership (64.41%), job security (53.0%) and
rewards (24.49%), show lower satisfaction levels. The global demanding index is -65.9%,
indicating that employees are not very demanding.
According to the weights computed by the MUSA system, the criterion of co-workers seems to be
the most important (63.25%) (Table 1). The weights of the other criteria are about 10% with little
fluctuation. This shows that administrative staffers give approximately the same importance to the
other aspects of their work and are not particularly demanding with regard to them. The criterion
of co-workers is the most important for the employees and this criterion also presented a higher
degree of satisfaction. At the level of sub-criteria it was observed that in regard to the criterion of
work conditions, employees are very satisfied with the physical work environment (91.34%) and
working hours (81.34%).They are also satisfied with their workload (73.79%) and the content of
their work (79.21%). With regard to the dimension of rewards, the satisfaction of employee
regarding free days oscillates at very satisfactory levels (82.33%), while a low degree of
satisfaction is observed for the sub-criterion of pay and benefits package (satisfaction indices:
3.7% and 25.95% respectively). Regarding job security, the administrative staff felt safe before
the implementation of a mobility and reallocation scheme. However, after the new evidence,
employees felt insecure. In addition, the sub-criteria regarding learning opportunities (17.18%)
and opportunity of promotion (45.69%) show very poor performance ratings. The partial
satisfaction index regarding the leadership style has low value as well. Moreover, the sub-criteria
regarding employees’ needs (66.58%) and opportunity for initiatives (62.35%) also performed
Book of Proceedings
273
poorly. Finally, regarding co-workers, the indices representing the satisfaction level of employees
concerning all the tree sub-criteria, had satisfactory values.
Criteria
Weights Average
satisfaction
index
Average
demanding index
Impact
Conditions of
work
10.85% 81.14% -27.14% 2.05%
Leadership 8.62% 64.41% -6.70% 3.07%
Rewards 8.13% 24.49% -0.86% 6.14%
Co-workers 63.25% 97.94% -87.50% 1.30%
Job security 9.14% 53.90% -10.86% 4.21%
Table 1. Main results for numerical example
5. Conclusions
The current survey illustrates the implementation of a preference dissagregation methodology for
measuring employee job satisfaction in a university. The global satisfaction index exceeds 88.05%,
indicating that administrative staffers at the university are satisfied with their jobs. The average
satisfaction indices regarding two of the criteria (conditions of work and co-workers) exceed 80%.
The criterion of rewards has the lowest value (24.49%). This may coincide with the country’s
economic situation that has led, in the last few years, to drastic cuts in public sector workers wages.
As far as the importance of the criteria, it is observed that good organizational relationships
(63.25%) is considered as the most important, while the other criteria average about 10%
considering as not so important. At the level of sub-criteria, it is observed that overall employees’
feelings about job security have dramatically changed after the crisis. Over the past decades, Greek
civil employees had jobs for life, ensuring security and stability. Nowadays, the economic crisis is
forcing the government to put public-sector employees on mobility and reallocation schemes.
The criteria of rewards and job security should be the first priorities for improvement in
the future, as they have the lowest partial satisfaction index. However, they are not in the university
leadership’s control, thus university authorities can only exert influence in the same extent as
government decisions. Given that, university should take effort to improve the criteria of
leadership and conditions of work. However, the criterion of condition of work is difficult to be
improved because the partial satisfaction index is high, the weight is slightly high, where as the
demanding index is slightly low.
Book of Proceedings
274
Acknowledgments
This research has been co-financed by the European Union (European Social Fund) and Greek
national funds through the Operational Program "Education and Lifelong Learning".
References
Boles, J.S., Wood, J.A. and Johnson, J. “Interrelationships of role conflict, role ambiguity, and work-family
conflict with different facets of job satisfaction and the moderating effects of gender”. Journal of Personal
Selling & Sales Management, Vol. 23, 2003, pp. 99-113.
Brown, S.P. and Peterson, R.A. “Antecedents and consequences of salesperson job satisfaction: meta
analysis and assessment of causal effects”. Journal of Marketing Research, Vol. 30, 1993, pp. 63-77.
Chen S.H., Yang C.C., Shiau J.Y., Wang, H.H. “The development of an employee satisfaction model for
higher education”. The TQM Magazine, Vol. 18, 2006, pp. 484-500.
D’Addio, A.C., Eriksson, T. and Frijters, P. “An analysis of the determinants of job satisfaction when
individuals’ baseline satisfaction levels may differ, Centre for Applied Micro econometrics (CAM)”,
Department of Economics, University of Copenhagen, 16, 2003, available at:
http://www.econ.ku.dk/CAM/Files/workingpapers/2003/2003-16.pdf
Eskildsen, J., Kristensen, K. and Antvor, H. G. “The relationship between job satisfaction and national
culture”, The TQM Journal, Vol. 22, 2010, pp. 369-378.
Gosse, B. and Hurson, Ch. “Assessment and improvement of employee job-satisfaction: a full-scale
implementation of MUSA methodology on recent employees in a major French organisation”, Working
Paper, NIMEC 2014, Universitéde Rouen, France.
Grigoroudis, E.and Siskos, Y. “Customer Satisfaction Evaluation. Methods for Measuring and
Implementing Service Quality”. International Series in Operations Research & Management Science, Vol.
139, 2010, pp. 171-216.
Robbins, S. και Judge, T. “Organisational Behaviour”, 2011, Prentice Hall.
Schlesinger, L.A. and Heskett, J.L. “Customer satisfaction is rooted in employee satisfaction”, Harvard
Business Review, November-December, 1991, pp. 149-81.
Zimmerman, R.D. and Darnold, T.C. 2009. “The impact of job performance on employee turnover
intentions and the voluntary turnover process: A meta-analysis and path model”, Personnel Review, Vol.
38, 2009, pp.142-158.
Book of Proceedings
275
Identifying factors of bank service quality during economic crisis in Greece
Nancy Bouranta
Department of Business Administration of Food and Agricultural Enterprises, University of
Patras, 2 G. Seferi Street, Agrinio 30 100 Greece, email: [email protected]
Christian Hurson
IAE, Université de Rouen, 3, Avenue Pasteur, 76186 Rouen Cedex, France.
Yannis Siskos
Department of Informatics, University of Piraeus, 80, Karaoli & Dimitriou Street, 18534 Piraeus,
Greece.
Abstract
Greek banking industry has been facing strong pressures resulting mainly from the country’s
economic concurrence. This crisis has led to structural changes in the banking sector, including
mergers and acquisitions. In addition, competition between banks for attracting new customers or
keeping the existing ones has become more intense. The unstable economic environment has a
negative impact on customer confidence and trust in the banking industry, leading customers to be
more demanding and careful in their bank selection. The purpose of this paper is to identify factors
that influence Greek customers’ evaluation of bank service quality during the economic crisis
period. A satisfaction questionnaire with self-determined scales was used and distributed to a
randomly selected sample of customers. The data was processed using the MUSA system of multi-
criteria satisfaction analysis. The results show that, generally, bank customers seem to be satisfied
with the service provided, as global satisfaction index is high; even the dissatisfied customers feel
that their expectations have been met to an extent.
Keywords: Service quality, Multicriteria decision analysis, MUSA system, banks.
Book of Proceedings
276
1. Introduction
The claim that service quality is among the most important selection criteria in bank selection has
been supported (Frango et al., 2012). Thus, delivering high-quality financial service that keeps
customers satisfied is viewed as an indispensable practice for staying alive and gaining a
sustainable advantage.
Vasiliadis (2009) pointed out that every market and every customer differs. Thus, the study of
the specificities of each market and client will be considered in this case. In terms of this concept,
it is worth assessing how Greek customers evaluate bank service quality during the crisis period.
The main objective of this paper is to determine the factors that affect customer satisfaction in the
Greek bank industry during the economic crisis period, providing managers with a more complete
view of which factors regulate customers’ satisfaction. Moreover, the comparison of the current
results with the results of older surveys may lead to interesting conclusions about customer
evaluation and how it is affected by external circumstances. To do so a satisfaction survey was
conducted in Greece using the MUSA multicriteria satisfaction analysis methodology. MUSA
system was tested in empirical surveys of the Greek bank sector (Mihelis et al., 2001; Bouranta et
al., 2002) and showed reliable measurements.
The remainder of the paper is organized as follows. Section 2 presents a brief literature review
about service quality in the bank sector. A brief sketch of the multicriteria methodological frame
is outlined in section 3. The research findings are presented in section 4. Finally section 5
concludes the paper.
2. Literature Review
In measuring service quality many frameworks have been developed. The most widely
acknowledged between academics and practitioners and applied within service industries models
are those of Grönroos (1990) and Parasuraman et al. (1988). Both models are based on the concept
that a customer judges the quality of provided services based on the discrepancy among
expectations and perceptions. However, some researchers have questioned SERVQUAL
measurement raising theoretical and operational criticisms. One of these criticisms refers to its
applicability to different service industries has been questioned in terms of the number and the
nature of its dimensions (Jabnoun and Khalifa, 2005). The discrimination of bank services has
been pointed by Vasiliadis, (2009, p. 89) who supported that “unlike many other products, a bank’s
products are characterized by low levels of standardisation, high need for adaptations, high
customer involvement in providing the service, and a need for a high volume of customers”.
Although no standard instrument for measuring service quality exists in the banking sector, most
studies in this field have adopted SERVQUAL or the alternative SERVPERF model as the
fundamental measure of service quality (Chen et al., 2012). Specifically, a modified version of
SERVQUAL is used by Fatina and Razzaque (2014) in their survey involving retail banking
services in Bangladesh. Ladhari et al. (2011), using the five SERVQUAL dimensions, compare
Book of Proceedings
277
Tunisian and Canadian customer perceptions in the banking sector. All of the SERVQUAL
dimensions were found to be important predictors of customer satisfaction and loyalty, while the
importance of each dimension differs according to consumers in two countries, mainly due to their
different economic and cultural environments. Mihelis et al. (2001) based on the assumption that
customers’ satisfaction is an assessment of a set of criteria and subcriteria, proposed four criteria
and nineteen sub-criteria. Their survey, which is conducted in the private bank sector of Greece
pointed that “customer satisfaction is a dynamic parameter … changes in the current market can
affect customers’ preferences and expectations” (p. 357). The reputation of this survey in a totally
different environment (during economic crisis) is among the purposes of this paper. The
comparison of the results may lead to interesting conclusions about customer evaluation and how
it is affected by external circumstances.
In terms of attrition, the recent country economic crisis has created an exceptional environment
that may determine customer evaluation of bank service quality. For example, Keisidou et al.
(2013) point out that customers care more than they did before the economic crisis about the price-
quality ratio, and they are not willing to pay a premium for the products and services they receive.
3. Methodological frame
3.1. Criteria of bank customer satisfaction
As it was mentioned the hierarchical structure of customers’ satisfaction dimensions, proposed by
Mihelis et al. (2001) was used as a base for this survey. The respondents were required to point
out their own judgements about global bank service quality with regard to the set of criteria. The
customer evaluation of each criterion was also measured using a single measurement index. The
self-administered questionnaire is contained also of a set of sub-criteria. All the item were in the
form of statements on Likert–type scale, where 1 refers to the statement “strongly disagree” and 5
to the statement “strongly agree”.
The main satisfaction criteria and sub-criteria consist of:
Personnel: concerning personnel skills and knowledge, communication and collaboration with
customers, as well as first line employee responsiveness.
Products: refers mainly to the variety and price of the products and service (cards, loans, bank-
assurance, etc) as well as to the special service (leasing, factoring, internet banking etc.)
Image: bank credibility (name, reputation), technological excellence (troubles in the service
system like strikes, damaged ATMs, etc.), along with stores appearance are included in this
criterion.
Service: refers to the service offered to the customers; it includes waiting time (queue, telephone,
etc.) and information provided (informing customers in an understandable way, explaining the
service and other relevant factors, informing for new products, etc.).
Book of Proceedings
278
Access: network bank expansion, branches location, as well as convenience are contained within
this criterion.
3.2. The MUSA system
The Multicriteria Satisfaction Analysis (MUSA) system of Grigoroudis and Siskos (2010) has
been used in order to measure customer or employee satisfaction, assuming that their global
satisfaction depends on a set of criteria representing service characteristic dimensions. Thus, the
global satisfaction is denoted as a variable Y and the set of criteria is denoted as a vector Χ = (Χ1,
Χ2, …,Χn).
MUSA system uses a multicriteria preference disaggregation logic. In the traditional approach,
the criteria aggregation model is known a priori, while the global preference is unknown. The
required information is collected via a simple questionnaire in which the customers evaluate the
provided product/service, i.e. they are asked to express their judgments, namely their global
satisfaction and their satisfaction with regard to the set of discrete criteria. Based on these
assumptions, the problem is approached as a problem of qualitative regression and solved via
special linear programming formulations where the sum of deviations between global satisfaction
evaluation explicitly expressed by customers and the one resulting from their multicriteria
satisfaction evaluations is minimized.
The main results from the aforementioned preference disaggregation approach are focused on
global and partial explanatory analysis. Global explanatory analysis lays emphasis on customers’
global satisfaction and its primary dimensions, while partial explanatory analysis focuses on each
criterion and their relevant parameters separately. Satisfaction analysis results, in more detail,
consist of:
Global satisfaction index: it shows in a range of 0-100% the level of global satisfaction of the
customers; it may be considered as the basic average performance indicator for the
organisation.
Global demanding index: it shows in a range of -100%-100% the demanding level of customers
according to the following: demanding index 100%: extremely demanding customers,
demanding index 0%: "normal" customers, demanding index -100%: non-demanding
customers.
Criteria/sub-criteria satisfaction indices: they show in a range of 0-100% the level of partial
satisfaction of the customers according to the specific criterion/sub-criterion, similarly to the
global satisfaction index.
Weights of criteria/sub-criteria: they show the relative importance within a set of criteria or
sub-criteria.
Demanding indices: they show in a range of -100%-100% the demanding level of customers
according to the specific criterion/sub-criterion, similarly to the global demanding index.
Book of Proceedings
279
3.3. Sampling
The survey was conducted in Greece. Respondents were customers of a particular bank and were
approached by specifically trained interviewers during selected times of the day throughout a
month-long period. The sampling method was similar to that of a mall intercept interview
(Malhotra, 2004). While more than 200 customers were approached, only 182 gave their consent
to participate. Finally, 151 usable questionnaires used for data analysis as thirty-one were excluded
because they provided answers that were uniformly positive or negative (skewed responses).
The survey took place in a provincial Greek city, similar in size population and structure, to
the comparative one. As far as the demographic characteristics of the sample are concerned, the
respondents are split reasonably evenly between males (53.6%) and females (46.4%). The
customer age groupings are 18-25 years (13.9%), 26-34 years (19.2%), 35-44 years (15.2%), 45-
55 (32.5%), and over 56 years of age (19.2%).
4. Results
The majority of the respondents declared that they were from moderately (49.7%) to very satisfied
(3.9%), while only 0.6% of them declared that they were very unsatisfied. Several of the
respondents (35.8%) adopted a neutral attitude by declaring that they were neither satisfied nor
unsatisfied. Globally, bank customers seemed to be satisfied with the service provided, as their
satisfaction index is high (84.18%). However, it should be noted that only 3.9% of the participants
stated that they are very satisfy. The global demanding index is -62.28%, indicating that customers
are not very demanding, and this is in accordance with the high global satisfaction index.
Bank customers are not demanding regarding the criterion of image, as the partial index is lower
(-87.9%) than the average global demanding index (62.28%); this result explains its very high
partial satisfaction index (95,69%). The partial impact index moves to very low levels (2.86%).
This criterion illustrates the bank competitive advantage, and it should not be concentrated on bank
efforts for improvement. The second benefit of the bank seems to be the criterion of accessibility,
as it presents a rather high satisfaction level (73.56%). The criterion of serviceability exhibits the
lowest satisfaction (45.69%), a level of demanding nature equal to 2.23%, and its partial impact
index is also higher than the others (4.44%). This means that there is room for improvement
regarding this criterion. The criteria related to personnel (62.10%) and financial products
(60.18%), show lower satisfaction levels. The weights of all of the criteria, except that of image,
fluctuate from 8.18% to 8.91%.
Comparing the results of this survey which is conducted during the economic crisis and a
relative survey by Mihelis et al. (2001) which is completed before the economic crisis, the
following observations can be made:
The global satisfaction index was before crisis 89.6% very close to the vale observed during the
crisis (84.18%).
Book of Proceedings
280
Skills and knowledge criterion seems to have greater important for the customers nowadays
(weight 83.47%) comparing with their evaluation before the crisis (weight 6.2%). The
personnel competiveness became more important than before as customers ask information
about more complicated bank products and demand personalized financial solutions. Thus,
branch personnel should be more focused on advice and expertise.
Customers want also more than before crisis a variety of financial products to choose the best
suited for them. Specifically, the criterion weight was 26.5% before crisis and 81.50% during
crisis.
One of the negative consequences of the financial crisis is customer feeling of insecurity.
Because of this, they seek more control over their finances and they are more conservative.
As customers’ lives become busier, they want not to wait in teller lines, for a financial advisor
or wander through employee desks. The waiting time criterion weight was 6.9% and inclines to
90.82%. The partial satisfaction index is declined from 66.2% to 4.45%.
5. Discussion and Conclusion
This criterion of brand name can be implied as the bank’s advantage over competition. During the
economic crisis, banks attempted to boost their brand names and rewind their reliability, as
customers began to be more cautious. The bank examined in this case has a well-known brand
name and a good reputation. However, the bank should take action in order to increase customer
satisfaction on the criteria of serviceability, financial product, and employees. Serviceability has
the lowest satisfaction rate compared to the other criteria. The bank should communicate
customers’ needs to their employees and train them in order to nurture their capability to customize
their bank services. Moreover, the bank should inform its employees on an ongoing basis about
new financial products or services so they can answer customer questions and solve their problems.
Employees should be encouraged to learn new skills, be alert to any external changes, be
empowered, and exercise the delegation to make decisions. Providing individual attention to each
customer, keeping promises, having the ability to conduct problem solving and decision making
will build long-term relationship with customers.
Acknowledgments
This research has been co-financed by the European Union (European Social Fund) and Greek
national funds through the Operational Program "Education and Lifelong Learning".
Book of Proceedings
281
References
Bouranta, A., Kouremenos, A., and Siskos, Y. “Comparative satisfaction measurement of atms vs
tellers”. in C. Zopounidis (ed), New Trends in Banking Management, Physica-Verlag Heidelberg,
N.Y., 2002, pp.37-52.
Chen, H.-G., Liu, J. Y.-CH., Sheu, T.S. and Yang, M.-H. “The impact of financial services quality
and fairness on customer satisfaction”. Managing Service Quality, Vol. 22, 2012, pp.399-421.
Fatima, J. K. and Razzaque, M.A. “Service quality and satisfaction in the banking sector”.
International Journal of Quality & Reliability Management, Vol. 31, 2014, pp. 367-379.
Frangos, C.C., Fragkos, K.C., Sotiropoulos, I., Manolopoulos, G. and Valvi, A.C. Journal of
Marketing Research & Case Studies,
http://www.ibimapublishing.com/journals/JMRCS/jmrcs.html, ID 927167, 201216 pages, DOI:
10.5171/2012.927167.
Grigoroudis, E. and Siskos, Y. “Customer Satisfaction Evaluation. Methods for Measuring and
Implementing Service Quality”. International Series in Operations Research & Management
Science, Vol. 139, 2010, pp. 171-216.
Grönroos, C. “Relationship approach to marketing in service contexts: the marketing and
organizational behaviour interface”. Journal of Business Research, Vol. 20, 1990, pp. 3-11.
Jabnoun, N. and Khalifa, A. “A customized measure of service quality in the UAE”. Managing
Service Quality, Vol. 15, 2005, pp. 374-388.
Keisidou, E. Sarigiannidis, L., Maditinos D.I. and. Thalassinos E.I. “Customer satisfaction, loyalty
and financial performance: A holistic approach of the Greek banking sector”. International
Journal of Bank Marketing, Vol. 31, 2013, pp. 259-288.
Ladhari, R., Ladhari, I. and Morales, M. “Bank service quality: comparing Canadian and Tunisian
customer perceptions”. International Journal of Bank Marketing, Vol. 29, 2011, pp. 224-246.
Malhotra, N.K. Marketing Research, 4th ed., Prentice-Hall, Upper Saddle River, NJ, 2004, pp.
268-582.
Mihelis, G., E. Grigoroudis, Y. Siskos, Y. Politis and Y. Malandrakis, “Customer satisfaction
measurement in the private bank sector”, European Journal of Operational Research, Vol. 130,
2001, pp. 347-360.
Parasuraman, A., Zeithaml, V.A. and Berry, L.L. “SERVQUAL: a multi-item scale for measuring
consumer perceptions of the service quality”. Journal of Retailing, Vol. 64, 1988, pp. 12-40.
Vasiliadis, L. “Greek banks’ internationalisation: a suggested modelling approach”. EuroMed
Journal of Business, Vol. 4, 2009, pp. 88-103.
Book of Proceedings
282
Touristic Guide: A prototype software for touristic journey planning
Lampros Mpizas
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Nestoras Tsoutsanis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Zoi Moza
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Olena Pechak
School of Chemical Engineering, National Technical University of Athens, Iroon Polytechniou 9,
Zografou 15780, Athens, Greece.
Dimitris Pantelis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Georgios K.D. Saharidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece, email: [email protected]
Abstract
Given a user profile, a well-defined network of sites and activities, and the availability of (real-
time) information, Touristic Guide software will return to the user a collection of sites and
activities that maximizes their preferences while respecting budget and time constraints at real
time. “Real-time” is practically translated to the possibility of the system to take into account
changes in real-time that affect the proposed plan (e.g. more time or cost spent in a site, a site is
closed etc.) and require re-calculation. The user profile will be processed to eventually define a set
of weights/preferences associated to each candidate site/activity that the system will make
available. For instance, a certain user may be a family man who loves ancient history and is
interested in agritourism and tours offering cultural and heritage experience. The prototype
software would then favor places and activities that authentically represent the stories and people
of the past, would promote visits to regions famous for their history, art, architecture, religion(s),
and assign high score of preference to touristic elements of peoples’ way of life and lifestyle.
Keywords: Touristic guide, Journey planning, Software, Multi criteria optimization.
Book of Proceedings
283
1. Introduction
Given a user profile, a well-defined network of sites and activities, and the availability of (real-
time) information, the intended system will return to the user a collection of sites and activities
that maximizes their preferences while respecting budget and time constraints at real time. “Real-
time” is practically translated to the possibility of the system to take into account changes in real-
time that affect the proposed plan (such as more time or cost spent in a site, a site is closed etc.)
and require re-calculation.
An already constructed network of sites and activities will feed the system with static
information (e.g. distances between the sites, expected travelling time from one site to the other)
and real-time information (e.g. related to availability, expected duration and visiting cost). An
underlying optimization model will be constructed that will take into account the user profile, the
network and the real-time information to output suggestions for visiting sites that maximize user
preferences in time and budget.
2. Modeling
The input and decision variables of Touristic Guide software are divided into two categories
following their nature:
• Variables and data related to costs
• Variables and data related to time/duration
The input and decision variables will be divided into two categories following their scope:
• Variables and data related to visits
• Variables and data related to transfers
Eventually, a mixed-integer mathematical programming model is devised that aims to:
• Select sites and activities, such that the sum of the preferences assigned to the selected sites
and activities is maximum.
• The total time spent on visits and transfers should be no greater to the time available by the
user following their input.
• The money spent on visits and transfers should be no greater to the budget available by the
user following their input.
Mathematically-wise, the underlying problem is a mixture of two notoriously difficult
problems (NP-hard): the travelling salesman problem and the knapsack problem. Anytime the user
invokes the software, the mathematical programming model will be solved to output the suggested
set of site/sight to visit and/or activities to undertake accompanied by the expected cost, expected
time to spend and driving directions on how to travel from one site to the next.
Unexpected events might occur during the tour and will be recorded in the network’s database.
For instance, a site or activity may be unavailable on the given day/time or the entrance cost may
have changed. Alternatively, real-time changes might occur because the user has decided to spend
more time than expected on a given site or changed their mind on what they want to see next. The
system will re-solve the model taking into account the current user status (sites and types of sites
Book of Proceedings
284
visited, available remaining budget and available remaining time to spend) in order to output the
next optimal set of suggestions.
3. Software interface
A beta version of the user interface of the prototype software is presented in the following Figure
1 where the user has introduced two numbers: The available time (ex. 300 minutes) and the
available budget (ex. 400 Euro).
Figure 1: Touristic Guide - User interface
The user has the possibility to setup his/her profile by using the button “PROFILE”. Using the
button “PROFILE” a new windows is popped-up (see Figure 2).
Figure 2: Profile setup
In “PROFILE” the user has to answer a number of question in order to give to the system the
possibility to construct his/her profile. Based on this answers the weights of the objective function
of the mathematical problem will be defined. Additionally, the user has the possibility to use two
other buttons: “SITE” and “ACTIVITIES” buttons. In both buttons the user could find additional
information about the available, by the software, sites and activities.
Book of Proceedings
285
4. Conclusion
In the current manuscript the Touristic Guide software is presented. The software returns to the
potential user (ex. a tourist) a collection of sites and activities that maximizes his/her preferences
while respecting budget and time constraints at real time. The prototype of the user interface of
the software is presented as well as the available options to maximize the satisfaction of the
potential user. The next step of this work is to introduce a critical mass of touristic information
and transport data for the region of Thessaly as well as to improve the user interface of the software.
The final outcome of this work is to develop a software full operational and available for free to
any potential user.
Acknowledge
The author gratefully acknowledges financial support from the Department of Mechanical
Engineering, University of Thessaly and InnovEco Company.
Book of Proceedings
286
FindMyWay: A prototype web-based platform for journey planning in Athens
city, Volos city and Crete island
George Konstatzos
InnovECO, 76, Kourtidou Str., 111 45, Athens, Greece, email: [email protected]
George Emmanouilidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Lampros Mpizas
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Nestoras Tsoutsanis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Zoi Moza
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Olena Pechak
School of Chemical Engineering, National Technical University of Athens, Iroon Polytechniou 9,
Zografou 15780, Athens, Greece.
Georgios K.D. Saharidis
Department of Mechanical Engineering, University of Thessaly, Leoforos Athinon, Pedion Areos,
38334 Volos, Greece.
Abstract
The objective of the proposed research is to develop a Decision Support System (DSS) for a web
based platform which will help individuals to move in Greece using public transportation means.
The developed platform includes mainly three prototype platforms corresponding to three different
regions of Greece. The first platform provides journey planning using public transport means in
the region of Athens as well as the connection between city of Volos and Athens. The second
platform provides journey planning in the city of Volos using the local buses and the third platform
provides journey planning between the cities of Crete island using intercity buses. The final
outcome of this research will be the development of a journey planner for the entire Greece.
Book of Proceedings
287
Keywords: Journey planner, transport, web platform
1. Introduction
Transport, the fastest growing sector in terms of energy use and emissions production, plays a
central role in the European economy and accounts for almost 20% of total gross energy
consumption in Europe, while 98% of the energy consumed in this sector is fossil fuel. Along with
nutrition and housing, transport is responsible for 70 – 80% of all environmental impacts in urban
zones.
Climate change is one of the biggest challenges that man will be facing in the coming years. For
several years, the European Union (EU) has committed to tackling climate change both internally
and internationally and is taking action to curb greenhouse gas emissions in all its areas of activity.
The Kyoto protocol is the first step towards cutting greenhouse gas emissions. This required
emissions in 2008-2012 to drop 8% below the 1990 levels. Furthermore, the EU's objective is to
ensure that the global average temperature will not increase more than 2°C above its pre-industrial
levels. To underpin these commitments, the European Commission established in 2011 two
roadmaps for a “Resource Efficient Europe” and “a low carbon economy in 2050”, in reference to
the objectives which should be reached by 2020 and the technological and structural changes that
needed up to 2050, respectively. According to them EU has set the most ambitious greenhouse gas
emissions reduction targets in the world, with binding mechanisms already in place that guarantee
a unilateral 20% reduction by 2020 compared to the 1990 levels. The EU is committed to increase
this to a 30% in 2030. This commitment will be reinforced if other developed countries commit
themselves to comparable reductions, and if economically more advanced developing countries
contribute adequately according to their responsibilities and respective capabilities. Therefore, it
is necessary for the EU Member States to promote policies that will result in the prevention of
climate change. Aiming at this direction, various industries have been called upon to measure their
carbon footprints – usually reported as greenhouse gas (GHG) emissions in CO2 equivalents.
As reported by the Roadmap for a Resource Efficient Europe (2011) a better implementation of
the existing legislation in combination with new science-based standards for air quality and the
transition to a low-carbon economy would ensure the air quality benefits. For this reason, in order
the EU to meet the air quality standards and to limit the significant impacts in the health and
environment sectors it decides to propose strategies for the air quality and emission standards,
further emissions reduction and implementation of air pollution and quality policies. Transport
accounts for around a third of all final energy consumption in the EEA member countries and for
more than a fifth of GHG emissions. It is also responsible for a large share of urban air pollution
as well as noise nuisance. The last 10 years (mainly since the 2001 White Paper on Transport), a
lot has been achieved in road and partly in rail transport. Nevertheless, according to the White
Paper (2011), if we stick to the business as usual approach, energy consumption and CO2 emissions
from transport instead of decreasing, would remain one third higher than their 1990 level.
Book of Proceedings
288
Moreover, mobility belongs to one of three key sectors (in combination with nutrition and housing)
which are responsible for 70%-80% of the environmental impacts in industrialized countries. As
a result in order someone to deal with the challenges in energy and climate change; he/she should
address the above sectors in collaboration with long term strategies, in reference to the
implementation of partnerships under the scopes of the Resource Efficiency.
FindMyWay in Greece is a demonstration project which will put into practice, methodologies and
approaches in reference to combination of co-modal transportation, GHG emissions’ reduction and
air quality improvement in the geographical region of Greece. While the idea of co-modal
transportation seems to be known in other European countries, its absence in Greece results in a
rather high gap in the transportation sector. The developed platform will be freely available to the
public in an attempt to achieve the following goals:
promote GHG emission minimisation and energy efficiency in transport
raise awareness on the environmentally-friendliest journey planning decisions for passengers
circulating in the Greek public transport network;
highlight the importance, in terms of impact and results for the environment, on preferring co-
modal transport options against using own car;
introduce an innovative policy in the Greek transport system, based on the efficient co-
modality scheme and EU emission regulations;
The aim of FindMyWay project is to enhance sustainable mobility in Greece and reduce pollution
and minimize public transport emissions in urban environment by delivering an environmentally
friendly co-modal transport planner for passengers. Additionally, FindMyWay project contribute
to the implementation of EU greenhouse gas emission reduction commitments under UNFCCC
Kyoto Protocol, and help establish by 2020 the framework for a European multimodal transport
information, management and payment system.
Following the aim of FindMyWay project, the specific project objectives are:
• To develop tools that will accurately measure the emissions of public vehicle circulating
in the EU public transport networks under consideration.
• To develop tools that will provide the environmentally friendliest journey from any point
of the network under consideration to another by using public transport and optimize passengers’
logistics
• To change the culture and the commuting habits of passengers by providing an easy-to use
service while raising awareness on the environmental benefits.
• To decrease the CO2-eq per passenger and kilometre.
• To bring together business, scientists, local and national authorities to deal with impacts of
transport on air pollution and climate change in urban zones.
2. Prototype platforms
In the frame of FindMyWay in Greece project the research team has developed 3 prototype
platforms for three specific regions of Greece. The first platform provides journey planning using
Book of Proceedings
289
public transport means in the region of Athens as well as the connection between city of Volos and
Athens. The second platform provides journey planning in the city of Volos using the local buses
and the third platform provides journey planning between the cities of Crete island using intercity
buses. All the platforms are available through the site: www.greenyourroute.com
The first platform is the one providing multimodal and intermodal routing solution. In the
following Figure 1 the user interface of the platform is presented.
Figure 1: FindMyWay in Athens
The potential user requests by the platform the optimal route starting from the city of Volos and
ending in a specific address in the city of Athens. The user has three different options suggested
by the platform. The total travel time as well as the total CO2 emission produced are given as
itinerary information.
The second platform is the one providing intermodal routing solution. In the following Figure 2
the user interface of the platform is presented.
Book of Proceedings
290
Figure 2: FindMyWay in Crete
The potential user requests by the platform the optimal route starting from one city or village of
Crete island and ending to another city of the island. The user gets the optimal route as well as
information about the waiting time to the transitional node if there is not a direct route connecting
the selected origin and destination.
Finally, the third platform is the one providing routing solution for the city of Volos. In the
following Figure 3 the user interface of the platform is presented.
Book of Proceedings
291
Figure 3: FindMyWay in Volos city
The potential user requests by the platform the optimal route starting from one address in the center
of the city of Volos and ending to another address. The user gets the optimal route represented in
the map as well as information about the local buses that he/she should use in order to arrive at
his/her destination.
3. Conclusion
In this research work three prototype platforms for three regions of Greece are presented. The
next step of this work is to collect transport information for additional regions of Greece and
introduced them in the database developed in the frame of FindMyWay project. The final outcome
of this research will be the development of a journey planner for the entire Greece connecting all
cities and villages having a population greater or equal to 50.000.
Acknowledge
The author gratefully acknowledges financial support from the Department of Mechanical
Engineering, University of Thessaly and InnovEco Company.
Book of Proceedings
292
i United Nations, Glossary for transport Statistics, 3rd edition 2003 ii S A I L project. Semitrailers in Advanced Intermodal Logistics, Analysis of the State of the
Art Deliverable 1, November 2000, http://www.zlw-ima.rwth-
aachen.de/forschung/projekte/sail/index.html. iii UIRR. Statistics 2007, UIRR International Union of combined Road-Rail transport
companies, http://www.uirr.com iv National Technical University of Athens, UIRR and CEMAT. Investigation of Greek
Transport demand for the Combined Transport corridor Greece-Italy-Germany, Pilot Project
Financed by European Commission DG VII, 1994. v RoadTransport.com EU Drivers' Hours explained, 06 May 2008,
http://www.roadtransport.com vi EC Regulation 561/2006 on the Harmonisation of Certain Social Legislation Relating to
Road Transport and Amending Council Regulations (EEC). No 3821/85 and (EC) No
2135/98 and Repealing Council Regulation (EEC) No 3820/85, 2006. vii Lowe D. Intermodal Freight Transport, Elsevier, 2005. viii Lohr Groupe, Modalohr Presentation. Available at: http://www.oevg ix Official site of ISU system at: http://isusystem.de x BRAVO project. Information available at: www.bravo-project.com . xi CREAM project “Customer-driven Rail-freight services on a European mega-corridor based on Advanced business and operating Models”.
Final Report available at: http://www.cream-project.eu/home/index.php