research article a variable neighborhood moea/d...
TRANSCRIPT
Research ArticleA Variable Neighborhood MOEAD for Multiobjective Test TaskScheduling Problem
Hui Lu Zheng Zhu Xiaoteng Wang and Lijuan Yin
School of Electronic and Information Engineering Beihang University Beijing 100191 China
Correspondence should be addressed to Hui Lu mluhuibuaaeducn
Received 26 October 2013 Revised 2 January 2014 Accepted 19 February 2014 Published 1 April 2014
Academic Editor Kui Fu Chen
Copyright copy 2014 Hui Lu et al This is an open access article distributed under the Creative Commons Attribution License whichpermits unrestricted use distribution and reproduction in any medium provided the original work is properly cited
Test task scheduling problem (TTSP) is a typical combinational optimization scheduling problem This paper proposes a variableneighborhood MOEAD (VNM) to solve the multiobjective TTSP Two minimization objectives the maximal completion time(makespan) and the mean workload are considered together In order to make solutions obtained more close to the real ParetoFront variable neighborhood strategy is adopted Variable neighborhood approach is proposed to render the crossover spanreasonable Additionally because the search space of the TTSP is so large that many duplicate solutions and local optima willexist the Starting Mutation is applied to prevent solutions from becoming trapped in local optima It is proved that the solutionsgot by VNM can converge to the global optimum by using Markov Chain and Transition Matrix respectively The experimentsof comparisons of VNM MOEAD and CNSGA (chaotic nondominated sorting genetic algorithm) indicate that VNM performsbetter than theMOEAD and theCNSGA in solving the TTSPThe results demonstrate that proposed algorithmVNM is an efficientapproach to solve the multiobjective TTSP
1 Introduction
During recent decades the manufacturing of electronicdevices has become highly integrated and increasinglycomplex As a result the resource and time consumptionexpended on the test of electronic devices became a crucialproblem in engineering application Therefore the researchfor improving the test efficiency is a topic that has attractedextensive attention To address this situation the objectiveof this research is to solve the test task scheduling problem(TTSP) more efficiently
The goal of the TTSP is to arrange the execution of 119899tasks on 119898 instruments It is a difficult nondeterministicpolynomial (NP) problem [1] for optimization TTSP hassome similarities with flexible job shop scheduling problem(FJSP) [2 3] but the resource configuration of the TTSPis more flexible For example in the TTSP one task canbe performed on more than one instrument at a time Theprecedence relationships in the TTSP resemble a networkOne task can have one or more former or latter tasks inthe TTSP Generally speaking feasible solutions are moredifficult to be obtained in the TTSP than that in the FJSP
TTSP FJSP and most scheduling problems belong tocombinational optimization problems For combinationaloptimization problems the search space is too large that thebest solution cannot be obtained by adopting the methodof enumeration for even small-scale problem Therefore theintelligent algorithms based on integer programming modelare devoted to solving these kinds of problems Take FJSPas the example genetic algorithm (GA) [4ndash7] simulatedannealing (SA) [8ndash10] and the tabu search (TS) [11] havebeen successfully applied in solving scheduling optimizationproblem FJSP receives extensive attention and researchesand many hybrid intelligent algorithms are invented forimproving the performance of the solutions For examplea combination of shuffled frog leaping and fuzzy logic isproposed to solve FJSP [12] A particle swarm optimization(PSO) algorithm and TS algorithm are combined to solve themultiobjective FJSP [13] A biogeography-based optimization(BBO) algorithm [14] was proposed for FJSP for findingoptimum or near-optimum solution Hybrid discrete parti-cle swarm optimization for multiobjective flexible job-shopscheduling problemwas proposed in article [15] especially forlarge-scale problems The objective functions are different in
Hindawi Publishing CorporationMathematical Problems in EngineeringVolume 2014 Article ID 423621 14 pageshttpdxdoiorg1011552014423621
2 Mathematical Problems in Engineering
each literature but themakespan the total tardiness the crit-ical machine workload and the total workload of machinesare frequently considered factors in those researches
Different from the research of FJSP the research of TTSPis relatively few because of the development of automatic testsystem However there are still some achievements in TTSPXia et al [16] proposed a method that combined GA andsimulated SA to optimize the parallel efficiency and speedup ratio of the multi-UUT parallel test Genetic Algorithm-AntColonyAlgorithm (GA-ACA) [17] hybrid particle swarmand tabu search [18] and Ant Colony Algorithm [19] are usedto solve parallel test tasks scheduling to obtain the mini-mum makespan A chaotic nondominated sorting geneticalgorithm is proposed to solve multiobjective TTSP [20]Thechaotic operations are combined with NSGA-II [21] in thisapproachThose algorithms have shown excellent property indecreasing costs and improving efficiencies in automatic testsystem
There are also some intelligent algorithms used to solvepower dispatch problems and other scheduling problemsFor example opposition-based learning is employed inopposition-based gravitational search algorithm (OGSA)to solve optimal reactive power dispatch [22] A fuzzifiedmultiobjective PSO (FMOPSO) algorithm is proposed andimplemented to dispatch the electric power [23] An inter-active artificial bee colony algorithm was proposed for themultiobjective environmentaleconomic dispatch problem[24]
In summary most of the researches of scheduling prob-lems focus on the single-objective problem or adopt weightedsum approach to convert the multiobjective problem into asingle-objective problemHowever theweighting coefficientsare difficult to choose and human factors will greatly impactthe performance of the algorithms In fact there are anothertwo kinds of methods for solving themultiobjective problemOne method is the non-Pareto approach utilizing operatorsfor processing the different objectives in a separated wayAnother is the Pareto approaches which are directly basedon the Pareto optimality concept They aim at satisfying twogoals converging towards the Pareto front and also obtainingdiversified solutions scattered all over the Pareto frontThosetwo kinds of methods mainly rely on the performanceand strategies of the algorithms used in the multiobjectiveproblems
In this paper the method based on Tchebycheff decom-position for multiobjective functions was adopted and thealgorithm named MOEAD is used to solve TTSP MOEADis a typical evolutionary algorithm based on decompositionproposed by Zhang and Li [25] This method decomposes amultiobjective optimization problem into a number of scalaroptimization subproblems and optimizes them simultane-ously The results show that MOEAD has a good perfor-mance for the ZDT and DTLZ test problems MOEAD isvery efficient in solving multiobjective problems Researchon MOEAD has also been performed in recent yearsFor example Tan et al [26] proposed a new version ofMOEAD with a uniform design to deal with the multiob-jective problem in higher-dimensional objective spaces This
method can render the distribution of the weighting vectorsmore uniform especially for problems with high dimensionChen et al [27] introduced a guided mutation operator andpriority update to enhance the ability ofMOEAD Stochasticranking and constraint-domination principle are adopted inMOEAD to improve the ability of the algorithm to dealwith constrained multiobjective optimization problems [28]Although these studies have improved the ability ofMOEADfor solving multi-objective problems MOEAD is mainlyused to solve standard test cases like ZDT DTLZ and F1However MOEAD is rarely used to solve combinationaloptimization problems such as FJSP TTSP Peng et al appliedMOEAD to solve Travelling Salesman Problem (TSP) [29]However there is no special improvement for MOEADaccording to the feature ofMOEAD and the property of TSP
The scheduling problems such as TTSP FJSP and TSPand power dispatching problem are a branch of combina-tional optimization problems Because of the properties of thecombinational optimization problems the final best solutionsonly account for a rather small subset of the search spaceHow to avoid the solutions obtained being trapped in localoptima is the key to improve the ability of algorithms todeal with combinational optimization problems Consideringthe fact that the size of the neighborhood is important inMOEAD [25] too large size will lead to degradation andtoo small size will weaken the effect of evolutionary processMoreover there will be many duplicate solutions due to theinfluence of neighborhood updating of MOEAD [25] Thepopulation diversity will decrease obviously Based on theanalyses above variable neighborhood based on a quadraticcurve is adopted to ensure that the crossover span is morereasonable and Gauss mutation is adopted at the beginningof iteration to maintain the diversity of the populationThese two improvements can efficiently enhance the abilityof MOEAD for avoiding the solutions obtained from beingtrapped in local optima The proposed approach cannot onlysolve TTSP but also deal with other scheduling problemsbecause the feasible solutions of TTSP are more difficult toobtain than most scheduling problems such as FJSP TSP
The organization of this paper is as follows A briefintroduction of TTSP is introduced in Section 2 The newmethod for TTSP variable neighborhood MOEAD (VNM)is proposed in Section 3The convergence analysis of VNM isalso presented in Section 4 A large number of experimentalresults and discussions are covered in Section 5 Conclusionsare given in Section 6
2 The Formulation of TTSP
21 The Mathematical Model for TTSP The goal of the TTSPis to arrange the execution of 119899 tasks on119898 instrumentsThereare three main mathematical models for TTSP One model isbased on Petri netThe second is based onGraph theory Andthe third model is based on integer programming Our workin this paper is mainly based on the integer programmingproposed by us in paper [20]
Mathematical Problems in Engineering 3
P1
P4
P2 P3
r1 r2 r3
t1 [r1 r2 r3]
(1 2 3)
12 3
Figure 1 The Petri net model for one task TTSP
211 The Petri Net Model for TTSP Petri net [30 31] wasproposed in 1962 Petri net focuses on the changes of thesystem the conditions for changes the influence of changesand the relationships between changes We assume that thereis one test task 119905
1in TTSP The instruments occupied for 119905
1
are 1199031 1199032 and 119903
3 The Petri net model for this TTSP can be
shown as Figure 1 In this model there are four places (1199011 1199012
1199013 and 119901
4) one transition (119905
1) three tokens (119903
1 1199032 and 119903
3)
three variables (V1 V2 and V
3) four arc expressions (V
1 V2 V3
and (V1 V2 V3)) and a guard ([119903
1 1199032 1199033]) where V
1 V2 and V
3
are bound to 1199031 1199032 and 119903
3
In Figure 1 at the beginning test resources 1199031 1199032 and 119903
3
are vacant The corresponding tokens for three places 1199011 1199012
and 1199013are 1199031 1199032 and 119903
3 respectively Therefore 119903
1 1199032 and
1199033can be allocated to 119905
1 When the 119905
1is finished the tokens
in 1199011 1199012 and 119901
3will be transferred to place 119901
4 The tokens
in 1199014are V1 V2 and V
3 This means that resources 119903
1 1199032 and
1199033are released The Petri net can describe the relationships
between tasks by the places and transitions but the complexmodels are needed to be establishedThe process will increasethe development cost and extend the development cycle
212TheGraphTheoryModel for TTSP Graph theory [32] isan important branch of mathematics By adopting the Graphtheory the complex project planning and processing can bedescribed using ldquographsrdquo In TTSP the vertexes of the graphrepresent the test tasks and the lines between vertexes meanthat some test instruments are common for these two tasksFor example there are four test tasks (119905
1 1199052 1199053 and 119905
4) and
four test instruments (1199031 1199032 1199033 and 119903
4) The instruments set
needed by 1199051 1199052 1199053 1199054are 1199031 1199032 1199032 1199034 1199033 1199034 and 119903
1 1199033
respectively The graph for this TTSP example is shown inFigure 2
Graph theory model can only be adopted by typicaloptimization methods With the increment of the scale ofTTSP the computation expense will greatly increase but
t1 t2
t3 t4
Figure 2 The Graph model for TTSP
typical optimization methods are not suitable for large-scaleTTSP problem Therefore Graph theory model cannot solvelarge-scale TTSP also
213 The Integer Programming Model for TTSP TTSP isa typical integer programming problem For the integerprogramming model for TTSP the TTSP can be describedas follows [20] assume that 119899 tasks and 119898 instrument areincluded in TTSP There is a task set 119879 = 119905
1 1199052 119905
119895 (1 le
119895 le 119899) and an instrument set 119877 = 1199031 1199032 119903
119894 (1 le 119894 le 119898)
119878119894
119895 119862119894119895 and 119875
119894
119895represent the test start time test finish time
and test consumed time of task 119905119895tested on instrument 119903
119894
respectively In the TTSP one task can be tested on morethan one instrument A judgment matrix is used to expresswhether instrument 119903
119894is needed for 119905
119895 The judgment matrix
is defined as the following
119874119894
119895=
1 if 119905119895occupies 119903
119894
0 others(1)
In general task 119905119895may have several possible test schemes
The set of test schemes for 119905119895is defined as 119882
119895= 1199081
119895 1199082
119895
119908
119896119895
119895 (119896119895is the number of test schemes for 119905
119895) The notation
119875119896
119895= max
119903119894isin119908119896119895119875119894
119895is used to express the test time of 119905
119895for 119908119896119895
The following describes the restriction of resources
119883119896119896lowast
119895119895lowast =
1 if 119908119896
119895cap 119908119896lowast
119895lowast = 0
0 others(2)
Basic hypothesis includes three factors At a given timean instrument can only execute one task each task must becompleted without interruption once it starts Assume 119875
119894
119895=
119875119896
119895 119862119894119895= 119878119894
119895+ 119875119894
119895to simplify the problem
22 The Objective Functions for TTSP The objective func-tions are very important in the study of multiobjectiveoptimization problem The makespan is very importantin scheduling problems such as TTSP and FJSP becausethe completion time is an essential factor for schedulingproblem in product process In additional for TTSP the testinstruments have high integration and the test instrumentshave become increasingly expensive Therefore the demand
4 Mathematical Problems in Engineering
for reducing the workload of the instruments and increasingthe service life of the test instruments has great significancein TTSPTherefore our work focuses on twomain objectivesOne is tominimize themaximal test completion time and theother is to minimize the mean workload of the instrumentsThese objectives are represented by 119891
1(119909) and 119891
2(119909)
(1)TheMaximal Test Completion Time 1198911(119909)The notification
119862119896
119895= max
119903119894isin119908119896119895119862119894
119895is the test completion time of 119905
119895for119908119896119895Thus
the maximal test completion time of all tasks can be definedas follows
1198911(119909) = max
1le119896le1198961198951le119895le119899
119862119896
119895 (3)
(2) The Mean Workload of the Instruments 1198912(119909) First a new
notation 119876 is introduced to describe the parallel steps Theinitial value of 119876 is 1 Assign the instruments for all of thetasks if 119883119896119896
lowast
119895119895lowast = 1 119876 = 119876 + 1 Therefore the mean workload
of the instruments can be defined as follows
1198912(119909) =
1
119876
119899
sum
119895=1
119898
sum
119894=1
119875119894
119895119874119894
119895 (4)
3 The Variable NeighborhoodMOEAD Algorithm
In this section we proposed a variable neighborhoodMOEAD algorithm (VNM) To obtain solutions close tothe real Pareto Front (PF) of the TTSP two strategies areadopted The variable neighborhood strategy helps to makethe crossover span more reasonable Moreover Gauss muta-tion is adopted at the beginning of the iteration to maintainthe diversity of the population
31TheMain Strategy of the VNM TheVNM is an evolution-ary algorithm based on decomposition The main strategyof the VNM is to decompose a multiobjective optimizationproblem into a number of scalar optimization subproblemsand optimize these subproblems simultaneouslyThe decom-position method used is the Tchebycheff approach [33] Eachsubproblem is bound with a weight vector and then eachsubproblem is updated by obtaining information from itsneighborhood [25] The neighborhood of each subproblemis determined by its weighting vector
Let 1205821 1205822 120582
119873 be a set of weight vectors and
119911lowast
= (119911lowast
1 119911lowast
2 119911
lowast
119898)119879 is defined as the reference point
The problem of the Pareto Front approximation can bedecomposed into 119873 scalar optimization subproblems usingthe Tchebycheff approach and the objective function of the119895th subproblem is defined as
119892te(119909 | 120582
119895 119911lowast) = max1le119894le119898
120582119895
119894
1003816100381610038161003816119891119894(119909) minus 119911
119894
lowast1003816100381610038161003816 119909 isin Ω (5)
Parameter initialization
Randomly generate Npoints in the search space
Update Update the neighborhood
Update EP
Stop criteria satisfied
Output EP
Yes
No
Staring mutation
Crossover
Mutation
zlowast
Update
Figure 3 The main procedure of the VNM
where Ω is the decision space and 120582119895
= (120582119895
1 120582119895
2 120582
119895
119898)119879
119911lowast
119894= min119891
119894(119909) | 119909 sub Ω for each 119894 = 1 2 119898 It is clear
that the VNM is able to minimize all 119873 objective functionssimultaneously in a single run
The main procedure of the VNM can be described asshown in Figure 3
In the part of parameter setting the iteration number119872 the subproblem number 119873 the size of neighborhood 119879
(which ranges from beginning size 119861 to stopping size 119878) andthe population for saving the optimal solutions EP are set
The crossover operation in VNM is as followsFor each individual 119909119905
119894in generation 119905 the child 119909
119905+1
119894can
be obtained by the following equation
119909119905+1
119894=
119909119905
119894+ 1198651times (119909119905
119894minus 119909119905
1198941) + 1198652times (119909119905
119894minus 119909119905
1198942) rand (1)ltCR
119909119905
119894rand (1)geCR
(6)
CR 1198651 and 119865
2are the three control variables for the
crossover 119909119905
1198941and 119909
119905
1198942are two individuals chosen in the
neighborhood of 119909119905119894This crossovermethod canmake full use
of the information from the neighborhood and render theinformation exchange more sufficient
Themain idea ofVNM is given above Two improvementsare involved in the VNM algorithm Variable neighborhoodstrategy is adopted to make the crossover span more rea-sonable Moreover Starting Mutation is used to enhance thediversity of the population
Mathematical Problems in Engineering 5
0 50 100 150 200 2505
10
15
20
25
30
Generations
Nei
ghbo
rhoo
d siz
e
Straight lineMonotone parabolaNonmonotone parabola
Figure 4 Three controlling curves for the neighborhood size
32 Variable Neighborhood In the VNM the size of theneighborhood 119879 has a high impact on the performance ofthe algorithm If 119879 is too large the two solutions chosen (119909119897
and 119909119896) for the genetic operation may be unsuitable for the
subproblem and degradation may occur during the progressof the evolution In contrast if119879 is too small the subproblemsare all similar The child individual will be so similar to itsparents that the crossover operation will have a weak effect
119879 is the neighborhood size which determines thecrossover and neighborhood updating span Too large andtoo small 119879 will both have a negative influence on VNMTherefore 119879 should be large enough at the beginning of theevolution period to ensure sufficient information exchange ofthe solutions and 119879 should be sufficiently small in the latterportion of the evolution period such that degradation can beavoided Motivated by this ideology we designed and testedthree curves to find the best 119879 controlling curve
The three curves are shown in Figure 4 In this figure theabscissa is the number of iterations and the ordinate is thesize of the neighborhood 1198721 1198722 and 1198723 represent thestraight line themonotonic parabolic and the nonmonotonicparabolic curves respectively It is worth noting that in curve1198722 the curvature will be 0 at the end of the evolution periodThis means that the rate of change of curvature for1198722 is thefastest of all of the concave monotone parabolas during theperiod of evolution Because the curvature goes to 0 in theend curve 1198722 is determined Assume that if the number ofiterations is 125 the neighborhood of curves 1198721 1198722 and1198723 are 119910
1 1199102 and 119910
3 respectively in accordance with the
equation 1199101minus 1199102
= 1199102minus 1199103 Thus curve 1198723 can be also
determined Curve 1198723 is a nonmonotonic parabolic curveA series of experiments should be performed to compare theinfluence of the three curves on the algorithm to identify thebest controlling curve
33 StartingMutation TheTTSP represents a typical combi-national optimization problem The final best solutions maybe limited to only several points in the solution space Becauseof the neighborhood updating effect of the VNM there will
be many duplicate solutions so that the crossover operationwill have little effectTherefore how to maintain the diversityof the population is the key question for enhancing thealgorithm effect
Motivated by the ideology above a starting Gauss muta-tion is adopted at the beginning of the iteration For a solution119909119894= (119909119894
1 119909119894
2 119909
119894
119872) (119872 is the number of variables) Gauss
mutation is described as the following
for 119895 = 1 2 119872 119909119894lowast
119895=
normal (119909119894119895 120590) rand (1) lt 119901
119909119894
119895rand (1) ge 119901
(7)
119909119894lowast
= (119909119894lowast
1 119909119894lowast
2 119909119894lowast
119872) represents the individual after muta-
tion 119901 is themutation probability normal (119909119894119895 120590) is a number
that obeys the normal distribution 119909119894119895is the mean value
and 120590 is the variance With Starting Mutation the problemwith the initially invalid crossover operation can be resolvedTherefore we can avoid the solutions from becoming trappedin local optima and thus solutions with higher quality areobtained
4 The Convergence Analysis of VNM
Theconvergence analysis of VNM in this section provides thetheory ground for its application The convergence behaviorof VNM is analyzed according to the Markov Chain and thetransfer matrix respectively
41 Strong and Weak Convergence This section proposes thebasic theories of convergence and proves the strong and weakconvergence of VNM from the perspective of Markov Chain
There is a global optimal solution set 119872 for MOPs(multiobjective problem) 119872 is defined as 119872 = 119883 forall119884 isin
119878 119891(119883) ge 119891(119884) It is assumed that (119899) is the populationin evolutionary algorithms
A detailed demonstration for the convergence of MOEAhas been proposed in paper [34] Based on it the definitionsare described as follows
Theorem 1 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
120573119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
119903119899= 12057311205732sim 120573119899
(8)
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionweakly It is defined as (119899) rarr 119872(119875119882)
Theorem 2 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
120573119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
119903119899= 12057311205732sim 120573119899
(9)
6 Mathematical Problems in Engineering
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionstrongly It is defined as (119899) rarr 119872(119875119878)
Based onTheorems 1 and 2 above the demonstration forthe convergence of VNM is described in the following Herelim119899rarrinfin
120573119899
= 0 lim119899rarrinfin
120573119899
= 0 describe the evolutionarytrend of VNMThere is lim
119899rarrinfin119903119899= 0 lim
119899rarrinfin119903119899= 0
Proof It is defined as 119875(119899) = 119875
997888
119883(119899) cap 119872 = 0Based on Bayesian we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872 = 0
= 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
+ 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
= 120572119899119875 (119899) cap 119872 = 0 + 120573
119899119875 (119899) cap 119872 = 0
(10)
Elitist strategy is adopted in VNM 120572119899= 0 Hence
119875 (119899 + 1) = 120573119899119875 (119899) (11)
Then
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0
lim119899rarrinfin
119875 (119899) cap 119872 = 0 = 1
(12)
Therefore we have
(119899) 997888rarr 119872(119875119882) (13)
It means that (119899) converges to global optimal solutionweakly
Similarly it is defined as 119875(119899) = 119875(119899) cap 119872119888
= 0By Bayesian formula we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872119888
= 0
= 120572119899119875 (119899) cap 119872
119888= 0 + 120573
119899119875 (119899)
(14)
Elitist strategy is adopted in VNM lim119899rarrinfin
120572119899= 0 Hence
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0 (15)
Then
lim119899rarrinfin
119875 (119899) cap 119872119888= 0 = 1
lim119899rarrinfin
119875 (119899) isin 119872 = 1
(16)
Therefore we have
(119899) 997888rarr 119872(119875119878) (17)
It means that (119899) converges to global optimal solutionstrongly
42 Convergence to Global Optimal This part focuses on theelitist strategy and proves that the VNM converges to theglobal optimum from the perspective of transfer matrix
Theorem 3 (see [35]) 119875 = (119862 0
119877 119879) is a reducible stochastic
matrix where 119862 119898 times 119898 is primitive stochastic matrix and119877 119879 = 0 Then
119875infin
= lim119896rarrinfin
119875119896= lim119896rarrinfin
(
119862119896
0
119896minus1
sum
119894=0
119879119894119877119862119896minus119894
119879119896)
= (
119862infin
0
119877infin
0)
(18)
where 119875infin is a stable stochastic matrix with 119875infin
= 11015840119901infin 119901infin =
1199010119875infin is unique regardless of the initial distributionThematrix
119901infin satisfies that 119901infin
119894gt 119901 for 1 le 119894 le 119898 and 119901
infin
119894= 0 for
119898 lt 119894 le 119899
According to the previous description of VNM theextended transition matrices for crossover 119862+ mutation119872
+
1
119872+
2 selection 119878
+ can be written as block diagonal matrix andupgrade matrix 119880 is lower triangular
119862+= (
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
) 119872+
1= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)
119878+= (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
) 119872+
2= (
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
119880 = (
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
(19)
119862+ 119872+1 119878+ 119872+
2 and 119880 are with 2
119899119897 square matrices 119862 1198721
1198722 119878 and119880
119886119887(1 le 119886 119887 le 2
119899119897) are all with the size of 119899times 119899 (119899is the number of individuals and 119897 is the number of individualattributes)
119886 119887 in 119880119886119887
represents the populationrsquos state sequencenumber (in the order of the populations of the pros andcons from 1 to 2
119899119897) So 119880 is used to represent populationrsquosselection process Each block matrix 119880
119886119887is a selection of
individuals The details in 119880119886119887
can be described as thereare some individuals to make 119906
119894119895= 1 established in each
row Firstly the first individual is compared with all otherindividuals 119906
1119895= 1 if 119895th individual is optimal (there may
be several optima) or 11990611
= 1 if no one is better than it Then
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
2 Mathematical Problems in Engineering
each literature but themakespan the total tardiness the crit-ical machine workload and the total workload of machinesare frequently considered factors in those researches
Different from the research of FJSP the research of TTSPis relatively few because of the development of automatic testsystem However there are still some achievements in TTSPXia et al [16] proposed a method that combined GA andsimulated SA to optimize the parallel efficiency and speedup ratio of the multi-UUT parallel test Genetic Algorithm-AntColonyAlgorithm (GA-ACA) [17] hybrid particle swarmand tabu search [18] and Ant Colony Algorithm [19] are usedto solve parallel test tasks scheduling to obtain the mini-mum makespan A chaotic nondominated sorting geneticalgorithm is proposed to solve multiobjective TTSP [20]Thechaotic operations are combined with NSGA-II [21] in thisapproachThose algorithms have shown excellent property indecreasing costs and improving efficiencies in automatic testsystem
There are also some intelligent algorithms used to solvepower dispatch problems and other scheduling problemsFor example opposition-based learning is employed inopposition-based gravitational search algorithm (OGSA)to solve optimal reactive power dispatch [22] A fuzzifiedmultiobjective PSO (FMOPSO) algorithm is proposed andimplemented to dispatch the electric power [23] An inter-active artificial bee colony algorithm was proposed for themultiobjective environmentaleconomic dispatch problem[24]
In summary most of the researches of scheduling prob-lems focus on the single-objective problem or adopt weightedsum approach to convert the multiobjective problem into asingle-objective problemHowever theweighting coefficientsare difficult to choose and human factors will greatly impactthe performance of the algorithms In fact there are anothertwo kinds of methods for solving themultiobjective problemOne method is the non-Pareto approach utilizing operatorsfor processing the different objectives in a separated wayAnother is the Pareto approaches which are directly basedon the Pareto optimality concept They aim at satisfying twogoals converging towards the Pareto front and also obtainingdiversified solutions scattered all over the Pareto frontThosetwo kinds of methods mainly rely on the performanceand strategies of the algorithms used in the multiobjectiveproblems
In this paper the method based on Tchebycheff decom-position for multiobjective functions was adopted and thealgorithm named MOEAD is used to solve TTSP MOEADis a typical evolutionary algorithm based on decompositionproposed by Zhang and Li [25] This method decomposes amultiobjective optimization problem into a number of scalaroptimization subproblems and optimizes them simultane-ously The results show that MOEAD has a good perfor-mance for the ZDT and DTLZ test problems MOEAD isvery efficient in solving multiobjective problems Researchon MOEAD has also been performed in recent yearsFor example Tan et al [26] proposed a new version ofMOEAD with a uniform design to deal with the multiob-jective problem in higher-dimensional objective spaces This
method can render the distribution of the weighting vectorsmore uniform especially for problems with high dimensionChen et al [27] introduced a guided mutation operator andpriority update to enhance the ability ofMOEAD Stochasticranking and constraint-domination principle are adopted inMOEAD to improve the ability of the algorithm to dealwith constrained multiobjective optimization problems [28]Although these studies have improved the ability ofMOEADfor solving multi-objective problems MOEAD is mainlyused to solve standard test cases like ZDT DTLZ and F1However MOEAD is rarely used to solve combinationaloptimization problems such as FJSP TTSP Peng et al appliedMOEAD to solve Travelling Salesman Problem (TSP) [29]However there is no special improvement for MOEADaccording to the feature ofMOEAD and the property of TSP
The scheduling problems such as TTSP FJSP and TSPand power dispatching problem are a branch of combina-tional optimization problems Because of the properties of thecombinational optimization problems the final best solutionsonly account for a rather small subset of the search spaceHow to avoid the solutions obtained being trapped in localoptima is the key to improve the ability of algorithms todeal with combinational optimization problems Consideringthe fact that the size of the neighborhood is important inMOEAD [25] too large size will lead to degradation andtoo small size will weaken the effect of evolutionary processMoreover there will be many duplicate solutions due to theinfluence of neighborhood updating of MOEAD [25] Thepopulation diversity will decrease obviously Based on theanalyses above variable neighborhood based on a quadraticcurve is adopted to ensure that the crossover span is morereasonable and Gauss mutation is adopted at the beginningof iteration to maintain the diversity of the populationThese two improvements can efficiently enhance the abilityof MOEAD for avoiding the solutions obtained from beingtrapped in local optima The proposed approach cannot onlysolve TTSP but also deal with other scheduling problemsbecause the feasible solutions of TTSP are more difficult toobtain than most scheduling problems such as FJSP TSP
The organization of this paper is as follows A briefintroduction of TTSP is introduced in Section 2 The newmethod for TTSP variable neighborhood MOEAD (VNM)is proposed in Section 3The convergence analysis of VNM isalso presented in Section 4 A large number of experimentalresults and discussions are covered in Section 5 Conclusionsare given in Section 6
2 The Formulation of TTSP
21 The Mathematical Model for TTSP The goal of the TTSPis to arrange the execution of 119899 tasks on119898 instrumentsThereare three main mathematical models for TTSP One model isbased on Petri netThe second is based onGraph theory Andthe third model is based on integer programming Our workin this paper is mainly based on the integer programmingproposed by us in paper [20]
Mathematical Problems in Engineering 3
P1
P4
P2 P3
r1 r2 r3
t1 [r1 r2 r3]
(1 2 3)
12 3
Figure 1 The Petri net model for one task TTSP
211 The Petri Net Model for TTSP Petri net [30 31] wasproposed in 1962 Petri net focuses on the changes of thesystem the conditions for changes the influence of changesand the relationships between changes We assume that thereis one test task 119905
1in TTSP The instruments occupied for 119905
1
are 1199031 1199032 and 119903
3 The Petri net model for this TTSP can be
shown as Figure 1 In this model there are four places (1199011 1199012
1199013 and 119901
4) one transition (119905
1) three tokens (119903
1 1199032 and 119903
3)
three variables (V1 V2 and V
3) four arc expressions (V
1 V2 V3
and (V1 V2 V3)) and a guard ([119903
1 1199032 1199033]) where V
1 V2 and V
3
are bound to 1199031 1199032 and 119903
3
In Figure 1 at the beginning test resources 1199031 1199032 and 119903
3
are vacant The corresponding tokens for three places 1199011 1199012
and 1199013are 1199031 1199032 and 119903
3 respectively Therefore 119903
1 1199032 and
1199033can be allocated to 119905
1 When the 119905
1is finished the tokens
in 1199011 1199012 and 119901
3will be transferred to place 119901
4 The tokens
in 1199014are V1 V2 and V
3 This means that resources 119903
1 1199032 and
1199033are released The Petri net can describe the relationships
between tasks by the places and transitions but the complexmodels are needed to be establishedThe process will increasethe development cost and extend the development cycle
212TheGraphTheoryModel for TTSP Graph theory [32] isan important branch of mathematics By adopting the Graphtheory the complex project planning and processing can bedescribed using ldquographsrdquo In TTSP the vertexes of the graphrepresent the test tasks and the lines between vertexes meanthat some test instruments are common for these two tasksFor example there are four test tasks (119905
1 1199052 1199053 and 119905
4) and
four test instruments (1199031 1199032 1199033 and 119903
4) The instruments set
needed by 1199051 1199052 1199053 1199054are 1199031 1199032 1199032 1199034 1199033 1199034 and 119903
1 1199033
respectively The graph for this TTSP example is shown inFigure 2
Graph theory model can only be adopted by typicaloptimization methods With the increment of the scale ofTTSP the computation expense will greatly increase but
t1 t2
t3 t4
Figure 2 The Graph model for TTSP
typical optimization methods are not suitable for large-scaleTTSP problem Therefore Graph theory model cannot solvelarge-scale TTSP also
213 The Integer Programming Model for TTSP TTSP isa typical integer programming problem For the integerprogramming model for TTSP the TTSP can be describedas follows [20] assume that 119899 tasks and 119898 instrument areincluded in TTSP There is a task set 119879 = 119905
1 1199052 119905
119895 (1 le
119895 le 119899) and an instrument set 119877 = 1199031 1199032 119903
119894 (1 le 119894 le 119898)
119878119894
119895 119862119894119895 and 119875
119894
119895represent the test start time test finish time
and test consumed time of task 119905119895tested on instrument 119903
119894
respectively In the TTSP one task can be tested on morethan one instrument A judgment matrix is used to expresswhether instrument 119903
119894is needed for 119905
119895 The judgment matrix
is defined as the following
119874119894
119895=
1 if 119905119895occupies 119903
119894
0 others(1)
In general task 119905119895may have several possible test schemes
The set of test schemes for 119905119895is defined as 119882
119895= 1199081
119895 1199082
119895
119908
119896119895
119895 (119896119895is the number of test schemes for 119905
119895) The notation
119875119896
119895= max
119903119894isin119908119896119895119875119894
119895is used to express the test time of 119905
119895for 119908119896119895
The following describes the restriction of resources
119883119896119896lowast
119895119895lowast =
1 if 119908119896
119895cap 119908119896lowast
119895lowast = 0
0 others(2)
Basic hypothesis includes three factors At a given timean instrument can only execute one task each task must becompleted without interruption once it starts Assume 119875
119894
119895=
119875119896
119895 119862119894119895= 119878119894
119895+ 119875119894
119895to simplify the problem
22 The Objective Functions for TTSP The objective func-tions are very important in the study of multiobjectiveoptimization problem The makespan is very importantin scheduling problems such as TTSP and FJSP becausethe completion time is an essential factor for schedulingproblem in product process In additional for TTSP the testinstruments have high integration and the test instrumentshave become increasingly expensive Therefore the demand
4 Mathematical Problems in Engineering
for reducing the workload of the instruments and increasingthe service life of the test instruments has great significancein TTSPTherefore our work focuses on twomain objectivesOne is tominimize themaximal test completion time and theother is to minimize the mean workload of the instrumentsThese objectives are represented by 119891
1(119909) and 119891
2(119909)
(1)TheMaximal Test Completion Time 1198911(119909)The notification
119862119896
119895= max
119903119894isin119908119896119895119862119894
119895is the test completion time of 119905
119895for119908119896119895Thus
the maximal test completion time of all tasks can be definedas follows
1198911(119909) = max
1le119896le1198961198951le119895le119899
119862119896
119895 (3)
(2) The Mean Workload of the Instruments 1198912(119909) First a new
notation 119876 is introduced to describe the parallel steps Theinitial value of 119876 is 1 Assign the instruments for all of thetasks if 119883119896119896
lowast
119895119895lowast = 1 119876 = 119876 + 1 Therefore the mean workload
of the instruments can be defined as follows
1198912(119909) =
1
119876
119899
sum
119895=1
119898
sum
119894=1
119875119894
119895119874119894
119895 (4)
3 The Variable NeighborhoodMOEAD Algorithm
In this section we proposed a variable neighborhoodMOEAD algorithm (VNM) To obtain solutions close tothe real Pareto Front (PF) of the TTSP two strategies areadopted The variable neighborhood strategy helps to makethe crossover span more reasonable Moreover Gauss muta-tion is adopted at the beginning of the iteration to maintainthe diversity of the population
31TheMain Strategy of the VNM TheVNM is an evolution-ary algorithm based on decomposition The main strategyof the VNM is to decompose a multiobjective optimizationproblem into a number of scalar optimization subproblemsand optimize these subproblems simultaneouslyThe decom-position method used is the Tchebycheff approach [33] Eachsubproblem is bound with a weight vector and then eachsubproblem is updated by obtaining information from itsneighborhood [25] The neighborhood of each subproblemis determined by its weighting vector
Let 1205821 1205822 120582
119873 be a set of weight vectors and
119911lowast
= (119911lowast
1 119911lowast
2 119911
lowast
119898)119879 is defined as the reference point
The problem of the Pareto Front approximation can bedecomposed into 119873 scalar optimization subproblems usingthe Tchebycheff approach and the objective function of the119895th subproblem is defined as
119892te(119909 | 120582
119895 119911lowast) = max1le119894le119898
120582119895
119894
1003816100381610038161003816119891119894(119909) minus 119911
119894
lowast1003816100381610038161003816 119909 isin Ω (5)
Parameter initialization
Randomly generate Npoints in the search space
Update Update the neighborhood
Update EP
Stop criteria satisfied
Output EP
Yes
No
Staring mutation
Crossover
Mutation
zlowast
Update
Figure 3 The main procedure of the VNM
where Ω is the decision space and 120582119895
= (120582119895
1 120582119895
2 120582
119895
119898)119879
119911lowast
119894= min119891
119894(119909) | 119909 sub Ω for each 119894 = 1 2 119898 It is clear
that the VNM is able to minimize all 119873 objective functionssimultaneously in a single run
The main procedure of the VNM can be described asshown in Figure 3
In the part of parameter setting the iteration number119872 the subproblem number 119873 the size of neighborhood 119879
(which ranges from beginning size 119861 to stopping size 119878) andthe population for saving the optimal solutions EP are set
The crossover operation in VNM is as followsFor each individual 119909119905
119894in generation 119905 the child 119909
119905+1
119894can
be obtained by the following equation
119909119905+1
119894=
119909119905
119894+ 1198651times (119909119905
119894minus 119909119905
1198941) + 1198652times (119909119905
119894minus 119909119905
1198942) rand (1)ltCR
119909119905
119894rand (1)geCR
(6)
CR 1198651 and 119865
2are the three control variables for the
crossover 119909119905
1198941and 119909
119905
1198942are two individuals chosen in the
neighborhood of 119909119905119894This crossovermethod canmake full use
of the information from the neighborhood and render theinformation exchange more sufficient
Themain idea ofVNM is given above Two improvementsare involved in the VNM algorithm Variable neighborhoodstrategy is adopted to make the crossover span more rea-sonable Moreover Starting Mutation is used to enhance thediversity of the population
Mathematical Problems in Engineering 5
0 50 100 150 200 2505
10
15
20
25
30
Generations
Nei
ghbo
rhoo
d siz
e
Straight lineMonotone parabolaNonmonotone parabola
Figure 4 Three controlling curves for the neighborhood size
32 Variable Neighborhood In the VNM the size of theneighborhood 119879 has a high impact on the performance ofthe algorithm If 119879 is too large the two solutions chosen (119909119897
and 119909119896) for the genetic operation may be unsuitable for the
subproblem and degradation may occur during the progressof the evolution In contrast if119879 is too small the subproblemsare all similar The child individual will be so similar to itsparents that the crossover operation will have a weak effect
119879 is the neighborhood size which determines thecrossover and neighborhood updating span Too large andtoo small 119879 will both have a negative influence on VNMTherefore 119879 should be large enough at the beginning of theevolution period to ensure sufficient information exchange ofthe solutions and 119879 should be sufficiently small in the latterportion of the evolution period such that degradation can beavoided Motivated by this ideology we designed and testedthree curves to find the best 119879 controlling curve
The three curves are shown in Figure 4 In this figure theabscissa is the number of iterations and the ordinate is thesize of the neighborhood 1198721 1198722 and 1198723 represent thestraight line themonotonic parabolic and the nonmonotonicparabolic curves respectively It is worth noting that in curve1198722 the curvature will be 0 at the end of the evolution periodThis means that the rate of change of curvature for1198722 is thefastest of all of the concave monotone parabolas during theperiod of evolution Because the curvature goes to 0 in theend curve 1198722 is determined Assume that if the number ofiterations is 125 the neighborhood of curves 1198721 1198722 and1198723 are 119910
1 1199102 and 119910
3 respectively in accordance with the
equation 1199101minus 1199102
= 1199102minus 1199103 Thus curve 1198723 can be also
determined Curve 1198723 is a nonmonotonic parabolic curveA series of experiments should be performed to compare theinfluence of the three curves on the algorithm to identify thebest controlling curve
33 StartingMutation TheTTSP represents a typical combi-national optimization problem The final best solutions maybe limited to only several points in the solution space Becauseof the neighborhood updating effect of the VNM there will
be many duplicate solutions so that the crossover operationwill have little effectTherefore how to maintain the diversityof the population is the key question for enhancing thealgorithm effect
Motivated by the ideology above a starting Gauss muta-tion is adopted at the beginning of the iteration For a solution119909119894= (119909119894
1 119909119894
2 119909
119894
119872) (119872 is the number of variables) Gauss
mutation is described as the following
for 119895 = 1 2 119872 119909119894lowast
119895=
normal (119909119894119895 120590) rand (1) lt 119901
119909119894
119895rand (1) ge 119901
(7)
119909119894lowast
= (119909119894lowast
1 119909119894lowast
2 119909119894lowast
119872) represents the individual after muta-
tion 119901 is themutation probability normal (119909119894119895 120590) is a number
that obeys the normal distribution 119909119894119895is the mean value
and 120590 is the variance With Starting Mutation the problemwith the initially invalid crossover operation can be resolvedTherefore we can avoid the solutions from becoming trappedin local optima and thus solutions with higher quality areobtained
4 The Convergence Analysis of VNM
Theconvergence analysis of VNM in this section provides thetheory ground for its application The convergence behaviorof VNM is analyzed according to the Markov Chain and thetransfer matrix respectively
41 Strong and Weak Convergence This section proposes thebasic theories of convergence and proves the strong and weakconvergence of VNM from the perspective of Markov Chain
There is a global optimal solution set 119872 for MOPs(multiobjective problem) 119872 is defined as 119872 = 119883 forall119884 isin
119878 119891(119883) ge 119891(119884) It is assumed that (119899) is the populationin evolutionary algorithms
A detailed demonstration for the convergence of MOEAhas been proposed in paper [34] Based on it the definitionsare described as follows
Theorem 1 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
120573119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
119903119899= 12057311205732sim 120573119899
(8)
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionweakly It is defined as (119899) rarr 119872(119875119882)
Theorem 2 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
120573119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
119903119899= 12057311205732sim 120573119899
(9)
6 Mathematical Problems in Engineering
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionstrongly It is defined as (119899) rarr 119872(119875119878)
Based onTheorems 1 and 2 above the demonstration forthe convergence of VNM is described in the following Herelim119899rarrinfin
120573119899
= 0 lim119899rarrinfin
120573119899
= 0 describe the evolutionarytrend of VNMThere is lim
119899rarrinfin119903119899= 0 lim
119899rarrinfin119903119899= 0
Proof It is defined as 119875(119899) = 119875
997888
119883(119899) cap 119872 = 0Based on Bayesian we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872 = 0
= 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
+ 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
= 120572119899119875 (119899) cap 119872 = 0 + 120573
119899119875 (119899) cap 119872 = 0
(10)
Elitist strategy is adopted in VNM 120572119899= 0 Hence
119875 (119899 + 1) = 120573119899119875 (119899) (11)
Then
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0
lim119899rarrinfin
119875 (119899) cap 119872 = 0 = 1
(12)
Therefore we have
(119899) 997888rarr 119872(119875119882) (13)
It means that (119899) converges to global optimal solutionweakly
Similarly it is defined as 119875(119899) = 119875(119899) cap 119872119888
= 0By Bayesian formula we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872119888
= 0
= 120572119899119875 (119899) cap 119872
119888= 0 + 120573
119899119875 (119899)
(14)
Elitist strategy is adopted in VNM lim119899rarrinfin
120572119899= 0 Hence
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0 (15)
Then
lim119899rarrinfin
119875 (119899) cap 119872119888= 0 = 1
lim119899rarrinfin
119875 (119899) isin 119872 = 1
(16)
Therefore we have
(119899) 997888rarr 119872(119875119878) (17)
It means that (119899) converges to global optimal solutionstrongly
42 Convergence to Global Optimal This part focuses on theelitist strategy and proves that the VNM converges to theglobal optimum from the perspective of transfer matrix
Theorem 3 (see [35]) 119875 = (119862 0
119877 119879) is a reducible stochastic
matrix where 119862 119898 times 119898 is primitive stochastic matrix and119877 119879 = 0 Then
119875infin
= lim119896rarrinfin
119875119896= lim119896rarrinfin
(
119862119896
0
119896minus1
sum
119894=0
119879119894119877119862119896minus119894
119879119896)
= (
119862infin
0
119877infin
0)
(18)
where 119875infin is a stable stochastic matrix with 119875infin
= 11015840119901infin 119901infin =
1199010119875infin is unique regardless of the initial distributionThematrix
119901infin satisfies that 119901infin
119894gt 119901 for 1 le 119894 le 119898 and 119901
infin
119894= 0 for
119898 lt 119894 le 119899
According to the previous description of VNM theextended transition matrices for crossover 119862+ mutation119872
+
1
119872+
2 selection 119878
+ can be written as block diagonal matrix andupgrade matrix 119880 is lower triangular
119862+= (
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
) 119872+
1= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)
119878+= (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
) 119872+
2= (
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
119880 = (
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
(19)
119862+ 119872+1 119878+ 119872+
2 and 119880 are with 2
119899119897 square matrices 119862 1198721
1198722 119878 and119880
119886119887(1 le 119886 119887 le 2
119899119897) are all with the size of 119899times 119899 (119899is the number of individuals and 119897 is the number of individualattributes)
119886 119887 in 119880119886119887
represents the populationrsquos state sequencenumber (in the order of the populations of the pros andcons from 1 to 2
119899119897) So 119880 is used to represent populationrsquosselection process Each block matrix 119880
119886119887is a selection of
individuals The details in 119880119886119887
can be described as thereare some individuals to make 119906
119894119895= 1 established in each
row Firstly the first individual is compared with all otherindividuals 119906
1119895= 1 if 119895th individual is optimal (there may
be several optima) or 11990611
= 1 if no one is better than it Then
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 3
P1
P4
P2 P3
r1 r2 r3
t1 [r1 r2 r3]
(1 2 3)
12 3
Figure 1 The Petri net model for one task TTSP
211 The Petri Net Model for TTSP Petri net [30 31] wasproposed in 1962 Petri net focuses on the changes of thesystem the conditions for changes the influence of changesand the relationships between changes We assume that thereis one test task 119905
1in TTSP The instruments occupied for 119905
1
are 1199031 1199032 and 119903
3 The Petri net model for this TTSP can be
shown as Figure 1 In this model there are four places (1199011 1199012
1199013 and 119901
4) one transition (119905
1) three tokens (119903
1 1199032 and 119903
3)
three variables (V1 V2 and V
3) four arc expressions (V
1 V2 V3
and (V1 V2 V3)) and a guard ([119903
1 1199032 1199033]) where V
1 V2 and V
3
are bound to 1199031 1199032 and 119903
3
In Figure 1 at the beginning test resources 1199031 1199032 and 119903
3
are vacant The corresponding tokens for three places 1199011 1199012
and 1199013are 1199031 1199032 and 119903
3 respectively Therefore 119903
1 1199032 and
1199033can be allocated to 119905
1 When the 119905
1is finished the tokens
in 1199011 1199012 and 119901
3will be transferred to place 119901
4 The tokens
in 1199014are V1 V2 and V
3 This means that resources 119903
1 1199032 and
1199033are released The Petri net can describe the relationships
between tasks by the places and transitions but the complexmodels are needed to be establishedThe process will increasethe development cost and extend the development cycle
212TheGraphTheoryModel for TTSP Graph theory [32] isan important branch of mathematics By adopting the Graphtheory the complex project planning and processing can bedescribed using ldquographsrdquo In TTSP the vertexes of the graphrepresent the test tasks and the lines between vertexes meanthat some test instruments are common for these two tasksFor example there are four test tasks (119905
1 1199052 1199053 and 119905
4) and
four test instruments (1199031 1199032 1199033 and 119903
4) The instruments set
needed by 1199051 1199052 1199053 1199054are 1199031 1199032 1199032 1199034 1199033 1199034 and 119903
1 1199033
respectively The graph for this TTSP example is shown inFigure 2
Graph theory model can only be adopted by typicaloptimization methods With the increment of the scale ofTTSP the computation expense will greatly increase but
t1 t2
t3 t4
Figure 2 The Graph model for TTSP
typical optimization methods are not suitable for large-scaleTTSP problem Therefore Graph theory model cannot solvelarge-scale TTSP also
213 The Integer Programming Model for TTSP TTSP isa typical integer programming problem For the integerprogramming model for TTSP the TTSP can be describedas follows [20] assume that 119899 tasks and 119898 instrument areincluded in TTSP There is a task set 119879 = 119905
1 1199052 119905
119895 (1 le
119895 le 119899) and an instrument set 119877 = 1199031 1199032 119903
119894 (1 le 119894 le 119898)
119878119894
119895 119862119894119895 and 119875
119894
119895represent the test start time test finish time
and test consumed time of task 119905119895tested on instrument 119903
119894
respectively In the TTSP one task can be tested on morethan one instrument A judgment matrix is used to expresswhether instrument 119903
119894is needed for 119905
119895 The judgment matrix
is defined as the following
119874119894
119895=
1 if 119905119895occupies 119903
119894
0 others(1)
In general task 119905119895may have several possible test schemes
The set of test schemes for 119905119895is defined as 119882
119895= 1199081
119895 1199082
119895
119908
119896119895
119895 (119896119895is the number of test schemes for 119905
119895) The notation
119875119896
119895= max
119903119894isin119908119896119895119875119894
119895is used to express the test time of 119905
119895for 119908119896119895
The following describes the restriction of resources
119883119896119896lowast
119895119895lowast =
1 if 119908119896
119895cap 119908119896lowast
119895lowast = 0
0 others(2)
Basic hypothesis includes three factors At a given timean instrument can only execute one task each task must becompleted without interruption once it starts Assume 119875
119894
119895=
119875119896
119895 119862119894119895= 119878119894
119895+ 119875119894
119895to simplify the problem
22 The Objective Functions for TTSP The objective func-tions are very important in the study of multiobjectiveoptimization problem The makespan is very importantin scheduling problems such as TTSP and FJSP becausethe completion time is an essential factor for schedulingproblem in product process In additional for TTSP the testinstruments have high integration and the test instrumentshave become increasingly expensive Therefore the demand
4 Mathematical Problems in Engineering
for reducing the workload of the instruments and increasingthe service life of the test instruments has great significancein TTSPTherefore our work focuses on twomain objectivesOne is tominimize themaximal test completion time and theother is to minimize the mean workload of the instrumentsThese objectives are represented by 119891
1(119909) and 119891
2(119909)
(1)TheMaximal Test Completion Time 1198911(119909)The notification
119862119896
119895= max
119903119894isin119908119896119895119862119894
119895is the test completion time of 119905
119895for119908119896119895Thus
the maximal test completion time of all tasks can be definedas follows
1198911(119909) = max
1le119896le1198961198951le119895le119899
119862119896
119895 (3)
(2) The Mean Workload of the Instruments 1198912(119909) First a new
notation 119876 is introduced to describe the parallel steps Theinitial value of 119876 is 1 Assign the instruments for all of thetasks if 119883119896119896
lowast
119895119895lowast = 1 119876 = 119876 + 1 Therefore the mean workload
of the instruments can be defined as follows
1198912(119909) =
1
119876
119899
sum
119895=1
119898
sum
119894=1
119875119894
119895119874119894
119895 (4)
3 The Variable NeighborhoodMOEAD Algorithm
In this section we proposed a variable neighborhoodMOEAD algorithm (VNM) To obtain solutions close tothe real Pareto Front (PF) of the TTSP two strategies areadopted The variable neighborhood strategy helps to makethe crossover span more reasonable Moreover Gauss muta-tion is adopted at the beginning of the iteration to maintainthe diversity of the population
31TheMain Strategy of the VNM TheVNM is an evolution-ary algorithm based on decomposition The main strategyof the VNM is to decompose a multiobjective optimizationproblem into a number of scalar optimization subproblemsand optimize these subproblems simultaneouslyThe decom-position method used is the Tchebycheff approach [33] Eachsubproblem is bound with a weight vector and then eachsubproblem is updated by obtaining information from itsneighborhood [25] The neighborhood of each subproblemis determined by its weighting vector
Let 1205821 1205822 120582
119873 be a set of weight vectors and
119911lowast
= (119911lowast
1 119911lowast
2 119911
lowast
119898)119879 is defined as the reference point
The problem of the Pareto Front approximation can bedecomposed into 119873 scalar optimization subproblems usingthe Tchebycheff approach and the objective function of the119895th subproblem is defined as
119892te(119909 | 120582
119895 119911lowast) = max1le119894le119898
120582119895
119894
1003816100381610038161003816119891119894(119909) minus 119911
119894
lowast1003816100381610038161003816 119909 isin Ω (5)
Parameter initialization
Randomly generate Npoints in the search space
Update Update the neighborhood
Update EP
Stop criteria satisfied
Output EP
Yes
No
Staring mutation
Crossover
Mutation
zlowast
Update
Figure 3 The main procedure of the VNM
where Ω is the decision space and 120582119895
= (120582119895
1 120582119895
2 120582
119895
119898)119879
119911lowast
119894= min119891
119894(119909) | 119909 sub Ω for each 119894 = 1 2 119898 It is clear
that the VNM is able to minimize all 119873 objective functionssimultaneously in a single run
The main procedure of the VNM can be described asshown in Figure 3
In the part of parameter setting the iteration number119872 the subproblem number 119873 the size of neighborhood 119879
(which ranges from beginning size 119861 to stopping size 119878) andthe population for saving the optimal solutions EP are set
The crossover operation in VNM is as followsFor each individual 119909119905
119894in generation 119905 the child 119909
119905+1
119894can
be obtained by the following equation
119909119905+1
119894=
119909119905
119894+ 1198651times (119909119905
119894minus 119909119905
1198941) + 1198652times (119909119905
119894minus 119909119905
1198942) rand (1)ltCR
119909119905
119894rand (1)geCR
(6)
CR 1198651 and 119865
2are the three control variables for the
crossover 119909119905
1198941and 119909
119905
1198942are two individuals chosen in the
neighborhood of 119909119905119894This crossovermethod canmake full use
of the information from the neighborhood and render theinformation exchange more sufficient
Themain idea ofVNM is given above Two improvementsare involved in the VNM algorithm Variable neighborhoodstrategy is adopted to make the crossover span more rea-sonable Moreover Starting Mutation is used to enhance thediversity of the population
Mathematical Problems in Engineering 5
0 50 100 150 200 2505
10
15
20
25
30
Generations
Nei
ghbo
rhoo
d siz
e
Straight lineMonotone parabolaNonmonotone parabola
Figure 4 Three controlling curves for the neighborhood size
32 Variable Neighborhood In the VNM the size of theneighborhood 119879 has a high impact on the performance ofthe algorithm If 119879 is too large the two solutions chosen (119909119897
and 119909119896) for the genetic operation may be unsuitable for the
subproblem and degradation may occur during the progressof the evolution In contrast if119879 is too small the subproblemsare all similar The child individual will be so similar to itsparents that the crossover operation will have a weak effect
119879 is the neighborhood size which determines thecrossover and neighborhood updating span Too large andtoo small 119879 will both have a negative influence on VNMTherefore 119879 should be large enough at the beginning of theevolution period to ensure sufficient information exchange ofthe solutions and 119879 should be sufficiently small in the latterportion of the evolution period such that degradation can beavoided Motivated by this ideology we designed and testedthree curves to find the best 119879 controlling curve
The three curves are shown in Figure 4 In this figure theabscissa is the number of iterations and the ordinate is thesize of the neighborhood 1198721 1198722 and 1198723 represent thestraight line themonotonic parabolic and the nonmonotonicparabolic curves respectively It is worth noting that in curve1198722 the curvature will be 0 at the end of the evolution periodThis means that the rate of change of curvature for1198722 is thefastest of all of the concave monotone parabolas during theperiod of evolution Because the curvature goes to 0 in theend curve 1198722 is determined Assume that if the number ofiterations is 125 the neighborhood of curves 1198721 1198722 and1198723 are 119910
1 1199102 and 119910
3 respectively in accordance with the
equation 1199101minus 1199102
= 1199102minus 1199103 Thus curve 1198723 can be also
determined Curve 1198723 is a nonmonotonic parabolic curveA series of experiments should be performed to compare theinfluence of the three curves on the algorithm to identify thebest controlling curve
33 StartingMutation TheTTSP represents a typical combi-national optimization problem The final best solutions maybe limited to only several points in the solution space Becauseof the neighborhood updating effect of the VNM there will
be many duplicate solutions so that the crossover operationwill have little effectTherefore how to maintain the diversityof the population is the key question for enhancing thealgorithm effect
Motivated by the ideology above a starting Gauss muta-tion is adopted at the beginning of the iteration For a solution119909119894= (119909119894
1 119909119894
2 119909
119894
119872) (119872 is the number of variables) Gauss
mutation is described as the following
for 119895 = 1 2 119872 119909119894lowast
119895=
normal (119909119894119895 120590) rand (1) lt 119901
119909119894
119895rand (1) ge 119901
(7)
119909119894lowast
= (119909119894lowast
1 119909119894lowast
2 119909119894lowast
119872) represents the individual after muta-
tion 119901 is themutation probability normal (119909119894119895 120590) is a number
that obeys the normal distribution 119909119894119895is the mean value
and 120590 is the variance With Starting Mutation the problemwith the initially invalid crossover operation can be resolvedTherefore we can avoid the solutions from becoming trappedin local optima and thus solutions with higher quality areobtained
4 The Convergence Analysis of VNM
Theconvergence analysis of VNM in this section provides thetheory ground for its application The convergence behaviorof VNM is analyzed according to the Markov Chain and thetransfer matrix respectively
41 Strong and Weak Convergence This section proposes thebasic theories of convergence and proves the strong and weakconvergence of VNM from the perspective of Markov Chain
There is a global optimal solution set 119872 for MOPs(multiobjective problem) 119872 is defined as 119872 = 119883 forall119884 isin
119878 119891(119883) ge 119891(119884) It is assumed that (119899) is the populationin evolutionary algorithms
A detailed demonstration for the convergence of MOEAhas been proposed in paper [34] Based on it the definitionsare described as follows
Theorem 1 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
120573119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
119903119899= 12057311205732sim 120573119899
(8)
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionweakly It is defined as (119899) rarr 119872(119875119882)
Theorem 2 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
120573119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
119903119899= 12057311205732sim 120573119899
(9)
6 Mathematical Problems in Engineering
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionstrongly It is defined as (119899) rarr 119872(119875119878)
Based onTheorems 1 and 2 above the demonstration forthe convergence of VNM is described in the following Herelim119899rarrinfin
120573119899
= 0 lim119899rarrinfin
120573119899
= 0 describe the evolutionarytrend of VNMThere is lim
119899rarrinfin119903119899= 0 lim
119899rarrinfin119903119899= 0
Proof It is defined as 119875(119899) = 119875
997888
119883(119899) cap 119872 = 0Based on Bayesian we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872 = 0
= 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
+ 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
= 120572119899119875 (119899) cap 119872 = 0 + 120573
119899119875 (119899) cap 119872 = 0
(10)
Elitist strategy is adopted in VNM 120572119899= 0 Hence
119875 (119899 + 1) = 120573119899119875 (119899) (11)
Then
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0
lim119899rarrinfin
119875 (119899) cap 119872 = 0 = 1
(12)
Therefore we have
(119899) 997888rarr 119872(119875119882) (13)
It means that (119899) converges to global optimal solutionweakly
Similarly it is defined as 119875(119899) = 119875(119899) cap 119872119888
= 0By Bayesian formula we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872119888
= 0
= 120572119899119875 (119899) cap 119872
119888= 0 + 120573
119899119875 (119899)
(14)
Elitist strategy is adopted in VNM lim119899rarrinfin
120572119899= 0 Hence
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0 (15)
Then
lim119899rarrinfin
119875 (119899) cap 119872119888= 0 = 1
lim119899rarrinfin
119875 (119899) isin 119872 = 1
(16)
Therefore we have
(119899) 997888rarr 119872(119875119878) (17)
It means that (119899) converges to global optimal solutionstrongly
42 Convergence to Global Optimal This part focuses on theelitist strategy and proves that the VNM converges to theglobal optimum from the perspective of transfer matrix
Theorem 3 (see [35]) 119875 = (119862 0
119877 119879) is a reducible stochastic
matrix where 119862 119898 times 119898 is primitive stochastic matrix and119877 119879 = 0 Then
119875infin
= lim119896rarrinfin
119875119896= lim119896rarrinfin
(
119862119896
0
119896minus1
sum
119894=0
119879119894119877119862119896minus119894
119879119896)
= (
119862infin
0
119877infin
0)
(18)
where 119875infin is a stable stochastic matrix with 119875infin
= 11015840119901infin 119901infin =
1199010119875infin is unique regardless of the initial distributionThematrix
119901infin satisfies that 119901infin
119894gt 119901 for 1 le 119894 le 119898 and 119901
infin
119894= 0 for
119898 lt 119894 le 119899
According to the previous description of VNM theextended transition matrices for crossover 119862+ mutation119872
+
1
119872+
2 selection 119878
+ can be written as block diagonal matrix andupgrade matrix 119880 is lower triangular
119862+= (
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
) 119872+
1= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)
119878+= (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
) 119872+
2= (
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
119880 = (
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
(19)
119862+ 119872+1 119878+ 119872+
2 and 119880 are with 2
119899119897 square matrices 119862 1198721
1198722 119878 and119880
119886119887(1 le 119886 119887 le 2
119899119897) are all with the size of 119899times 119899 (119899is the number of individuals and 119897 is the number of individualattributes)
119886 119887 in 119880119886119887
represents the populationrsquos state sequencenumber (in the order of the populations of the pros andcons from 1 to 2
119899119897) So 119880 is used to represent populationrsquosselection process Each block matrix 119880
119886119887is a selection of
individuals The details in 119880119886119887
can be described as thereare some individuals to make 119906
119894119895= 1 established in each
row Firstly the first individual is compared with all otherindividuals 119906
1119895= 1 if 119895th individual is optimal (there may
be several optima) or 11990611
= 1 if no one is better than it Then
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
4 Mathematical Problems in Engineering
for reducing the workload of the instruments and increasingthe service life of the test instruments has great significancein TTSPTherefore our work focuses on twomain objectivesOne is tominimize themaximal test completion time and theother is to minimize the mean workload of the instrumentsThese objectives are represented by 119891
1(119909) and 119891
2(119909)
(1)TheMaximal Test Completion Time 1198911(119909)The notification
119862119896
119895= max
119903119894isin119908119896119895119862119894
119895is the test completion time of 119905
119895for119908119896119895Thus
the maximal test completion time of all tasks can be definedas follows
1198911(119909) = max
1le119896le1198961198951le119895le119899
119862119896
119895 (3)
(2) The Mean Workload of the Instruments 1198912(119909) First a new
notation 119876 is introduced to describe the parallel steps Theinitial value of 119876 is 1 Assign the instruments for all of thetasks if 119883119896119896
lowast
119895119895lowast = 1 119876 = 119876 + 1 Therefore the mean workload
of the instruments can be defined as follows
1198912(119909) =
1
119876
119899
sum
119895=1
119898
sum
119894=1
119875119894
119895119874119894
119895 (4)
3 The Variable NeighborhoodMOEAD Algorithm
In this section we proposed a variable neighborhoodMOEAD algorithm (VNM) To obtain solutions close tothe real Pareto Front (PF) of the TTSP two strategies areadopted The variable neighborhood strategy helps to makethe crossover span more reasonable Moreover Gauss muta-tion is adopted at the beginning of the iteration to maintainthe diversity of the population
31TheMain Strategy of the VNM TheVNM is an evolution-ary algorithm based on decomposition The main strategyof the VNM is to decompose a multiobjective optimizationproblem into a number of scalar optimization subproblemsand optimize these subproblems simultaneouslyThe decom-position method used is the Tchebycheff approach [33] Eachsubproblem is bound with a weight vector and then eachsubproblem is updated by obtaining information from itsneighborhood [25] The neighborhood of each subproblemis determined by its weighting vector
Let 1205821 1205822 120582
119873 be a set of weight vectors and
119911lowast
= (119911lowast
1 119911lowast
2 119911
lowast
119898)119879 is defined as the reference point
The problem of the Pareto Front approximation can bedecomposed into 119873 scalar optimization subproblems usingthe Tchebycheff approach and the objective function of the119895th subproblem is defined as
119892te(119909 | 120582
119895 119911lowast) = max1le119894le119898
120582119895
119894
1003816100381610038161003816119891119894(119909) minus 119911
119894
lowast1003816100381610038161003816 119909 isin Ω (5)
Parameter initialization
Randomly generate Npoints in the search space
Update Update the neighborhood
Update EP
Stop criteria satisfied
Output EP
Yes
No
Staring mutation
Crossover
Mutation
zlowast
Update
Figure 3 The main procedure of the VNM
where Ω is the decision space and 120582119895
= (120582119895
1 120582119895
2 120582
119895
119898)119879
119911lowast
119894= min119891
119894(119909) | 119909 sub Ω for each 119894 = 1 2 119898 It is clear
that the VNM is able to minimize all 119873 objective functionssimultaneously in a single run
The main procedure of the VNM can be described asshown in Figure 3
In the part of parameter setting the iteration number119872 the subproblem number 119873 the size of neighborhood 119879
(which ranges from beginning size 119861 to stopping size 119878) andthe population for saving the optimal solutions EP are set
The crossover operation in VNM is as followsFor each individual 119909119905
119894in generation 119905 the child 119909
119905+1
119894can
be obtained by the following equation
119909119905+1
119894=
119909119905
119894+ 1198651times (119909119905
119894minus 119909119905
1198941) + 1198652times (119909119905
119894minus 119909119905
1198942) rand (1)ltCR
119909119905
119894rand (1)geCR
(6)
CR 1198651 and 119865
2are the three control variables for the
crossover 119909119905
1198941and 119909
119905
1198942are two individuals chosen in the
neighborhood of 119909119905119894This crossovermethod canmake full use
of the information from the neighborhood and render theinformation exchange more sufficient
Themain idea ofVNM is given above Two improvementsare involved in the VNM algorithm Variable neighborhoodstrategy is adopted to make the crossover span more rea-sonable Moreover Starting Mutation is used to enhance thediversity of the population
Mathematical Problems in Engineering 5
0 50 100 150 200 2505
10
15
20
25
30
Generations
Nei
ghbo
rhoo
d siz
e
Straight lineMonotone parabolaNonmonotone parabola
Figure 4 Three controlling curves for the neighborhood size
32 Variable Neighborhood In the VNM the size of theneighborhood 119879 has a high impact on the performance ofthe algorithm If 119879 is too large the two solutions chosen (119909119897
and 119909119896) for the genetic operation may be unsuitable for the
subproblem and degradation may occur during the progressof the evolution In contrast if119879 is too small the subproblemsare all similar The child individual will be so similar to itsparents that the crossover operation will have a weak effect
119879 is the neighborhood size which determines thecrossover and neighborhood updating span Too large andtoo small 119879 will both have a negative influence on VNMTherefore 119879 should be large enough at the beginning of theevolution period to ensure sufficient information exchange ofthe solutions and 119879 should be sufficiently small in the latterportion of the evolution period such that degradation can beavoided Motivated by this ideology we designed and testedthree curves to find the best 119879 controlling curve
The three curves are shown in Figure 4 In this figure theabscissa is the number of iterations and the ordinate is thesize of the neighborhood 1198721 1198722 and 1198723 represent thestraight line themonotonic parabolic and the nonmonotonicparabolic curves respectively It is worth noting that in curve1198722 the curvature will be 0 at the end of the evolution periodThis means that the rate of change of curvature for1198722 is thefastest of all of the concave monotone parabolas during theperiod of evolution Because the curvature goes to 0 in theend curve 1198722 is determined Assume that if the number ofiterations is 125 the neighborhood of curves 1198721 1198722 and1198723 are 119910
1 1199102 and 119910
3 respectively in accordance with the
equation 1199101minus 1199102
= 1199102minus 1199103 Thus curve 1198723 can be also
determined Curve 1198723 is a nonmonotonic parabolic curveA series of experiments should be performed to compare theinfluence of the three curves on the algorithm to identify thebest controlling curve
33 StartingMutation TheTTSP represents a typical combi-national optimization problem The final best solutions maybe limited to only several points in the solution space Becauseof the neighborhood updating effect of the VNM there will
be many duplicate solutions so that the crossover operationwill have little effectTherefore how to maintain the diversityof the population is the key question for enhancing thealgorithm effect
Motivated by the ideology above a starting Gauss muta-tion is adopted at the beginning of the iteration For a solution119909119894= (119909119894
1 119909119894
2 119909
119894
119872) (119872 is the number of variables) Gauss
mutation is described as the following
for 119895 = 1 2 119872 119909119894lowast
119895=
normal (119909119894119895 120590) rand (1) lt 119901
119909119894
119895rand (1) ge 119901
(7)
119909119894lowast
= (119909119894lowast
1 119909119894lowast
2 119909119894lowast
119872) represents the individual after muta-
tion 119901 is themutation probability normal (119909119894119895 120590) is a number
that obeys the normal distribution 119909119894119895is the mean value
and 120590 is the variance With Starting Mutation the problemwith the initially invalid crossover operation can be resolvedTherefore we can avoid the solutions from becoming trappedin local optima and thus solutions with higher quality areobtained
4 The Convergence Analysis of VNM
Theconvergence analysis of VNM in this section provides thetheory ground for its application The convergence behaviorof VNM is analyzed according to the Markov Chain and thetransfer matrix respectively
41 Strong and Weak Convergence This section proposes thebasic theories of convergence and proves the strong and weakconvergence of VNM from the perspective of Markov Chain
There is a global optimal solution set 119872 for MOPs(multiobjective problem) 119872 is defined as 119872 = 119883 forall119884 isin
119878 119891(119883) ge 119891(119884) It is assumed that (119899) is the populationin evolutionary algorithms
A detailed demonstration for the convergence of MOEAhas been proposed in paper [34] Based on it the definitionsare described as follows
Theorem 1 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
120573119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
119903119899= 12057311205732sim 120573119899
(8)
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionweakly It is defined as (119899) rarr 119872(119875119882)
Theorem 2 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
120573119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
119903119899= 12057311205732sim 120573119899
(9)
6 Mathematical Problems in Engineering
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionstrongly It is defined as (119899) rarr 119872(119875119878)
Based onTheorems 1 and 2 above the demonstration forthe convergence of VNM is described in the following Herelim119899rarrinfin
120573119899
= 0 lim119899rarrinfin
120573119899
= 0 describe the evolutionarytrend of VNMThere is lim
119899rarrinfin119903119899= 0 lim
119899rarrinfin119903119899= 0
Proof It is defined as 119875(119899) = 119875
997888
119883(119899) cap 119872 = 0Based on Bayesian we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872 = 0
= 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
+ 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
= 120572119899119875 (119899) cap 119872 = 0 + 120573
119899119875 (119899) cap 119872 = 0
(10)
Elitist strategy is adopted in VNM 120572119899= 0 Hence
119875 (119899 + 1) = 120573119899119875 (119899) (11)
Then
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0
lim119899rarrinfin
119875 (119899) cap 119872 = 0 = 1
(12)
Therefore we have
(119899) 997888rarr 119872(119875119882) (13)
It means that (119899) converges to global optimal solutionweakly
Similarly it is defined as 119875(119899) = 119875(119899) cap 119872119888
= 0By Bayesian formula we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872119888
= 0
= 120572119899119875 (119899) cap 119872
119888= 0 + 120573
119899119875 (119899)
(14)
Elitist strategy is adopted in VNM lim119899rarrinfin
120572119899= 0 Hence
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0 (15)
Then
lim119899rarrinfin
119875 (119899) cap 119872119888= 0 = 1
lim119899rarrinfin
119875 (119899) isin 119872 = 1
(16)
Therefore we have
(119899) 997888rarr 119872(119875119878) (17)
It means that (119899) converges to global optimal solutionstrongly
42 Convergence to Global Optimal This part focuses on theelitist strategy and proves that the VNM converges to theglobal optimum from the perspective of transfer matrix
Theorem 3 (see [35]) 119875 = (119862 0
119877 119879) is a reducible stochastic
matrix where 119862 119898 times 119898 is primitive stochastic matrix and119877 119879 = 0 Then
119875infin
= lim119896rarrinfin
119875119896= lim119896rarrinfin
(
119862119896
0
119896minus1
sum
119894=0
119879119894119877119862119896minus119894
119879119896)
= (
119862infin
0
119877infin
0)
(18)
where 119875infin is a stable stochastic matrix with 119875infin
= 11015840119901infin 119901infin =
1199010119875infin is unique regardless of the initial distributionThematrix
119901infin satisfies that 119901infin
119894gt 119901 for 1 le 119894 le 119898 and 119901
infin
119894= 0 for
119898 lt 119894 le 119899
According to the previous description of VNM theextended transition matrices for crossover 119862+ mutation119872
+
1
119872+
2 selection 119878
+ can be written as block diagonal matrix andupgrade matrix 119880 is lower triangular
119862+= (
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
) 119872+
1= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)
119878+= (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
) 119872+
2= (
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
119880 = (
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
(19)
119862+ 119872+1 119878+ 119872+
2 and 119880 are with 2
119899119897 square matrices 119862 1198721
1198722 119878 and119880
119886119887(1 le 119886 119887 le 2
119899119897) are all with the size of 119899times 119899 (119899is the number of individuals and 119897 is the number of individualattributes)
119886 119887 in 119880119886119887
represents the populationrsquos state sequencenumber (in the order of the populations of the pros andcons from 1 to 2
119899119897) So 119880 is used to represent populationrsquosselection process Each block matrix 119880
119886119887is a selection of
individuals The details in 119880119886119887
can be described as thereare some individuals to make 119906
119894119895= 1 established in each
row Firstly the first individual is compared with all otherindividuals 119906
1119895= 1 if 119895th individual is optimal (there may
be several optima) or 11990611
= 1 if no one is better than it Then
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 5
0 50 100 150 200 2505
10
15
20
25
30
Generations
Nei
ghbo
rhoo
d siz
e
Straight lineMonotone parabolaNonmonotone parabola
Figure 4 Three controlling curves for the neighborhood size
32 Variable Neighborhood In the VNM the size of theneighborhood 119879 has a high impact on the performance ofthe algorithm If 119879 is too large the two solutions chosen (119909119897
and 119909119896) for the genetic operation may be unsuitable for the
subproblem and degradation may occur during the progressof the evolution In contrast if119879 is too small the subproblemsare all similar The child individual will be so similar to itsparents that the crossover operation will have a weak effect
119879 is the neighborhood size which determines thecrossover and neighborhood updating span Too large andtoo small 119879 will both have a negative influence on VNMTherefore 119879 should be large enough at the beginning of theevolution period to ensure sufficient information exchange ofthe solutions and 119879 should be sufficiently small in the latterportion of the evolution period such that degradation can beavoided Motivated by this ideology we designed and testedthree curves to find the best 119879 controlling curve
The three curves are shown in Figure 4 In this figure theabscissa is the number of iterations and the ordinate is thesize of the neighborhood 1198721 1198722 and 1198723 represent thestraight line themonotonic parabolic and the nonmonotonicparabolic curves respectively It is worth noting that in curve1198722 the curvature will be 0 at the end of the evolution periodThis means that the rate of change of curvature for1198722 is thefastest of all of the concave monotone parabolas during theperiod of evolution Because the curvature goes to 0 in theend curve 1198722 is determined Assume that if the number ofiterations is 125 the neighborhood of curves 1198721 1198722 and1198723 are 119910
1 1199102 and 119910
3 respectively in accordance with the
equation 1199101minus 1199102
= 1199102minus 1199103 Thus curve 1198723 can be also
determined Curve 1198723 is a nonmonotonic parabolic curveA series of experiments should be performed to compare theinfluence of the three curves on the algorithm to identify thebest controlling curve
33 StartingMutation TheTTSP represents a typical combi-national optimization problem The final best solutions maybe limited to only several points in the solution space Becauseof the neighborhood updating effect of the VNM there will
be many duplicate solutions so that the crossover operationwill have little effectTherefore how to maintain the diversityof the population is the key question for enhancing thealgorithm effect
Motivated by the ideology above a starting Gauss muta-tion is adopted at the beginning of the iteration For a solution119909119894= (119909119894
1 119909119894
2 119909
119894
119872) (119872 is the number of variables) Gauss
mutation is described as the following
for 119895 = 1 2 119872 119909119894lowast
119895=
normal (119909119894119895 120590) rand (1) lt 119901
119909119894
119895rand (1) ge 119901
(7)
119909119894lowast
= (119909119894lowast
1 119909119894lowast
2 119909119894lowast
119872) represents the individual after muta-
tion 119901 is themutation probability normal (119909119894119895 120590) is a number
that obeys the normal distribution 119909119894119895is the mean value
and 120590 is the variance With Starting Mutation the problemwith the initially invalid crossover operation can be resolvedTherefore we can avoid the solutions from becoming trappedin local optima and thus solutions with higher quality areobtained
4 The Convergence Analysis of VNM
Theconvergence analysis of VNM in this section provides thetheory ground for its application The convergence behaviorof VNM is analyzed according to the Markov Chain and thetransfer matrix respectively
41 Strong and Weak Convergence This section proposes thebasic theories of convergence and proves the strong and weakconvergence of VNM from the perspective of Markov Chain
There is a global optimal solution set 119872 for MOPs(multiobjective problem) 119872 is defined as 119872 = 119883 forall119884 isin
119878 119891(119883) ge 119891(119884) It is assumed that (119899) is the populationin evolutionary algorithms
A detailed demonstration for the convergence of MOEAhas been proposed in paper [34] Based on it the definitionsare described as follows
Theorem 1 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
120573119899= 119875 (119899 + 1) cap 119872 = 0 (119899) cap 119872 = 0
119903119899= 12057311205732sim 120573119899
(8)
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionweakly It is defined as (119899) rarr 119872(119875119882)
Theorem 2 120572119899 120573119899 and 119903
119899are defined as
120572119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
120573119899= 119875 (119899 + 1) cap 119872
119888= 0(119899) cap 119872
119888= 0
119903119899= 12057311205732sim 120573119899
(9)
6 Mathematical Problems in Engineering
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionstrongly It is defined as (119899) rarr 119872(119875119878)
Based onTheorems 1 and 2 above the demonstration forthe convergence of VNM is described in the following Herelim119899rarrinfin
120573119899
= 0 lim119899rarrinfin
120573119899
= 0 describe the evolutionarytrend of VNMThere is lim
119899rarrinfin119903119899= 0 lim
119899rarrinfin119903119899= 0
Proof It is defined as 119875(119899) = 119875
997888
119883(119899) cap 119872 = 0Based on Bayesian we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872 = 0
= 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
+ 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
= 120572119899119875 (119899) cap 119872 = 0 + 120573
119899119875 (119899) cap 119872 = 0
(10)
Elitist strategy is adopted in VNM 120572119899= 0 Hence
119875 (119899 + 1) = 120573119899119875 (119899) (11)
Then
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0
lim119899rarrinfin
119875 (119899) cap 119872 = 0 = 1
(12)
Therefore we have
(119899) 997888rarr 119872(119875119882) (13)
It means that (119899) converges to global optimal solutionweakly
Similarly it is defined as 119875(119899) = 119875(119899) cap 119872119888
= 0By Bayesian formula we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872119888
= 0
= 120572119899119875 (119899) cap 119872
119888= 0 + 120573
119899119875 (119899)
(14)
Elitist strategy is adopted in VNM lim119899rarrinfin
120572119899= 0 Hence
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0 (15)
Then
lim119899rarrinfin
119875 (119899) cap 119872119888= 0 = 1
lim119899rarrinfin
119875 (119899) isin 119872 = 1
(16)
Therefore we have
(119899) 997888rarr 119872(119875119878) (17)
It means that (119899) converges to global optimal solutionstrongly
42 Convergence to Global Optimal This part focuses on theelitist strategy and proves that the VNM converges to theglobal optimum from the perspective of transfer matrix
Theorem 3 (see [35]) 119875 = (119862 0
119877 119879) is a reducible stochastic
matrix where 119862 119898 times 119898 is primitive stochastic matrix and119877 119879 = 0 Then
119875infin
= lim119896rarrinfin
119875119896= lim119896rarrinfin
(
119862119896
0
119896minus1
sum
119894=0
119879119894119877119862119896minus119894
119879119896)
= (
119862infin
0
119877infin
0)
(18)
where 119875infin is a stable stochastic matrix with 119875infin
= 11015840119901infin 119901infin =
1199010119875infin is unique regardless of the initial distributionThematrix
119901infin satisfies that 119901infin
119894gt 119901 for 1 le 119894 le 119898 and 119901
infin
119894= 0 for
119898 lt 119894 le 119899
According to the previous description of VNM theextended transition matrices for crossover 119862+ mutation119872
+
1
119872+
2 selection 119878
+ can be written as block diagonal matrix andupgrade matrix 119880 is lower triangular
119862+= (
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
) 119872+
1= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)
119878+= (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
) 119872+
2= (
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
119880 = (
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
(19)
119862+ 119872+1 119878+ 119872+
2 and 119880 are with 2
119899119897 square matrices 119862 1198721
1198722 119878 and119880
119886119887(1 le 119886 119887 le 2
119899119897) are all with the size of 119899times 119899 (119899is the number of individuals and 119897 is the number of individualattributes)
119886 119887 in 119880119886119887
represents the populationrsquos state sequencenumber (in the order of the populations of the pros andcons from 1 to 2
119899119897) So 119880 is used to represent populationrsquosselection process Each block matrix 119880
119886119887is a selection of
individuals The details in 119880119886119887
can be described as thereare some individuals to make 119906
119894119895= 1 established in each
row Firstly the first individual is compared with all otherindividuals 119906
1119895= 1 if 119895th individual is optimal (there may
be several optima) or 11990611
= 1 if no one is better than it Then
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
6 Mathematical Problems in Engineering
If lim119899rarrinfin
119903119899
= 0 (119899) converges to global optimal solutionstrongly It is defined as (119899) rarr 119872(119875119878)
Based onTheorems 1 and 2 above the demonstration forthe convergence of VNM is described in the following Herelim119899rarrinfin
120573119899
= 0 lim119899rarrinfin
120573119899
= 0 describe the evolutionarytrend of VNMThere is lim
119899rarrinfin119903119899= 0 lim
119899rarrinfin119903119899= 0
Proof It is defined as 119875(119899) = 119875
997888
119883(119899) cap 119872 = 0Based on Bayesian we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872 = 0
= 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
+ 119875 (119899 + 1) cap 119872 = 0(119899) cap 119872 = 0
sdot 119875 (119899) cap 119872 = 0
= 120572119899119875 (119899) cap 119872 = 0 + 120573
119899119875 (119899) cap 119872 = 0
(10)
Elitist strategy is adopted in VNM 120572119899= 0 Hence
119875 (119899 + 1) = 120573119899119875 (119899) (11)
Then
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0
lim119899rarrinfin
119875 (119899) cap 119872 = 0 = 1
(12)
Therefore we have
(119899) 997888rarr 119872(119875119882) (13)
It means that (119899) converges to global optimal solutionweakly
Similarly it is defined as 119875(119899) = 119875(119899) cap 119872119888
= 0By Bayesian formula we have
119875 (119899 + 1) = 119875 (119899 + 1) cap 119872119888
= 0
= 120572119899119875 (119899) cap 119872
119888= 0 + 120573
119899119875 (119899)
(14)
Elitist strategy is adopted in VNM lim119899rarrinfin
120572119899= 0 Hence
lim119899rarrinfin
119875 (119899 + 1) = lim119899rarrinfin
120573119899119875 (119899) = lim
119899rarrinfin119903119899119875 (0) = 0 (15)
Then
lim119899rarrinfin
119875 (119899) cap 119872119888= 0 = 1
lim119899rarrinfin
119875 (119899) isin 119872 = 1
(16)
Therefore we have
(119899) 997888rarr 119872(119875119878) (17)
It means that (119899) converges to global optimal solutionstrongly
42 Convergence to Global Optimal This part focuses on theelitist strategy and proves that the VNM converges to theglobal optimum from the perspective of transfer matrix
Theorem 3 (see [35]) 119875 = (119862 0
119877 119879) is a reducible stochastic
matrix where 119862 119898 times 119898 is primitive stochastic matrix and119877 119879 = 0 Then
119875infin
= lim119896rarrinfin
119875119896= lim119896rarrinfin
(
119862119896
0
119896minus1
sum
119894=0
119879119894119877119862119896minus119894
119879119896)
= (
119862infin
0
119877infin
0)
(18)
where 119875infin is a stable stochastic matrix with 119875infin
= 11015840119901infin 119901infin =
1199010119875infin is unique regardless of the initial distributionThematrix
119901infin satisfies that 119901infin
119894gt 119901 for 1 le 119894 le 119898 and 119901
infin
119894= 0 for
119898 lt 119894 le 119899
According to the previous description of VNM theextended transition matrices for crossover 119862+ mutation119872
+
1
119872+
2 selection 119878
+ can be written as block diagonal matrix andupgrade matrix 119880 is lower triangular
119862+= (
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
) 119872+
1= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)
119878+= (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
) 119872+
2= (
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
119880 = (
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
(19)
119862+ 119872+1 119878+ 119872+
2 and 119880 are with 2
119899119897 square matrices 119862 1198721
1198722 119878 and119880
119886119887(1 le 119886 119887 le 2
119899119897) are all with the size of 119899times 119899 (119899is the number of individuals and 119897 is the number of individualattributes)
119886 119887 in 119880119886119887
represents the populationrsquos state sequencenumber (in the order of the populations of the pros andcons from 1 to 2
119899119897) So 119880 is used to represent populationrsquosselection process Each block matrix 119880
119886119887is a selection of
individuals The details in 119880119886119887
can be described as thereare some individuals to make 119906
119894119895= 1 established in each
row Firstly the first individual is compared with all otherindividuals 119906
1119895= 1 if 119895th individual is optimal (there may
be several optima) or 11990611
= 1 if no one is better than it Then
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 7
the second individual is compared with all other individualsexcept the first individualThe best individual119892th individualis chosen set 119906
2119892= 1 if 119892th individual is optimal or 119906
22= 1 if
there is no one better than the second individualThe sortingprocess continues until all individuals are sorted To simplifythe difficulty of the problem assume that the there is only oneglobal optimal solution set Then only 119880
11is a unit matrix
whereas all matrices 119880119886119886with 119886 ge 2 are not unit matrices
In VNM the populations go through Gauss mutation119872+
1 crossover 119862
+ mutation 119872+
2 selection 119878
+ and EPupgradematrix119880 It is worth of noticing that (120583+120582) selectionmode is not used in the evolutionary process of VNMand thenumber of individuals remains unchanged This means that119878+= 119868 The transition matrix 119875
+ for VNM is
119875+= 119872+
1119862+119872+
2119878+119880
= (
1198721
sdot sdot sdot 0
d
0 sdot sdot sdot 1198721
)(
119862 sdot sdot sdot 0
d
0 sdot sdot sdot 119862
)(
1198722
sdot sdot sdot 0
d
0 sdot sdot sdot 1198722
)
times (
119878 sdot sdot sdot 0
d
0 sdot sdot sdot 119878
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
11987211198621198722
sdot sdot sdot 0
d
0 sdot sdot sdot 11987211198621198722
)(
11988011
sdot sdot sdot 0
d
11988021198991198971
sdot sdot sdot 11988021198991198972119899119897
)
= (
1198721119862119872211988011
0 0
sdot sdot sdot 0
1198721119862119872211988021198991198971
sdot sdot sdot 1198721119862119872211988021198991198972119899119897
)
= (
11987511
sdot sdot sdot 0
d
11987521198991198971
sdot sdot sdot 11987521198991198972119899119897
)
(20)
There is 11987511
gt 0 in the transition matrix 119875+ The submatrices
119875119886119897which is with 119886 ge 2 may be gathered in a rectangular
matrix 119877 = 0 so that Theorem 3 can be used to prove that thecorresponding VNM converges to the global optimum [36]
5 Experimental Results and Analysis
Computational experiments are carried out to compare theapproaches and to evaluate the efficiency of the proposedmethodThere are two objectives to minimize the makespanand the mean workload of the instruments In this sectionthe performance metric coverage metric 119862 is introducedfirst There are two experimental instances adopted in thissectionThey are instances of 30 taskswith 12 instruments and40 tasks with 12 instruments which are real-world examplestaken from a missile system The instance of 40 tasks with12 instruments is displayed in Table 1 The instance of 30
Table 1 The instance of 40 tasks with 12 instrumentsTask Scheme Resource Time
1199051
1199081
11199031 1199037
51199082
11199033 1199035
51199083
11199036 11990310
4
1199052
1199081
21199032 11990311
51199082
21199034 1199039
41199083
21199035 1199036
61199084
21199033 1199037
4
1199053
1199081
31199033
71199082
311990312
5
1199054
1199081
41199039
251199082
411990310
221199055
1199081
511990312
14
1199056
1199081
61199031 1199034
71199082
61199033 1199037
81199083
61199036 1199038
8
1199057
1199081
71199031 1199032
41199082
71199033 1199038
21199083
71199037 11990311
3
1199058
1199081
81199031 1199033
51199082
81199036 11990310
41199083
81199037 11990312
7
1199059
1199081
91199031 1199034
111199082
91199037 1199039
131199083
91199038 11990311
12
11990510
1199081
101199032
91199082
101199034
101199083
1011990310
10
11990511
1199081
111199032 1199037
61199082
111199033 11990312
91199083
111199038 1199039
8
11990512
1199081
121199032
111199082
121199035
131199083
1211990311
15
11990513
1199081
131199032
41199082
131199038
51199083
131199039
7
11990514
1199081
141199033
71199082
1411990311
101199083
1411990312
811990515
1199081
1511990312
2
11990516
1199081
161199032
91199082
161199035
71199083
161199038
6
11990517
1199081
171199031 11990310
101199082
171199035 1199039
121199083
1711990311 11990312
1111990518
1199081
181199036
15
11990519
1199081
191199032
81199082
191199035
71199083
1911990310
71199084
1911990312
6
11990520
1199081
201199033
61199082
201199036
41199083
201199039
5
11990521
1199081
211199031 1199034
21199082
211199033 1199035
51199083
211199036 1199038
3
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
8 Mathematical Problems in Engineering
Table 1 Continued
Task Scheme Resource Time
11990522
1199081
221199032
31199082
221199034
41199083
221199036
31199084
2211990310
4
11990523
1199081
231199033
51199082
2311990312
5
11990524
1199081
241199034
141199082
2411990311
1711990525
1199081
251199037
19
11990526
1199081
261199031 1199034
71199082
261199033 1199037
81199083
261199036 1199038
10
11990527
1199081
271199031 1199032
21199082
271199031 1199037
21199083
271199033 1199038
4
11990528
1199081
281199031 1199033
51199082
281199034 1199035
41199083
281199037 11990312
2
11990529
1199081
291199031 1199034
111199082
291199033 1199034
151199083
291199037 1199038
12
11990530
1199081
301199031
91199082
301199034
121199083
3011990312
10
11990531
1199081
311199032 1199033
61199082
311199035 11990311
81199083
311199036 1199039
8
11990532
1199081
321199032
111199082
321199035
131199083
321199036
17
11990533
1199081
331199032
61199082
331199036
51199083
3311990311
4
11990534
1199081
341199033
71199082
341199037
81199083
3411990312
1011990535
1199081
351199039
2
11990536
1199081
361199032
91199082
361199035
71199083
3611990310
6
11990537
1199081
371199031 1199032
101199082
371199037 11990311
71199083
371199035 11990312
1111990538
1199081
3811990310
15
11990539
1199081
391199034
81199082
391199036
71199083
391199039
71199084
3911990310
6
11990540
1199081
401199033
61199082
401199036
51199083
401199039
5
Table 2 Parameters setting
Population Generation CR 1198651 1198652 119875
100 250 05 1 1 005
tasks with 12 instruments is the first 30 tasks in Table 1 Theexperiment of selection of controlling curve for neighbor-hood size is shown in Section 52 The verification of theimprovements of the algorithm is displayed in Section 53 InSection 53 VNM is comparedwithMOEAD In Section 54the proposed algorithm (VNM) is compared with the varia-tions of CNSGA using real-world TTSP problems All of thealgorithms are executed using 50 independent runs In allof the experiments the better performances are denoted inbold The basic algorithm parameter settings are displayed inTable 2 CR 119865
1 and 119865
2are the three control variables for the
crossover 119901 is the mutation probability
51 PerformanceMetric Formultiobjective optimization theconvergence to the Pareto-optimal set is the most importanttarget to be considered There are mainly two metrics toevaluate the convergence One is convergence metric 120574 andthe other is convergence metric 119862 The true set of Pareto-optimal solutions is necessary for the calculation of 120574 How-ever the solutions space of TTSP is so large that the true set ofPareto-optimal solutions cannot be obtained by enumerationThe metric 119862 can be used to compare the performancesof the two solutions sets obtained by different algorithmsThe calculation of 119862 needs only the information of thetwo solutions sets Therefore in this paper the convergencemetric 119862 is used to evaluate the performance of the proposedalgorithm
Assume that 119860 and 119861 are two sets of nondominatedsolutions and 119862(119860 119861) is the ratio of the solutions in 119861 thatare dominated by at least one solution in 119860 Hence
119862 (119860 119861) =
1003816100381610038161003816119909 isin 119861 | exist119910 isin 119860 119910 dominates 119909
1003816100381610038161003816
|119861|
(21)
119862(119860 119861) = 1means that all of the solutions in119861 are dominatedby solutions in 119860 and 119862(119860 119861) = 0 means that there isno solution in 119861 dominated by a solution in 119860 Generallyspeaking if 119862(119860 119861) gt 119862(119861 119860) then solution set 119860 is betterthan solution set 119861
52 The Selection of Controlling Curve In this sectionthree curves are designed and tested to identify the best 119879controlling curve 1198721 1198722 and 1198723 respectively representthe straight line monotonic parabolic and nonmonotonicparabolic curves shown in Figure 4 In curve 1198722 the cur-vature will be 0 at the end of the evolution period Becauseof the influence of neighboring updating in MOEAD manyduplicate solutions will be presented in the final evolutionprocess of MOEAD Therefore Starting Mutation is appliedto the beginning of the next iteration to maintain the
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 9
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 30lowast12
Figure 5 The boxplot of three curves for 30 lowast 12 instance
Table 3 Comparison of influence of three curves for 30 lowast 12
instance
Average Times119862(11987211198722) 02213 13119862(11987221198721) 05196 37119862(11987221198723) 04964 36119862(11987231198722) 02069 14
population diversity Tables 3 and 4 show the comparisonof the influence of the three curves on the algorithm usingtwo instancesThe results show that the monotonic paraboliccurve 1198722 has the best performance This means that themonotonic curve with the fastest rate of change of curvatureis themost useful for the algorithm And the boxplots of threecurves for 30 lowast 12 and 40 lowast 12 instances in Figures 5 and 6also give the same conclusion
53 Experiments for Comparisons of VNM and MOEADIn order to verify the improvement of VNM 30 lowast 12 and40 lowast 12 instances are used to test the performance of VNMandMOEADThemonotonic parabolic curve1198722 is selectedas the controlling curve in VNM The neighborhood size inMOEAD is 20 119881 and 119872 respectively represent VNM andMOEADThe results in Tables 5 and 6 show that the concavecurve with the fastest rate of change of curvature obtainedimprovement for VNM The selected curve renders the sizeof the neighborhood more suitable than before
The results of the two independent experiments forcomparison of VNM and MOEAD are shown in Figures 7and 8 for the 30 lowast 12 and 40 lowast 12 instances respectivelyAs shown in the figures the solutions obtained by theVNM dominate most of the solutions obtained by MOEADVariable neighborhood and Starting Mutation improve theperformance of MOEAD efficiently
Figures 9 and 10 are the boxplots for comparison of VNMand MOEAD It shows that the data distribution of VNMis superior to MOEAD VNM has the better performance
0010203040506070809
1
C(M2M1) C(M1M2) C(M2M3) C(M3M2)
C-m
etric
TTSP 40lowast12
Figure 6 The boxplot of three curves for 40 lowast 12 instance
Table 4 Comparison of influence of three curves for 40 lowast 12
instance
Average Times119862(11987211198722) 0244178 14119862(11987221198721) 0501508 36119862(11987221198723) 0533806 38119862(11987231198722) 0242146 12
Table 5 Comparison of VNM and MOEAD for 30 lowast 12 instance
Average Times119862(119881119872) 04845 35119862(119872119881) 02104 15
Table 6 Comparison of VNM and MOEAD for 40 lowast 12 instance
Average Times119862(119881119872) 05256 40119862(119872119881) 01949 10
because of application of variable neighborhood and StartingMutation
54 Experiments for Comparisons of VNM and CNSGA Inthis section the VNM is compared with the CNSGA forTTSP CNSGA is based on NSGA-II NSGA-II has beensuccessfully applied to job shop scheduling problems [37]reactive power dispatch problems [38] and many otherapplications CNSGA has successfully been adopted to solveTTSP [20] Therefore a comparison of VNM and CNSGA iscarried out to test the performance of the proposed algorithmVNM
There are two chaotic sequences logistic map and catmap and the chaotic sequences can be applied in threepositions population initialization crossover and mutationTherefore there are six combinations for CNSGA Thenomenclatures for six variants of CNSGA are shown in
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
10 Mathematical Problems in Engineering
40 50 60 70 80 90 10016
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(a)
40 50 60 70 80 9016
17
18
19
20
21
22
23
24
25
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 30lowast12
(b)
Figure 7 The comparison of VNM and MOEAD for 30 lowast 12 instance
50 55 60 65 70 75 80 85 9017
18
19
20
21
22
23
24
25
26
27
Makespan
Mea
n w
orkl
oad
TTSP
VNMMOEAD
40lowast12
(a)
50 55 60 65 70 75 8017
18
19
20
21
22
23
Makespan
Mea
n w
orkl
oad
VNMMOEAD
TTSP 40lowast12
(b)
Figure 8 The comparison of VNM and MOEAD for 40 lowast 12 instance
Table 7 Tables 8 and 9 show the comparison of VNM andCNSGA for 30 lowast 12 and 40 lowast 12 instances The VNM isrepresented by 119881 All the comparisons between VNM andthe variations of CNSGA are based on 50 independent exper-iments The average of 119862 metric and better computationaltimes are used for the performance analysis The results fromTables 8 and 9 show that VNMprovides the best performancenot only for the average metric 119862 but also in terms of bettercomputational times than CNSGA
Figures 11 and 12 show the comparisons of VNM and the6 variations of CNSGA for 30lowast 12 instanceThe figures show
Table 7 Nomenclature for six variants of the CNSGA
The logistic map The cat mapInitial population 119871
11198621
Crossover operator 1198712
1198622
Mutation operator 1198713
1198623
that the solutions obtained by the VNMdominatemost of thesolutions obtained by the 6 variations of CNSGA Thereforethe VNM obtains the best performance
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 11
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
30lowast12
Figure 9 The boxplot of comparison of VNM and MOEAD for30 lowast 12 instance
0010203040506070809
1
C-m
etric
TTSP
C(VM) C(MV)
40lowast12
Figure 10 The boxplot of comparison of VNM and MOEAD for40 lowast 12 instance
The results of the comparison of VNM and CNSGA forthe 40 lowast 12 instance are shown in Figures 13 and 14The resultsalso show that the solutions obtained by VNM have higherquality
In addition box plots are used to display the perfor-mances of the algorithmsThe box plots of119862metric for VNMandCNSGAwith 30lowast12 and 40lowast12 instances are shown fromFigures 15 16 17 and 18
From the box plots of 119862metric it is clear that the medianof VNM is larger than that of the variations of CNSGA inboth the 30 lowast 12 and 40 lowast 12 instances and the datadistribution of VNM ismore reasonable as well Additionallythe average of VNM is also superior The results showthat VNM demonstrates better performance than CNSGAin solving the multiobjective TTSP The performance ofsolutions obtained by VNM is better than that obtained byCNSGA because that the variable neighborhood is adoptedin VNMThe span of information exchange in VNM changesfollowing the evolutionary process but that in CNSGA stays
Table 8 Comparison of VNM and six variations of CNSGA for 30lowast 12 instance
Average Times119862(119881 119871
1) 04206 36
119862(1198711 119881) 02200 14
119862(119881 1198712) 04077 34
119862(1198712 119881) 02648 16
119862(119881 1198713) 04182 36
119862(1198713 119881) 02248 14
119862(119881 1198621) 04638 35
119862(1198621 119881) 02210 15
119862(119881 1198622) 04602 34
119862(1198622 119881) 02288 16
119862(119881 1198623) 04128 35
119862(1198623 119881) 02525 15
Table 9 Comparison of VNM and six variations of CNSGA for 40lowast 12 instance
Average Times119862(119881 119871
1) 05243 37
119862(1198711 119881) 02264 13
119862(119881 1198712) 05359 36
119862(1198712 119881) 02282 14
119862(119881 1198713) 05218 38
119862(1198713 119881) 02338 12
119862(119881 1198621) 05044 36
119862(1198621 119881) 02138 14
119862(119881 1198622) 04844 35
119862(1198622 119881) 02169 15
119862(119881 1198623) 05116 37
119862(1198623 119881) 02055 13
the sameThe information from the process of evolution helpsVNM get better performance
The variable neighborhood provides a strategy to improvethe performance of the algorithm For different problemswith different scales the controlling curves for the neigh-borhood size will be different The Starting Mutation canbe also applied to solve other optimization problem in theevolution process The strategies proposed in this paper canbe investigated in other scheduling problem similar to TTSP
6 Conclusion
How to improve the test efficiency is more and more impor-tant in modern industry TTSP has important application
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
12 Mathematical Problems in Engineering
40 50 60 70 80 90 100 11012
14
16
18
20
22
24
26
28
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-L1
CNSGA-L2
CNSGA-L3
30lowast12
Figure 11 The comparison of VNM and three variations withlogistic map for CNSGA for 30 lowast 12 instance
30 40 50 60 70 80 90 100 11014
16
18
20
22
24
26
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
30lowast12
Figure 12 The comparison of VNM and three variations with catmap for CNSGA for 30 lowast 12 instance
value in modern manufacturing process TTSP is combina-tional optimization problem The final best solutions onlyaccount for a rather small subset of the search space Inorder to help the solutions avoid being trapped in localoptima this paper proposed a new genetic evolutionarymultiobjective optimization algorithm (VNM) to solve theTTSP The variable neighborhood and Starting Mutationstrategies are adopted in VNM to make the crossover spanmore suitable and improve the diversity of populationThree controlling curves for neighborhood size are studiedThe experimental results have shown that the monotonic
40 50 60 70 80 90 100 11014
16
18
20
22
24
26
28
30
32
Makespan
Mea
n w
orkl
oad
VNMCNSGA-
CNSGA-CNSGA-
TTSP
L1
L2
L3
40lowast12
Figure 13 The comparison of VNM and three variations withlogistic map for CNSGA for 40 lowast 12 instance
40 50 60 70 80 90 100 110 12016
18
20
22
24
26
28
30
Makespan
Mea
n w
orkl
oad
VNM
TTSP
CNSGA-C1
CNSGA-C2
CNSGA-C3
40lowast12
Figure 14 The comparison of VNM and three variations with catmap for CNSGA for 40 lowast 12 instance
parabolic has the best performance From the experimentconducted for comparison of VNM and MOEAD we seethat the improved algorithm has made great progress insolving the TTSP problem And the experiment conductedfor comparison of VNM and CNSGA also shows that theimproved algorithm is superior to CNSGA in solving TTSPVNM can also be applied to solve other combinationaloptimization problems such as FJSP and TSP Future workwill focus on two objectives the precedence constraint willbe added to the TTSP and information regarding bottlenecktasks will be considered
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Mathematical Problems in Engineering 13
TTSP
0010203040506070809
1
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
30lowast12
Figure 15 The boxplot of VNM and three variations with logisticmap for CNSGA for 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
30lowast12
Figure 16 The boxplot of VNM and three variations with cat mapfor CNSGA for TTSP 30 lowast 12 instance
0010203040506070809
1TTSP
C-m
etric
C(V L1) C(V L2) C(V L3)C(L1 V) C(L2 V) C(L3 V)
40lowast12
Figure 17 The boxplot of VNM and three variations with logisticmap for CNSGA for 40 lowast 12
TTSP
0010203040506070809
1
C-m
etric
C(V C1) C(V C2) C(V C3)C(C1 V) C(C2 V) C(C3 V)
40lowast12
Figure 18 The boxplot of VNM and three variations with cat mapfor CNSGA for 40 lowast 12
Conflict of Interests
The authors declare that there is no conflict of interestsregarding the publication of this paper
Acknowledgments
This research is supported by the National Natural Sci-ence Foundation of China under Grant no 61101153 andthe National 863 Hi-Tech R and D Plan under Grant2011AA110101
References
[1] A Radulescu C Nicolescu A J C van Gemund and P PJonker ldquoCPR mixed task and data parallel scheduling for dis-tributed systemsrdquo in Proceeding of 15th Parallel and DistributedProcessing Symposium pp 1ndash9 San Francisco Calif USA April2001
[2] Z Yin J Cui Y Yang and YMa ldquoJob shop scheduling problembased on DNA computingrdquo Journal of Systems Engineering andElectronics vol 17 no 3 pp 654ndash659 2006
[3] Y Zuo H Y Gu and Y G Xi ldquoModified bottleneck-basedheuristic for large-scale job-shop scheduling problem with asingle bottleneckrdquo Journal of Systems Engineering and Electron-ics vol 18 no 3 pp 556ndash565 2007
[4] N Al-Hinai and T Y Elmekkawy ldquoRobust and stable flexiblejob shop scheduling with random machine breakdowns usinga hybrid genetic algorithmrdquo International Journal of ProductionEconomics vol 132 no 2 pp 279ndash291 2011
[5] J C Chen C-C Wu C-W Chen and K-H Chen ldquoFlexiblejob shop scheduling with parallel machines using geneticalgorithm and grouping genetic algorithmrdquo Expert Systems withApplications vol 39 no 11 pp 10016ndash10021 2012
[6] L De Giovanni and F Pezzella ldquoAn improved genetic algorithmfor the distributed and flexible job-shop scheduling problemrdquoEuropean Journal of Operational Research vol 200 no 2 pp395ndash408 2010
[7] D Lei ldquoCo-evolutionary genetic algorithm for fuzzy flexible jobshop schedulingrdquo Applied Soft Computing Journal vol 12 no 8pp 2237ndash2245 2012
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
14 Mathematical Problems in Engineering
[8] R Zhang and C Wu ldquoA hybrid immune simulated annealingalgorithm for the job shop scheduling problemrdquo Applied SoftComputing Journal vol 10 no 1 pp 79ndash89 2010
[9] P Damodaran andM C Velez-Gallego ldquoA simulated annealingalgorithm to minimize makespan of parallel batch processingmachines with unequal job ready timesrdquo Expert Systems withApplications vol 39 no 1 pp 1451ndash1458 2012
[10] K Li Y Shi S-L Yang and B-Y Cheng ldquoParallel machinescheduling problem to minimize the makespan with resourcedependent processing timesrdquo Applied Soft Computing Journalvol 11 no 8 pp 5551ndash5557 2011
[11] Q Zhang H Manier and M-A Manier ldquoA genetic algorithmwith tabu search procedure for flexible job shop schedulingwith transportation constraints and bounded processing timesrdquoComputers and Operations Research vol 39 no 7 pp 1713ndash17232012
[12] W Teekeng and A Thammano ldquoA combination of shuffledfrog leaping and fuzzy logic for flexible job-shop schedulingproblemsrdquo Procedia Computer Science vol 6 pp 69ndash75 2011
[13] G Zhang X Shao P Li and LGao ldquoAn effective hybrid particleswarm optimization algorithm for multi-objective flexible job-shop scheduling problemrdquo Computers and Industrial Engineer-ing vol 56 no 4 pp 1309ndash1318 2009
[14] S H A Rahmati and M Zandieh ldquoA new biogeography-based optimization (BBO) algorithm for the flexible job shopscheduling problemrdquo The International Journal of AdvancedManufacturing Technology vol 58 no 9ndash12 pp 1115ndash1129 2012
[15] X Y Shao W Q Liu Q Liu and C Y Zhang ldquoHybriddiscrete particle swarmoptimization formulti-objective flexiblejob-shop scheduling problemrdquo The International Journal ofAdvancedManufacturing Technology vol 67 no 9ndash12 pp 2885ndash2901 2013
[16] R Xia M Xiao J Cheng and X Fu ldquoOptimizing the multi-UUT parallel test task scheduling based on multi-objectiveGASArdquo in Proceedings of the 8th International Conference onElectronicMeasurement and Instruments (ICEMI rsquo07) pp 4839ndash4844 Xirsquoan China August 2007
[17] J Y Fang H H Xue and M Q Xiao ldquoParallel test tasksscheduling and resources configuration based on GA-ACArdquoJournal of Measurement Science and Instrumentation vol 2 no4 pp 321ndash326 2011
[18] H Lu X Chen and J Liu ldquoParallel test task scheduling withconstraints based on hybrid particle swarm optimization andtaboo searchrdquo Chinese Journal of Electronics vol 21 no 4 pp615ndash618 2012
[19] X Liang B G Dong H Gao and D S Yan ldquoParallel test taskscheduling of aircraft electrical system based on cost constraintmatrix and ant colony algorithmrdquo in Proceeding of IEEE 10thInternational Conference on Industrial Informatics pp 178ndash183Beijing Chinese July 2012
[20] H Lu R Y Niu J Liu and Z Zhu ldquoA chaotic non-dominatedsorting genetic algorithm for the multi-objective automatic testtask scheduling problemrdquoApplied Soft Computing vol 13 no 5pp 2790ndash2802 2013
[21] K Deb A Pratap S Agarwal and T Meyarivan ldquoA fastand elitist multiobjective genetic algorithm NSGA-IIrdquo IEEETransactions on Evolutionary Computation vol 6 no 2 pp 182ndash197 2002
[22] B Show V Mukherjee and S P Ghoshal ldquoSolution of reactivepower dispatch of power systems by an opposition-basedgravitational search algorithmrdquo Electrical Power and EnergySystems vol 55 pp 29ndash40 2014
[23] L Wang and C Singh ldquoEnvironmentaleconomic power dis-patch using a fuzzifiedmulti-objective particle swarmoptimiza-tion algorithmrdquo Electric Power Systems Research vol 77 no 12pp 1654ndash1664 2007
[24] O Abedinia D Garmarodi R Rahbar and F JavidzadehldquoMulti-objective environmentaleconomic dispatch using inter-active artificial bee colony algorithmrdquo Journal of Basic andApplied Scientific Research vol 2 no 11 pp 11272ndash11281 2012
[25] Q Zhang and H Li ldquoMOEAD a multiobjective evolutionaryalgorithm based on decompositionrdquo IEEE Transactions onEvolutionary Computation vol 11 no 6 pp 712ndash731 2007
[26] Y-Y Tan Y-C Jiao H Li and X-K Wang ldquoMOEAD+uniform design a new version of MOEAD for optimizationproblems with many objectivesrdquo Computers and OperationsResearch 2012
[27] C-M Chen Y-P Chen and Q Zhang ldquoEnhancing MOEADwith guided mutation and priority update for multi-objectiveoptimizationrdquo in Proceedings of the IEEE Congress on Evolution-ary Computation (CEC rsquo09) pp 209ndash216 Trondheim NorwayMay 2009
[28] M A Jan and R A Khanum ldquoA study of two penalty-parameterless constraint handling techniques in the frameworkofMOEADrdquoApplied Soft Computing vol 13 no 1 pp 128ndash1482012
[29] W Peng Q Zhang and H Li ldquoComparison betweenMOEADand NSGA-II on the multi-objective travelling salesman prob-lemrdquo Studies in Computational Intelligence vol 171 pp 309ndash3242009
[30] P A Carl Communication with Automata Applied DataResearch Princeton NJ USA 1966
[31] T Zhang B Guo and Y Tan ldquoCapacitated stochastic colouredPetri net-based approach for computing two-terminal reliabilityof multi-state networkrdquo Journal of Systems Engineering andElectronics vol 23 no 2 pp 304ndash313 2012
[32] B Bollobas GraphTheory Springer New York NY USA 1979[33] K Miettinen Nonlinear Multiobjective Optimization Kluwer
Academic Boston Mass USA 1999[34] Y-R Zhou H-Q Min X-Y Xu and Y-X Li ldquoMulti-objective
evolutionary algorithm and its convergencerdquo Chinese Journal ofComputers vol 27 no 10 pp 1415ndash1421 2004
[35] M Iosifescu Finite Markov Processes and Their ApplicationsDover 1980
[36] Q Zhang L Dong F Jiang and X J Zhu ldquoConvergence ofmulti-objective evolutionary computation to its pareto optimalsetrdquo Systems Engineering and Electronics vol 22 no 8 pp 17ndash212000
[37] E Moradi S M T Fatemi Ghomi and M Zandieh ldquoBi-objective optimization research on integrated fixed time inter-val preventive maintenance and production for schedulingflexible job-shop problemrdquo Expert Systems with Applicationsvol 38 no 6 pp 7169ndash7178 2011
[38] S Jeyadevi S Baskar C K Babulal and M Willjuice Irutha-yarajan ldquoSolving multiobjective optimal reactive power dis-patch usingmodifiedNSGA-IIrdquo International Journal of Electri-cal Power and Energy Systems vol 33 no 2 pp 219ndash228 2011
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of
Submit your manuscripts athttpwwwhindawicom
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical Problems in Engineering
Hindawi Publishing Corporationhttpwwwhindawicom
Differential EquationsInternational Journal of
Volume 2014
Applied MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Probability and StatisticsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Mathematical PhysicsAdvances in
Complex AnalysisJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
OptimizationJournal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
CombinatoricsHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Operations ResearchAdvances in
Journal of
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Function Spaces
Abstract and Applied AnalysisHindawi Publishing Corporationhttpwwwhindawicom Volume 2014
International Journal of Mathematics and Mathematical Sciences
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
The Scientific World JournalHindawi Publishing Corporation httpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Algebra
Discrete Dynamics in Nature and Society
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Decision SciencesAdvances in
Discrete MathematicsJournal of
Hindawi Publishing Corporationhttpwwwhindawicom
Volume 2014 Hindawi Publishing Corporationhttpwwwhindawicom Volume 2014
Stochastic AnalysisInternational Journal of