multi-agent modelling for simulation of customer-centric supply …acl 06... · 2018-02-18 ·...

1 International Journal of Simulation & Process Modelling, Vol. 2, Nos. 3/4, 2006

Multi-Agent Modelling forSimulation of Customer-CentricSupply ChainOlivier Labarthe*, Alain Ferrarini, Bernard EspinasseLSIS, Laboratoire des Sciences de l’Information et des Systèmes,Université Aix-Marseille III, 13397, Marseille – FranceE-mail: [email protected] E-mail: [email protected]:[email protected]*Corresponding author

Benoit MontreuilCENTOR, Network Organization Technology Research Center,Université Laval, Pavillon Palasis Prince, G1K 7P4, Québec, Qc – CanadaE-mail: [email protected]

Abstract: The analysis of coordination among the supply chain members is a key factor forimproving its performance. The highly dynamic demand induced by the volatility ofcustomers requires defining tools to improve the performance. This paper presents anagent-based modelling concept for simulation of customer-centric supply chain. Wedescribe the specificities of such supply chains at the enterprise modelling domain level andpropose an approach permitting to explicitly take into account the customers and theirbehaviours. The approach uses three levels of modelling, successively leading to a domainmodel, a conceptual agent model and an operational agent model. A case study from thepersonalized golf club industry illustrates the proposed approach.

Keywords: Customer-Centric Supply Chain, Multi-Agents Systems, Agent-OrientedSimulation

Reference to this paper should be made as follows: Labarthe, O., Ferrarini, A., Espinasse,B. and Montreuil, B. (2006) ‘Multi-Agent Modelling for Simulation of Customer-CentricSupply Chain’, International Journal of Simulation & Process Modelling, Vol. 2, Nos. 3/4,pp. XX–XX.

Biographical notes: O. Labarthe is currently a PhD candidate in computer science at theUniversity of Aix-Marseille III and in operations management at the Université Laval. Hereceived a BSc in Industrial Engineering from the University of Bordeaux I and a MSc inProduction System Computing from the University of Pau et Pays de l’Adour.

A. Ferrarini is an Associate Professor of Industrial Engineering and Computer Science inthe Ecole Polytechnique Universitaire de Marseille since 1997 and a researcher at LSISresearch laboratory. He has an Engineer diploma in Automatics and Computer Science, anda PhD in Automation from the University of Aix-Marseille III.

B. Espinasse received an engineer diploma from Ecole Nationale Supérieure d'Arts etMetiers (1977) and a Doctorate in Information Systems from Aix-Marseille University(1981). He is currently a Full Professor in Computer Sciences at Aix-Marseille Universityand a team leader at LSIS research laboratory. He was Associate Professor at the UniversitéLaval (1983-1987). His main research topics are Agent-Based Simulation, DecisionSupport Systems and Information Systems Engineering.

B. Montreuil is Professor of Operations and Decision Systems in the Faculty ofAdministration Sciences at the Université Laval since 1988. He holds both the CanadaResearch Chair in Enterprise Engineering and the NSERC/Bell/Cisco Research Chair inBusiness Design. He is a founding member of the research center CENTOR. He has aMSIE and a PhD in Industrial Engineering from the Georgia Institute of Technology. Hehas extensive research and consulting experience in design, optimization and agent basedmodelling and simulation of demand and supply chains.

MULTI-AGENT MODELLING FOR SIMULATION OF CUSTOMER-CENTRIC SUPPLY CHAIN 2

1 INTRODUCTION

In the increasingly competitive context of globalization, it isbecoming more and more difficult for companies topreserve and conquer market shares. Strategically it isimportant for them to insure that their Supply Chain(Ganeshan et al., 1998) becomes an integrated networkedorganization capable of responding to the volatility ofcustomers needs. To grasp these needs, the structure of theirSupply Chain (SC) has to be adapted to targeted product /market segments and the overall management of the SC hasto become demand driven and customer centric (growingvariability of products adapted to numerous differentiatedcategories of customers). Such a customer-centric SCbecomes characterized by a highly dynamic demand.

This context requires companies to develop flexiblecapabilities enabling them fast reactivity to changes in theenvironment. To deliver customizable products withinshorter delays it is necessary for them to develop adaptivestrategies for resource allocation and flow management.Such adaptive strategies may include combinations ofMake-To-Order, To-Stock, … Overall this has a significantimpact on both the structure and dynamics of the SC.

Estimating the future key performance indicators of aclient centric SC in a highly dynamic context is important toinsure that client expectations are met profitably. It is keywhen a SC transformation project is proposed andimplemented. It is also key to help determine up to whichconditions the current SC can perform adequately, and thenwhich elements should be improved or altered to go beyondthat. Classic forecasting models turn out to have limitedapplicability for such performance estimation (Fisher,1997). In order to estimate the key performance indices of acustomer-centric SC, it is necessary to develop elaboratetools such as a SC simulator. However building a tool tosimulate the behaviour of the SC in a customer-centriccontext is not a trivial matter. It requires the elaboration of arepresentative model and the execution of this modelaccording to a set of hypotheses (Zeigler et al., 2000).

Multi-Agent Systems, developed in the DistributedArtificial Intelligence field, seem particularly well suited forthe modelling and the simulation of SC. This leads us tosuggest an agent-based modelling and simulation approachto facilitate the management of Customer-Centric SC.

In section two, we present a review of previous researchon agent-based modelling and simulation of SC. In sectionthree, we define at the enterprise modelling level thespecificities of customer-centric SC. In order to explicitlytake into account the customers and their behaviours wepropose a domain representation of such networks. A casestudy from the golf club industry illustrating this approachis then presented. Section four presents an agent-basedmodel of customer-centric SC, derived from the domainrepresentation and composed of two models: a conceptualagent model and an operational agent model. We emphasizethe modelling and the simulation of central actors in the SC:the customers. Finally, we present conclusive remarks andavenues for further research.

2 AGENT-BASED MODELLING AND SIMULATION OFSUPPLY CHAINS

Parunak et al. (1998) contrast three main modellingapproaches: (i) the Control Theory approach, based ondifferential equations, (ii) the Operational Researchapproach, which relies on optimization theories, and (iii) theSimulation approach. Simulation relies on experimentationthrough executable models for understanding and predictingthe behaviour of systems (Courtney et al., 1997). It has beenrecognized for its capability to allow more realisticobservation of the SC behaviours and dynamics over time. Ithelps to understand the organizational decision-makingprocesses and to analyze the dynamic interdependenciesbetween actors and the consistency of coordination modesthrough the SC.

When using the Simulation approach the literaturedistinguishes between two main types of modelling(Parunak et al., 1998): equation-based modelling and agent-based modelling. The former represents observables asquantifiable characteristics of one or more individuals. Itexpresses the relations between the observables throughequations. The model is executed by iteratively evaluatingthese equations and observing the evolution of theobservables. The latter approach encapsulates in individuals(agents) the behaviour of each actor of the system. Theexecution of the model is a behavioural simulation, lettingthe agents interact with each other and the environment, andmonitoring their behaviour and the observables affected byagent actions. Several research studies in the correlatedmanufacturing and SC domains (Shen and Norrie, 1999;Strader et al., 1999; Fox et al., 2000; Sadeh et al., 2003)point to the fact that the agent-based modelling approachmay be well suited for realizing models enabling theanalysis of the SC behaviour.

2.1 The Agent approach

An agent can be defined as an entity, theoretical, virtual orphysical, capable of acting on itself and on the environmentin which it evolves and to communicate with other agents,(Jennings et al., 1998). Its behaviour is the consequence ofits observations, knowledge and interactions. Wooldridgeand Jennings (1995) propose the following properties of anagent:• Autonomy: an agent operates without human being or

other direct intervention and neither the actions itrealizes nor its internal state are submitted to anycontrol.

• Reactivity: an agent perceives its environment andreacts in an appropriate way.

• Pro-activi ty: an agent must be able to developbehaviours directed by internal goals.

• Sociability: the agents interact with each other usingcommunication languages and common sociabilityrules.

Two kinds of agents are distinguished according to theirdegree of "intelligence": the reactive and cognitive agents.

3 O. LABARTHE, A. FERRARINI, B. ESPINASSE AND B. MONTREUIL

The reactive agents follow the stimulus/action law and usereduced communication languages. The cognitive agentshave reasoning capabilities and have an explicitrepresentation of the environment in which they evolve(Moulin and Chaib-Draa, 1996).

2.2 Relevance of the agent approach

Multi-Agent Systems (MAS) and SC can both be seen asnetworks of entities which cooperate in order to solveproblems that are beyond their individual abilities orknowledge (Wu et al., 2000). The analogical thread betweenMAS and SC relates to the coordination needs andcapabilities, between agents in a MAS and between actors ina SC. Table 1 summarizes important analogies betweenMAS and SC.

Table 1Analogies between SC and MAS, adapted from(Yuan et al., 2002)

Both MAS and SC are composed of entities which interactaccording to their roles and abilities within organizationalstructures. Agents and SC actors have means, resources andcapabilities allowing them to carry out various functions,tasks or activities in an individual or collective way. Theseanalogies naturally lead to the use of the Multi-Agentapproach to represent SC. Empirical evidence shows thatMAS make the study by simulation of the SC more realistic,as reported through many projects of agent-based modellingand simulation (Kjenstad, 1998; Strader et al., 1999; Fox etal., 2000; Sadeh et al., 2003; Brueckner et al., 2003). Theagent based approach allows to taking the following factorsinto account easily:• The distributed nature of SC and the non linearity of

their behaviours;• The modifications of the environments, since the agent

autonomy allows the model to change by updatingthrough addition and removal of agents;

• The decision complexity and variety, through the MAScapability to solve problems by cooperation in adistributed way;

• The presence of cooperative, yet not altruist, behaviourby actors, thanks to the definition of the agents'properties through interaction modes.

2.3 Using multi-agent systems for modelling supplychains

MAS offer a behavioural representation in a modular wayand allow simulation models to have evolutionarytransformation capabilities due to their ability to adaptthemselves to changes and to structural modificationswithout an entire alteration of the system. Agent-basedmodelling is especially well suited to SC problems since itallows the actors' behaviours and the interactions to berepresented (Moulin and Chaib-Draa, 1996). Previousresearch in SC domain based on the agent paradigm hasbeen mainly focused on the management aspects. Somenoteworthy examples of this focus are found in the works of(Strader et al., 1999; Swaminathan et al., 1998; Gjerdrum etal., 2001), the MASCOT (Sadeh et al., 2003) and theDASCh (Parunak et al., 1998) project.

In these projects, the scope of the study, the simulationobjectives and the abstraction level influence the multi-agent organization design which represents the SC. Asrevealed in (Gjerdrum et al., 2001) and (Swaminathan et al.,1998), the representation has not been systematicallyisomorphic. For example, each plant site or manufacturingresource is not necessarily represented by an agent. TheDASCh project (Parunak et al., 1998) and the MASCOTproject (Sadeh et al., 2003) and several other works rely ondifferent combination of agents by company modelled. Theabstraction level influences the number of agents and thegranularity of their behaviours.

Depending on the project, an agent in the model hasbehaviours, communicational and decisional capabilities,which range from the management of a resource to theplanning of a company (Gjerdrum et al., 2001). Forexample, the agents attempt to locally optimize theirschedule or estimate various scheduling policies by usingnegotiation processes or protocols. The DASCh project(Parunak et al., 1998) defines a Company agent whichinteracts with the Production Planning and InventoryControl agent when a problem occurs in order to compute anew plan.

2.4 Current limits of Agent-Based Supply ChainModelling

At the modelling level, the analysis of previous researchworks highlights the fact that the agents are used to modelthe existing actors and / or decisional processes. However,these works differ relative to the approaches applied for thedesign of the simulation model. They do not offer modellingframeworks based on domain modelling which sufficientlyprovide for the variety of SC configurations, and moregenerally for the physical characteristics of the real systems.

At the model design level, the works of (Swaminathan etal., 1998) are centred on the modularity and the re-use ofconceptual models using a library of SC components. Thislibrary is made of components: structural which define theactors of the SC, and functional which describe thebehaviour of each actor by specifying the decision processesand the actions to realize. The use of such a library can be


restrictive if the elements needed for designing a new modeldo not match the library components (for example,customers do not belong to the structural components).Moreover, the number of available actor roles appears to belimited with respect to possible multiple relationships whichinfluence the SC structure and the actors' behaviour.

Finally, in the analyzed projects, demand estimates arespecified as fixed input parameters of the simulation model.A representation of the customer behaviour is needed totake the dynamics of this demand into account in customercentric SC. This is especially true in a productpersonalization context where demand is difficult to predictby classical forecasting methods due to the lack of data or ofhistory or due to the lack of adequate granularity.

2.5 Conclusion

The analysis of the coordination and the synchronization ofthe SC actors is a key step towards improving SCperformance. It is difficult to realize such an analyticalstudy directly on the real system, paving the way for using asimulation model.

The agent-oriented approach is both relevant andapplicable for SC modelling and simulation. Each analyzedresearch project had its own aims in modelling andsimulation and consequently supplied specific decisionsupport to the improvement of the SC performance.

Our research is concerned with a specific kind of SC:customer-centric SC. None of the research works studied inthis section has covered this kind of SC. Such SC havespecific characteristics due to the role that the customerplays in the chain. The agent approach appears to berelevant to model and simulate these SC.

3 MODELLING CUSTOMER-CENTRIC SUPPLYCHAINS

In customer-centric SC, the customers have to be consideredas the "prime contractors" actors. These SC may have toface an erratic and highly dynamic demand induced by thevolatility of the customers needs. Consequently, a relevantmodel has to explicitly take into account the customer, itsdemands, its interactions with the other actors of the chain.It also has to consider various multiple supplier/clientrelationships between these actors. If the agent-orientedapproach is relevant for SC representation, it appears to benecessary for the customer-centric SC description.

This section starts by defining the specificities of suchchain at the enterprise modelling domain level in order topropose an agent representation of customer-centric SC.Then we propose a specific domain modelling approachbased on two different models: a "Structural Model" and a"Dynamic Model". This approach allows, in particular, toexplicitly take into account the customer and its behaviours.A case study from the golf club industry is used to illustratethe domain modelling approach.

3.1 Mass customization

Mass customization is a key answer to take into account thedynamics of customers’ requirements. Mass customizationconsists of making products in large volume and adapting toindividual customer needs, in accordance with massproduction principles.

To cope with the volatility of customers needs,characterized by a dynamic evolution of the products, SChave to adapt their structure and their coordinationprocesses. The organizational structure of the companies hasto be adapted in order to deliver personalized products inshorter delays than those necessary for their manufacturing.In this context, the period between the order receipt and thedelivery date corresponds to time necessary for the productcustomization and delivery.

The processes and products need to adapt to meetobjectives discussed above. The use of standardizedmethods and modular products contributes to the reductionof delivery periods and to the acceleration ofpersonalization. SC directed by the production anddistribution of personalized products adapt theirorganizational structure as well as the relationships amongcompanies, altering material, information and monetaryflows. To give an efficient response in shorter delays, thepersonalized product transformation activities becomestrictly make-to-order.

3.2 Domain Modelling

The level of abstraction used in domain modelling has tocorrespond with the objective of formalizing the domainknowledge so as to describe the structure and the behaviourof the considered SC. At this level we use models andspecific formalisms in the domain of industrial networks.The domain modelling is based on approaches proposed in(Montreuil and Lefrançois, 1996; Montreuil et al., 2000).The representation of the SC structure is based on theconcept of responsibility network that we have extended tothe customer actor.

We proposed in (Labarthe et al., 2003) a classification ofthe activities of actors in a SC according to eight levels ofpersonalization illustrated in Figure 1. Each actor ischaracterized by its role and its responsibilities. The set ofthese characteristics defines an activity network. Each actorof the network has to coordinate its activities with the otheractors to meet the demand. This demand enters the chainthrough orders. The entry point of an order in the SC istermed the order penetration point (Hoover et al., 2001).This point corresponds to a node in the activity network ofthe SC and depends on the level of personalization requiredby the client. Locating the order penetration point consistsof determining the point of order entry on the physical flowof products.


Figure 1 Reference model for the actors’ coordination of SC

Identifying the personalization level allows positioning thede-coupling point (Christopher and Towill, 2000) on theactors' network. This identification defines the managementpolicies and the relations among actors upstream anddownstream of this point. The nature of orders defines thetypology of the SC by products’ personalization level. Thisimplies specific information and material flows. Thecustomer relationship and the location of the orderpenetration point define the structure and the behaviour ofthe SC actors. The structure and behaviour are representedusing two complementary models, the Structural Model andthe Dynamic Model described below.

3.2.1 The Structural ModelThe structural model describes the elements and the linkswhich form the SC. The representation of the SC is based onthe concept of "responsibility network". The SC iscomposed of a set of autonomous actors who have roles andadditional responsibilities in line with their competenciesand the activities which they are able to perform for theproduct transformation. A responsibility network isrepresented in each decision-making level according to an“inter” and “intra” organizational decomposition (inter-enterprise, inter-department ...). This refinement offers arecursive representation of the organization (SC,Enterprises, Departments, Cells and Resources).

A responsibility network can be defined at the decision-making and/or operational levels. It formalizes the roles ofeach element of the SC. Roles are defined according to theelements in interaction (supplier/client links) and to theactivities assigned to each actor (assumed as responsibility).The responsibility network according to the Inter-Enterpriselevel is decomposed into a new responsibility network at theEnterprise level. This decomposition implies redefinition of

objectives by level, highlighting the coordination of actorsaccording to their activities. It specifies a network for eachdecision-making level.

3.2.2 The Dynamic ModelThe dynamic model defines the behaviour of each elementof the structural model and identifies the interaction modesused within the SC. The design step of the Dynamic Modelis based on the NETMAN (NETworked MANufacturing)methodological framework (Montreuil et al., 2000),developed in our laboratories. This framework concerns theconception and the implementation of the organizationalstructure of hierarchical or heterarchical industrial networks.It allows economic planning, control and management ofthe activities of the actors of the network in a dynamicenvironment. Each independent business unit of the actor'snetwork is represented as a center. These centers collaborateand coordinate with the other centers by interactions offlows (information and physical). The external actors in theconsidered organization are customers and suppliers. Theinformation flows used to associate with the dynamics ofthe system are:

• The needs are represented as orders and forecastscharacterized by the personalization levels.

• The offers are expressed by personalization levels anddelays per market zone (Poulin et al., 2006).

• The coordination data are expressed from theperformance analysis defined from a local and globalpoint of view (actor and actors' network) (Labarthe etal. , 2003). The performance analysis providesinformation concerning actions carried out by actors.

• The coordination models are developed according totwo approaches: the network models and the centermodels. The network models represent the activity

S P A G F E D S C

Supplier Manufacturer Manufacturer Distribution Center

Retailer Demand

ActorsNetwork

Supplying ProducingAct ivity Network

Order Penetration PointPersonalizationLevels

Assembling Grouping Finishing Packaging Distributing Sales Buying

Leve

l VII

Leve

l VI

Lev e

l V

Leve

l IV

Leve

l III

Leve

l II

Leve

l I

L eve

l 0

DP

DP

DP

DP

DP

DP

DP

DP Design To Order

Buy To Order

Make To Order

Assemble To Order

Group To Order

Finish To Order

Package To Order

Deliver To Order

Customer s Relations

Push Flow

Pull Flow

Transport Physical Flow Order DP Decoupling Point S Supplier C Customer


coordination within the SC. The center models describethe coordination of the actors within the center. Thesemodels are shared according to two modes oftransmission, upstream and downstream. We specifythe contents of information exchanged between actorsto obtain efficient coordination.

3.3 The industrial case

The illustrative SC is inspired by major manufacturers fromthe golf club industry (Poulin et al., 2006). Depicted inFigure 2, this chain is characterized by specificcombinations of product personalization levels beingoffered to specific markets. The demand is the marketresponse to these offers, translated into customers in eachmarket each deciding or not to purchase a specific set ofgolf clubs at a specific time. Actors who compose the SCare considered autonomous and collectively responsible tosupply final products on all market zones. Upstream in thechain, a set of suppliers is responsible to deliver componentsand raw materials to the Golf Club Manufacturer.

3.3.1 The Structural ModelThe structural model is built around a set of actor typescomprising producers, assemblers, processors, fulfillers,distributors, retailers and customers. Producers have theresponsibility to deliver raw materials and/or components tothe Assembler. Processors are responsible for performingspecified operations on parts. The Assembler and theFulfiller both have responsibilities to transform rawmaterials and/or components into final products, thedifference lying mostly in the fact that fulfillers makepersonalized products, strictly to order, while assemblerscan assemble standard products and modules, and producethem either to stock or to order.

Final products are delivered to the Distributors andRetailers. Distributors deliver final products to Retailers.Retailers sell final products to the Customers. Retailers forma physical and organizational interface between Customersand the manufacturing network. At the level of Retailers,final products are stored and then delivered to customers.The number of Retailers is specified by market zone. Eachcustomer is located in a specific market zone. Figure 3represents the structural model of the industrial case at theInter-Enterprise level through a responsibility network.

3.3.2 The Dynamic ModelThe dynamic model depicted in Figure 4 at the inter-enterprise level is based on a part of the responsibilitynetwork described in the structural model (Shaft Producer,Golf Club Assembler, Golf Club Fulfiller, Distributors,Retailers and Customers). Customers express their needs bysending orders to the Retailers. These needs rise up alongthe responsibility network to the appropriate de-couplingpoint. The Customer relationship described in Figure 4 is anAssemble-To-Order relationship. The de-coupling point issituated at the Golf Club Assembler.

Transformation activities are then made in a pulled flowfrom the de-coupling point to the Customer. Figure 4illustrates how network models and center models areshared between SC’s actors to detail their coordination. Thisrepresentation shows SC actor’s behaviours in term ofCustomers needs translating the dynamics of the demand.Customers can purchase popular products stored at theRetailer or place an order for an out-of-stock popularproduct or for a personalized product fitted to hisspecifications. The retailer can transfer customer orders toothers Retailers, to Distributors, or to the Golf ClubAssembler depending on the established business rules.

Figure 2 An industrial case from the golf club industry

x 50 000x 300

x 4 Markets Zones

x 1 European Market

x 50 000x 500

x 1 US Market

Molds Sub-Contractor

Molded Heads Supplier

Shafts Supplier

Grips Supplier

Forged Heads Supplier

Finished Products Distributor

Customized Golf Club Manufacturer x 4 Markets Zones

DEMAND

Golf Club Manufacturer

Golf Club Sub -Contractor

Others MarketsSupplier Manufacturer Distributor Retailer CustomerPhysical

Flow

Informational Flow


Figure 3 A Structural Model at the Inter-Enterprise level

Figure 4 A Dynamic Model at the Inter-Enterprise level

4 AGENT-BASED MODELLING AND SIMULATION OFA CUSTOMER-CENTRIC SUPPLY CHAIN

This section presents an agent-based modelling approach forthe simulation of customer-centric SC. The proposedapproach feeds from the domain model and permits an agentbased simulation of the SC on a specific agent simulationplatform. First we define the agent based modelling processaccording to two different abstraction levels: the conceptualand operational levels. At each level a specific agent modelis defined. Then for each of these models, we define the

modelling objectives and process, and we illustrate itthrough the case study introduced in the previous section.Finally, we present the adopted multi-agent platform for ourprototype implementation and give some details on thesimulation of the customer actor of the SC.

4.1 Agent-Based Modelling process

We propose to model and simulate customer-centric SCbased on the agent paradigm. The proposed agent-basedmodelling process represents the SC as a network of agentswhich adapt their behaviours to the evolution of theenvironment. Environmental dynamics are mainly

x 50 000x 300

x 4 zones

x 1European Market

DEMAND

Others Markets

Customer

ActorDemandPhysical Flow

x n Entity NumberCustomer

Processor ProducerAssemblerFullfiler

DistributorRetailer

VRetailer

DDistributor

x 50 000x 500

x 4 zones

x 1 US Market

CustomerVRetailer

DDistributor

DDistributor

F EFullfiler

Business Process (Assembling, Stocking, Finishing, Producing, Transport, Grouping, Sales , Packaging , Distributing)

AA GP

Assembler

S

S

S

S

T

T

S

S

PProcessor

S S T

TT

T

S S S S

S

PS S TProducer

PS S TProducer

PS S TProducer

PS S TProducer

PProcessor

S S T

EV

Informationnal Flow

MoldsProcessor

Molded HeadsProducer

ShaftsProducer

Grips Producer

Forged HeadsProducer

Fullfiler Golf Club

Golf Club Processor

Final ProductsDistributorGolf Club

Assembler

DPDP

NetMan Center

NetMan Center Models

Material Flows

Needs Expression

Information Sharing

Raw Material

Work In Process

Final Product

De-coupling Point

Handling and transport

Business Process (Assembling , ...)

A

Models Transmission

Demand Models

Manufacturers Network NetMan

CustomerProducer

(Producing)

P

RetailerV

Customer

DistributorD

Fullfiler (Finishing, Packaging)

F EAssembler (Producing, AssemblingAssembling, Grouping)

A

DPDP

GP

VV

DFFModels Periodical

Update

Centers Models

Network Model

DDEF EFCenters Models

Network Models

Distributor ModelFullfiler ModelAssembler Model

A G

DPDP

A G

DPDP

VRetailer

VRetailer

Customer


characterized by the specification of the customer'sbehaviours. In section three, at the domain level, we havedescribed the structure and the dynamic of the consideredSC. The structure is represented by various decision-makinglevels in a unified framework describing behaviours andinteractions of the actors according to these levels.

The agent-based process proposed for customer-centricSC representation is based on a specific domain modelpresented in section 3 (input of the process) and iscomposed of two agent models associated to two abstractionlevels:

• The conceptual level translates the domain model intoan agent-based model disregarding technicalconsiderations associated with the execution of anysimulation. The conceptual model obtained is definedby several sub-models that together specify thearchitecture of the MAS. This approach also specifiesfor each agent its behaviours as well as its interactions.

• The operational level renders the previously obtainedsimulation of the conceptual agent model operational.At the conceptual level, the set of agents is defined anda general decision-making behaviour is described foreach agent. At the operational level, specific actions aredetermined corresponding to the prescribed behaviourof each agent. At this level, the internal softwarearchitecture of each agent is also defined as acomposition of interacting elementary software agents.At this level we consider specificities and constraintsrelated to the implementation environment.

Figure 5 illustrates the proposed modelling approachhighlighting various abstract levels. In the following sub-sections we develop each of the agent-based modellinglevels by clarifying their objectives.

Figure 5 Proposed multi-agent modelling process

4.2 Conceptual Agent Modelling

Conceptual Agent Modelling transposes the domain modelinto the agent paradigm. In order to introduce theConceptual Agent Modelling we present the objectives ofthe process and an illustration.

4.2.1 Modelling objectives and processThe actors and their behaviours, captured in the Structuraland Dynamic Models, are translated as a set of agents. Eachagent is defined according to its beliefs, behaviours, andinteractions links, hence specifying the overall architectureof the MAS.

We represent actors of the real world associated to the SCby agents. Thus Conceptual Agent Model identifies agentsassociated to the actors of the SC, including the customers.Actions or operating behaviours of these agents arespecified for simulation through the Operational AgentModel.

4.2.2 Illustration of Conceptual Agent ModellingTo illustrate the Conceptual Agent Modelling, we focushere on the Shaft Producer. The interaction with Customersis represented by elementary activities such as ordersending. The specification phase allows us to formalize theDomain Model using the agent paradigm. As one possibleillustrative modelling alternative, we propose to model theShaft Producer through a team of five SC agents. As shownin Figure 6, four agents are respectively responsible for eachof the main activities actually touching the products: rawshaft supply, shaft transformation, shaft supply and shaftdistribution. The fifth SC agent is the Shaft Producer agent,responsible for managing demand and planning production.Each SC agent is characterized by its role, itsresponsibilities and by the activities it is able to realize.As an example, the Raw Shaft Supply Agent has aDistributor role. It is responsible for purchasing, receiving,delivering and controlling the inventory of raw shaft. Itsactivities are to: (i) receive and stock raw shaft section, (ii)manage raw shaft purchases (reorder method), (iii) controlraw shaft inventory, and, (iv) distribute raw shaft to theTransformation Agent. We identify the interaction links(physical and information) existing between agents. Thephysical links allow objects (raw shaft and shaft) tocirculate along the physical flows.

4.3 Operational Agent Modelling

The objective of Operational Agent Modelling is to createan executable model from the Conceptual Agent Model(CAM). Designing the Operational Agent Model involvesthe integration of individual agent characteristics defined atthe conceptual level: social properties, organizationalstructure, interaction protocols and environmentalconstraints. Each characteristic implies a top-downdescription. The actions of each agent associated with thedecision-making behaviour are specified. Specificities andconstraints related to the adopted implementationenvironment are also taken into account at this level.

Conceptual Agent Modelling

Operational Agent Modelling

Conceptual Level

Operational Level

Domain Level

Agent-Based Modelling

Execution Level

Implementation within a multi-agent Platform

Domain Model of the Supply Chain

Multi-Agent Simulation


Figure 6 Conceptual Agent Model of the Shaft Producer

4.3.1 Modelling objectives and processThe Operational Agent Model (OAM) is derived from theCAM. Each “conceptual” agent of the CAM associated to aSC actor generates an “operational” agent of the OAM. This“operational” agent is defined as a composition ofinteracting cognitive or reactive agents according to the"Agent Actor" model defined in (Labarthe et al., 2003). Anoperational agent is composed of one cognitive agent and/ornone or several reactive agents. Cognitive and reactiveagents are defined below:

• A cognitive agent deals with the decisional activity of aconceptual agent. Such cognitive agent is a deliberativeagent, it can play one or several role, and its decision-making behaviour are specified by Behavioural Plans(BP). BP allow the agent to act so as to reach its aims.Decisions can require communication for theinformation request or for carrying out actions.

• A reactive agent is responsible for an operative activityassociated to with the represented conceptual agent.The behaviour of a reactive agent is defined as statecharts, and is linked to the decision-making behaviourof the responsible cognitive agent. The states specifyactions carried out after the reception of messages fromthe cognitive agent or of signals resulting from theenvironment.

Two different “operational” agents of an OAM can interactat two levels: between their cognitive agents by exchangesof messages in the context of BP associated to specificroles, and directly between their reactive agents byexchanges of signals. The first interaction mode concernsthe information flows and the second interaction modeconcerns the physical flows and their coordination (physicallinks).

In the context of SC modelling, for a given operationalagent, the decisional activity of its cognitive agent isassociated to the decision making of the SC actor, and theoperative activities of its reactive agents are associated tothe physical activities of this actor.

Thus the OAM is composed of a set of agents, whichrepresent the actors of the SC. To these agents, we allocatebehaviour similar to those of the actors that they aremodelling. Our objectives are to study the interlacedbehaviours of a physical system in which complex logisticaland production processes are running, and a decision-making system in which complex decisions processes arerunning. On one hand we have to define behaviours andreactions of the physical system based on decisions of itsdecision-making system. On the other hand we have todefine how these decisions also depend on the state of thephysical system.

Figure 7 From SC Actor to SC Agent

Figure 7 presents the mapping of SC actors to SC agentsbased on the process modelling, (Labarthe et al., 2003). Theencapsulation principle defines the following relation: thereactive agents perform operational activities from decisions

Supply Chain Actor

Agent Actor

Decision Making

Operative ActivityOperative

Activity

Deliberative Activity

Operative ActivityOperative Activity

Cognitive Agent

Reactive Agent

Encapsulation

Operative Behavior Representation+ Physical Links

Deliberative Behavior Representation+ Communication

Domain Level Conceptual Level Operational Level

Producer (Producing )

P

Shaft Producer

Specification

Raw Shaft Supply Agent (Activities : Raw ShaftPurchasing , Stock Managing , Raw Shaft Stocking and RawShaft Destocking )

Shaft Transformation Agent(Activities : Scheduling and Shaft Producing )

Shaft Producer Agent(Activities : Production Planning,

and Demand Managing )

Shaft Distributor Agent (Activities : Transport Planning,and Shaft Transporting )

Raw Shaft Object

Shaft Object

Shaft Supply Agent(Activities : Stock Managing,Shaft Stocking and Shaft Destocking )

Shaft Producer Environment


taken by the cognitive agent. The OAM is composed of twoagents' societies, one of cognitive agents and the second ofreactive agents. This decomposition allows differentiatingbehaviours, roles and competences adopted by the actorswithin the organization of the system. The aim of theimplementation phase is to carry out the OAM on adedicated Multi-Agent platform.

4.3.2 Illustration of the Operational Agent ModellingThe OAM is composed of two agents' societies, onecognitive and one reactive. This decomposition allowsdifferentiating the behaviour, the roles and the competencesadopted by agents. The OAM is formed by all the agentscomposing the CAM. Figure 8 shows the OAM of the ShaftProducer actor previously described.

The previously introduced Raw Shaft Supply Agent iscomposed of two agents, a cognitive agent and a reactiveagent. The cognitive part of this SC Agent is represented bythe Raw Shaft Supply Agent. This agent is responsible forthe realization of the decision-making activities, involvingcomplex behaviours. The reactive part of this SC Agent isrepresented by the Raw Shaft Dispatch Agent. This agent isresponsible for the realization of the operational activities,involving physical activities.

Agents are represented with a Belief Desire Intentionarchitecture (Kinny et al., 1992) based on plans. These plansare specified by state charts according to the AgentBehavior Representation (ABR) formalism (Tranvouez,2001). Figure 9 illustrates the behavioural plan for the "RawShaft Purchasing" Agent.

Figure 8 Operational Agent Model of the Producer Shaft Agent

Figure 9 Behavioural Plan "Raw Shaft Purchasing" for the Raw Shaft Supply Agent

Cognitive Society

Reactive Society

Customer Agent (Customer)(Behaviors : Initiate orders,

counter-proposal and cancel orders)

Shaft Producer Agent (Producer)(Behaviors : Manage the Demand and Plan production)

Shaft Distributor Agent (Distributor)(Behavior : Plan transport)

Shaft Supply Agent (Distributor)(Behavior : Manage stock)

Shaft Transformation Agent (Producer)(Behavior : Schedule)

Raw Shaft Supply Agent (Distributor)(Behaviors : Manage purchasing and stock)

Raw Shaft Dispatch Agent (Distributor)(Behaviors : Receive Raw Shaft and Send Raw Shaft)

Raw Shaft Cut Agent (Producer)(Behavior : Produce intermediary Shaft)

Shaft Dispatch Agent (Distributor)

(Behaviors : Receive Shaft and Send Shaft)

Shaft Distributor Agent (Distributor)

(Behavior : Deliver Shaft)

Shaft Polish Agent (Producer)

(Behavior : Finish Shaft)

Message

Signal

Agent Actor

Reactive Agent

Cognitive Agent

BP starts

Call for proposal

Analyze counter proposal Activate orders

management Activate Knowledge Update BP

Plan Activation signal

Counter -proposalReject proposal

Accept proposal

success

Message sent

Message sent

Wait for answers

Message from logistic agent

Analyze response

Refuse counter proposal

Solution Rejected Solution Accepted Orders failed

BP ends

BP starts

Call for proposal

Analyze counter proposal Activate orders

management Activate Knowledge Update BP

Plan Activation signal

Counter -proposalReject proposal

Accept proposal

success

Message sent

Message sent

Wait for answers

Message from logistic agent

Analyze response

Refuse counter proposal

Solution Rejected Solution Accepted Orders failed

BP endsInitial State

Internal Transition

Composite Action State

Final State

Elementary Action State

Communication State

Unlimited wait state

Limited wait state

External Transition

State


Figure 10 Behavioural Representation of the Raw Shaft Cut Agent

The Raw Shaft Supply Agent has a second Behavioural Planallowing it to establish and to manage communication linkswith an agent responsible for the raw shaft transport(Logistic Agent). These communication links areparticularly achieved by sending order propositions, ordersconfirmation or refusal according to the receivedcounterproposals.

Reactive agent architecture is based on reflex states. Theagents' behaviours are implemented with state chartdiagrams defined using Unified Modeling Language (UML)/ Agent Unified Modeling Language (AUML) (Odell et al.,2001). Figure 10 illustrates the behaviour of the Raw ShaftCut Agent.

Received events trigger the activation of the state chart andthe Sent events result from actions of the agent. The ShaftTransformation Agent responsible for the transformation ofraw shaft into intermediary shaft, sends orders to the RawShaft Cut Agent. In answer to this event and according tostates and transitions activated in the state chart, the RawShaft Cut Agent will send out one of three specified events.

4.4 The Multi-Agent Simulation Platform

In this section we detail the characteristics of a simulationplatform. The simulation makes possible the comprehensionand prediction of phenomena based on the experimentationsthrough the execution of different scenarios. The simulationtool has to support the understanding by the user of thesimulated SC behaviours. A Multi-Agent SimulationPlatform is typically composed of the following modules: (i)the model parameters expressing the needs of the study, (ii)the user interface, (iii) the service agents for simulationcomposed of generic elements such as the Simulation Clock

and the Events Administrator, and (iv) the OperationalAgent Model.

In our current implementation, the service agents include:(i) the Agent Name Server, (ii) the Register Agent, (iii) theClock Agent for time synchronization, (iv) the EventsScheduler Agent for events synchronization, and, (v) theSimulation Controller Agent responsible for themanagement of simulation parameters. Figure 11 presentsthe general architecture of our Multi-Agent SimulationPlatform prototype, depicting the modules and the inter-module links. The main modules are the Database,MAJORCA for the cognitive agent society, Anylogic for thereactive agent society, and the Service Agents.

The Database provides the interface between the user andthe platform for the definition of simulation parameters, therecording of simulation data and the display of results.Cognitive agents are based on MAJORCA agents' models,Jess (Friedman-Hill, 1998) Engine Agents Oriented AgentBehavior Representation, (Tranvouez, 2001). MAJORCA isa platform which proposes a Multi-Agent developmentenvironment based on Behavioral Plans specified with theABR formalism. Reactive agents are implemented withinthe Anylogic environment (simulation of discrete,continuous and hybrid systems). Figure 12 presents adevelopment interface showing the conception of thereactive agent society. Every agent is built according to itsrole (one or several behaviours implemented with statecharts).

This model is obtained by positioning the agents composingthe reactive agent society (defined by their roles) from theaccessible roles in the responsibility network. Oncepositioned in the network, the agents are interconnectedaccording to supplier/client links.

Idle

Receive order (product, delay)

Preparation

Transformation

Report Emission

Resource prepared (product)

Product transformed

Resource unloaded (product)

Report send (order)

Load

Unload

Breakdown

Maintenance

Resource loaded (product)

CancelFailure resource prepared (product)

Failure resource loaded (product)

Failure product transformed

Failure product unloaded

Order not completed

Order Canceled

Identify breakdown

Maintenance done

Idle

Receive order (product, delay)

Preparation

Transformation

Report Emission

Resource prepared (product)

Product transformed

Resource unloaded (product)

Report send (order)

Load

Unload

Breakdown

Maintenance

Resource loaded (product)

CancelFailure resource prepared (product)

Failure resource loaded (product)

Failure product transformed

Failure product unloaded

Order not completed

Order Canceled

Identify breakdown

Maintenance done

Shaft Cut Resource

Events Received

Events Sent

Order (operator)

Resource Loaded

Resource prepared

Resource Unloaded

Breakdown (environment)


Figure 11 The Multi-Agent Simulation Platform

Figure 12 Reactive Society Implementation

4.5 Simulation of the Demand

Demand for a given personalization level is considered oneof the essential elements forming the dynamic environmentof the SC. The personalization level is described accordingto the relation between Customers and the SC. Customerscan either acquire products from Retailers who generateorders in case of unavailability or the need for apersonalized product, or place orders via a Business-To-Consumer e-commerce interface. Personalization offersinclude the archetypical options presented in (Poulin et al.,2006): popularizing, varietizing, accessorizing,parametering and tailoring.

In this effort we put particular emphasis on the simulationof the Customers, taking into account personalization levelsand market zones. Table 2 presents the characteristicsassociated to products per personalization level for aspecific market. For every personalization level, anappropriate mix of products is proposed. Table 2 computesthe number of product combinations offered in the Canadianmarket for each personalization option, through the

combinatorial multiplication of their intrinsic productcomponents.

A Customer Agent represents a human customer in a marketzone. This agent generates an order if an offer made to himby a retailer or an e-store satisfies his needs and meets hisexpectations. The offers made to him vary depending on themarket zone and the expression of his needs. The demand isthe number of demanded products, specified per timeperiod, per personalization level and per market, prior tonegotiation with the retailer regarding availability issues.Demand is thus clearly potentially different from sales.

For simulation purposes, it has to be estimated throughsome appropriate probability distribution. Figure 13illustrates such time-phased demand estimation for eachpersonalization level. It exhibits a strong seasonal pattern asis the norm in the golf club industry. These estimatesbecome the core based on which the demand instances aregenerated, translating in customers dynamically enteringretailers or logging in an e-commerce site. These demandcurves are recorded in the Database which allows the user toobtain an aggregated vision of the demand. However the SCAgents in the simulation do not have access to the generateddemand distributions.

AnylogicMajorca

Service Agents for Simulation

Clock Agent

Events Scheduler Agent

Register Agent

Agent Name Server

SimulationController Agent

User(s)

Parameters

Data

Results

Database

Multi -agent Simulation Platform

Customer

Responsibility

Shaft Section Manage Agent

Implementation of the reactive society

Behavior


Table 2 Offers versus personalization level, source (Poulin et al., 2006)

Figure 13 Estimated demand per market per personalization level

5 CONCLUSION

The new economy is characterized by stiff competition andhigh market and technological turbulence, with an emphasison ever delighting customers. For an enterprise to maintaina competitive position, it is fundamental for it to be demanddriven and customer centric, exploiting the dynamic andstochastic nature of the demand, stemming from fastevolving customer expectations. It is essential to developand exploit agility and intelligence in its SC to meet the goalof delighting its customers while maintaining profitability.

In this paper we have proposed an agent-based modellingapproach for simulation of customer-centric SC. Thisapproach allows a satisfactory representation of thestructure and behaviour of the SC, specifically taking intoaccount the operating policies induced by the expectedshape of the demand. Agents have the ability to adapt theirbehaviours to the environment modifications. Hence, theyare able to play different roles depending on the various

management policies. The choice of the adapted role isguided by the decoupling point position in the SC. It allowsthe agents to determine the appropriate policy to ensure thecustomer relationship according to the personalization level.

The approach is based on three modelling levels: DomainModelling, Conceptual Agent Modelling and OperationalAgent Modelling. Each actor of the SC is represented byone cognitive agent and/or none or several reactive agents.Interactions between cognitive agents correspond to actors’information exchange whereas interactions between reactiveagents concern physical flows. The agent-based operationalmodel of the SC implemented on the simulation platformsimulates the interactions of the two societies of reactiveand cognitive agents.

The developed prototype provides a facility forexperimentation and analysis of the impact of alternativemanufacturing network configurations, managementstrategies and policies. It generates results that focus on thebehaviour and performance of the SC as a whole and of

Canadian market Number of options per parameterParameters for iron sets offer Popularizing Varietizing Accessorizing Parametering TailoringClub head models 2 4 4 4 4

Metal alloy 1 1 1 2 4Availability both sides per model 2 2 2 2 2

Lofts per model 1 1 1 3 7Lie angles per model 1 1 5 9 13Sole grinds per model 1 2 2 3 6Shaft type/flex per model 3 12 20 40 60Length per shaft 1 1 5 15 30

Adjusted weight per shaft 1 1 5 5 19Grip types per shaft 1 3 10 20 35Grip size per grip 1 1 2 6 6

Number of iron set combinations 12 576 800 000 466 560 000 125 483 904 000Service Offer (average figures)Delivery delay (days) 1 3 5 10 30Delivery reliability (fulfill rate) 95% 95% 90% 90% 75%Penalty for late delivery (per day late) $20 $10 $10 $10 $25Price $850 $900 $1 000 $1 200 $2 000


each actor involved in it, especially in a customer-centriccontext.

REFERENCES

Brueckner, S., Baumgaertel, H., Parunak, H.V.D., Vanderbok, R.and Wilke, J. (2003) ‘Agent Models of Supply NetworkDynamics: Analysis, Design, and Operation’, in Harrison,T.P., Lee, H.L. and Neale, J.J. (Eds), The Practice of SupplyChain Management: Where Theory and ApplicationConverge, Springer-Verlag.

Christopher, M. and Towill, D.R. (2000) ‘Supply chain migrationfrom lean and functional to agile and customised’,International Journal of Supply Chain Management, Vol. 5,No. 4, pp. 206-213.

Courtney, H., Kirkland, J. and Viguerie, P. (1997) ‘Strategy UnderUnce r t a in ty ’ , H a r v a r d B u s i n e s s R e v i e w,November/December, pp. 67-79.

Fisher, M. (1997) ‘What is the right supply chain for yourproduct’, Harvard Business Review, March, pp. 105-116.

Fox, M.S., Barbuceanu, M. and Teigen, R. (2000) ‘Agent-OrientedSupply-Chain Management’, International Journal ofFlexible Manufacturing Systems, Vol. 12, No. 2-3, pp. 165-188.

Friedman-Hill, E.J. (1998) Jess, The Java Expert System Shell.herzberg.ca.sandia.gov/jess/

Ganeshan, R., Jack, E., Magazine, M.J. and Stephens, P. (1998) ‘Ataxonomic review of supply chain management research’, inTayur, S., Ganeshan, R. and Magazine, M. (Eds),Quantitative Models for Supply Chain Management,Kluwer.

Gjerdrum, J., Shah, N. and Papageorgiou, L.G. (2001) ‘Acombined optimization and agent-based approach to supplychain modelling and performance assessment’, ProductionPlanning & Control, Vol. 12, No. 1, pp. 81-88.

Hoover, W., Eloranta, E., Holmstrom, J. and Huttunen, K. (2001)Managing The Demand Supply Chain: Value Innovationsfor Customer Satisfaction, John Wiley & Sons.

Jennings, N., Sycara, K. and Wooldridge, M. (1998) ‘A roadmapof agent research and development’, Autonomous Agentsand Multi-Agent Systems, Vol. 1, No. 1, pp. 7- 38.

Kinny, D., Ljungberg, M., Rao, A. S., Sonenberg, E., Tidhar, G.,and Werner, E. (1992). “Planned team activity”, LNAI 830,pp. 226-256.

Kjenstad, D. (1998) Coordinated Supply Chain Scheduling, PhDThesis, Norwegian University of Science and Technology.

Labarthe, O., Tranvouez, E., Ferrarini, A., Espinasse, B. andMontreuil B. (2003) ‘A Heterogeneous Multi-AgentModelling for Distributed Simulation of Supply Chains’,Proceedings of the International Conference on Holonicand Multi-Agent Systems for Manufacturing, CzechRepublic.

Labarthe, O., Ferrarini, A., Montreuil, B. and Espinasse, B. (2003)‘Cadre de coordination distribué de chaîne logistique parmesure des performances’, Proceedings of CongrèsInternational en Génie Industriel, Canada.

Montreuil, B. and Lefrançois, P. (1996) ‘Organizing factories asresponsibility networks’, in Graves, R.J. et al. (Eds),Progress in Material Handling Research, Braum-Brunfield.

Montreuil, B., Frayret, J.-M. and D’Amours, S. (2000) ‘A StrategicFramework for Networked Manufacturing’, Computers inIndustry, Vol. 42, No. 2-3, pp. 299-317.

Moulin, B. and Chaib-Draa, B. (1996) ‘An Overview ofDistributed Artificial Intelligence’, in O'Hare, G. andJennings, N. (Eds), Foundations of Distributed ArtificialIntelligence, John Wiley & Sons.

Odell, J., Parunak, H.V.D. and Bauer, B. (2001) ‘Representingagent interaction protocols in UML’, in Ciancarini, P. andWooldridge M. (Eds), Agent-Oriented SoftwareEngineering, Springer-Verlag.

Parunak, H.V.D., Savit, R. and Riolo, R.L. (1998) ‘Agent-BasedModeling vs. Equation-Based Modeling: A case Study andUser’s Guide’, Proceedings of Multi-agent systems andAgent-based Simulation, LNAI 1534, Springer-Verlag.

Poulin, M., Montreuil, B. and Martel, A. (2006) ‘Implications ofpersonalization offers on demand and supply networkdesign: A case from the golf club industry’, EuropeanJournal of Operational Research, Vol. 169, No. 3, pp. 996-1009.

Sadeh, N.M., Hildum, D.W. and Kjenstad, D. (2003) ‘Agent-basede-Supply Chain Decision Support’, Journal ofOrganizational Computing and Electronic Commerce, Vol.13, No. 3-4, pp. 225-241.

Shen, W. and Norrie, D.H. (1999) ‘Agent-Based Systems forIntelligent Manufacturing: A State-of-the-Art Survey’International Journal of Knowledge and InformationSystems, Vol. 1, No. 2, pp. 129-156.

Strader, T.J., Lin, F.-R. and Shaw, M.J. (1999) ‘The impact ofinformation sharing on order fulfillment in divergentdifferentiation supply chains’, Journal of GlobalInformation Management, Vol. 7, No 1, pp. 16-25.

Swaminathan, J.M., Smith, S.F. and Sadeh, N.M. (1998)‘Modeling Supply Chain Dynamics: A MultiagentApproach’, Decision Sciences, Vol. 29, No. 3, pp. 607-632.

Tranvouez, E. (2001) IAD et Ordonnancement : une approchecoopérative du réordonnancement par systèmes multi-agents, Thesis, Université d’Aix-Marseille III.

Wooldridge, M. and Jennings, N. (1995) ‘Intelligent Agent:Theory and practice’, Knowledge Engineering Review, Vol.10, No. 2, pp. 115-142.

Wu, J., Ulieru, M., Cobzaru, M. and Norrie, D. (2000) ‘Agent-based Supply Chain Management Systems: State of the Artand Implementation Issues’, Proceedings of theInternational Congress on Intelligent Systems andApplications, Australia.

Yuan, Y., Liang, T.P. and Zhang, J.J. (2002) ‘Using AgentTechnology to Support Supply Chain Management:Potentials and Challenges’, Working Paper No. 453,McMaster University.

Zeigler, B.P., Praehofer, H.P. and Kim, T.G. (2000) Theory ofModeling and Simulation, Academic Press.

multi-agent modelling for simulation of customer-centric supply …acl 06... · 2018-02-18 ·...

Documents