啤酒游戏

Click to edit Master title style,Click to edit Master text styles,Second level,Third level,Fourth level,Fifth level,*,ArtificialAgents Play the Beer Game Eliminate theBullwhipEffect and Whip theMBAs,Steven O.Kimbrough,D.-J.Wu,FangZhong,FMEC,Philadelphia,June2000;file: beergameslides.ppt,The MIT Beer Game,Players,Retailer,Wholesaler, Distributor and Manufacturer.,Goal,Minimize system-wide(chain) long-run averagecost.,Information sharing:Mail.,Demand:Deterministic.,Costs,Holding cost: $1.00/case/week.,Penalty cost: $2.00/case/week.,Leadtime:2 weeks physical delay,Timing,1. New shipments delivered.,2. Ordersarrive.,3. Fill ordersplusbacklog.,4. Decidehow much to order.,5. Calculate inventory costs.,GameBoard,The Bullwhip Effect,Ordervariability isamplifiedupstreamin the supply chain.,Industry examples (P&G, HP).,Observed Bullwhip effectfromundergraduatesgameplaying,Bullwhip EffectExample (P & G),Lee et al., 1997,SloanManagement Review,Analytic Results: Deterministic Demand,Assumptions,:,Fixedleadtime.,Players work asa team.,Manufacturer has unlimited capacity.,“1-1”policy isoptimal- order whatever amountis orderedfromyourcustomer.,Analytic Results: Stochastic Demand,(Chen, 1999,ManagementScience,),Additionalassumptions:,Onlythe Retailer incurspenalty cost.,Demand distributionis commonknowledge.,Fixedinformation lead time.,Decreasingholding costsupstream in thechain.,Order-up-to (base stock installation) policyis optimal.,Agent-Based Approach,Agents work asa team.,No agent has knowledge ondemand distribution.,No informationsharing among agents.,Agents learn via geneticalgorithms.,Fixedor stochasticleadtime.,ResearchQuestions,Cantheagents trackthedemand?,Cantheagents eliminatetheBullwhip effect?,Cantheagents discovertheoptimalpoliciesiftheyexist?,Cantheagents discoverreasonably goodpoliciesunder complex scenarioswhere analytical solutionsarenotavailable?,Flowchart,Agents Coding Strategy,Bit-stringrepresentationwithfixedlength,n,.,Leftmostbitrepresentsthesignof,“,“,+,”or,“,“,-,”.,Therestbitsrepresenthowmuchtoorder.,Rule,“,“,x+1,”means,“,“ifdemandis,x,thenorder,x+1,”.,Rulesearchspaceis,2,n-1,1.,Experiment1a:FirstCup,Environment:,Deterministicdemandwithfixedleadtime.,FixthepolicyofWholesaler,DistributorandManufacturertobe,“,“1-1,”,”.,OnlytheRetaileragentlearns.,Result:RetailerAgentfinds,“,“1-1,”,”.,Experiment1b,AllfourAgentslearnundertheenvironmentofexperiment1a.,ber,rulefortheteam.,Allfouragentsfind,“,“1-1,”,”.,ResultofExperiment1b,Allfouragentscanfindtheoptimal,“,“1-1,”,”policy,ArtificialAgentsWhiptheMBAsandUndergraduatesinPlayingtheMITBeerGame,Stability(Experiment1b),Fix any three agentsto be “1-1”, and allow the fourthagentto learn.,Thefourthagentminimizesitsownlong-runaveragecostratherthantheteamcost.,Noagenthasanyincentivetodeviateoncetheothersareplaying,“,“1-1,”,”.,Therefore,“,“1-1,”,”isapparentlyNash.,Experiment2:SecondCup,Environment:,Demanduniformlydistributedbetween0,15.,Fixedleadtime.,AllfourAgentsmaketheirowndecisionsasinexperiment1b.,AgentseliminatetheBullwhipeffect.,Agentsfindbetterpoliciesthan“1-1,”,”.,ArtificialagentseliminatetheBullwhipeffect.,Artificialagentsdiscoverabetterpolicythan,“,“1-1,”,”whenfacingstochasticdemandwithpenaltycostsforallplayers.,Experiment3:ThirdCup,Environment:,Leadtimeuniformlydistributedbetween0,4.,Therestasinexperiment2.,Agentsfindbetterpoliciesthan,“,“1-1”.,NoBullwhipeffect.,Thepolicesdiscovered by agentsareNash.,Artificial agentsdiscoverbetterand stablepoliciesthan “1-1,”,” whenfacing stochasticdemand andstochastic lead-time.,Artificial Agentsareabletoeliminatethe Bullwhip effectwhen facingstochastic demandwith stochasticleadtime,.,Agentslearning,TheColumbiaBeerGame,Environment:,Informationlead time: (2,2,2,0).,Physicallead time: (2,2,2,3).,Initial conditionsset as Chen(1999).,Agentsfindtheoptimalpolicy: order whatever is orderedwithtime shift,i.e.,Q,1,= D(t-1), Q,i,= Q,i-1,(t, l,i-1,).,Ongoing Research:More Beer,Valueofinformationsharing.,Coordinationand cooperation.,Bargaining andnegotiation.,Alternativelearningmechanisms:Classifier systems.,Summary,Agentsare capableofplayingtheBeerGame,Trackdemand.,Eliminatethe Bullwhip effect.,Discovertheoptimalpoliciesifexist.,Discovergood policies under complexscenarios where analyticalsolutions notavailable.,Intelligentandagilesupply chain.,Multi-agententerprise modeling.,A frameworkformulti-agentintelligententerprisemodeling,