产品需求预测的演化神经网络算法MLPES

第14卷第3期运筹与管理Vol.14,No.32005年6月OPERATIONSRESEARCHANDMANAGEMENTSCIENCEJun.2005:20041210:国家自然科学基金资助项目(70028102):王瑛(1967-),女,博士,研究方向为供应链管理MLPES王瑛(,100084):,,,,MLPES,BP,,,:;;;;;:F272:A:10073221(2005)03000505ForecastofManufacturersProductDemandBasedonMLPESAlgorithmWANGYing(TsinghuaUniversityHongtaGroupPostdoctoralStation,Beijing100084,China)Abstract:Productdemandisthestartingpointandmotivationinthedemand-pulledsupplychains.Becauseofitsspecialpositioninthesupplychain,themanufacturerbecomesthedecouplingpointinthechangeofsupplychainfromdemanddriventoforecastdriven.Accuratemanufacturer-centricdemandforecastcantosomeextentreducetheaffectoftheuncertaintyofdemand.Furthermore,onthebasisofthemultilayerperceptronmodel,aneuralnetworkforecastingalgorithm(MultiLayerPerceptronbasedonEvolutionStrategy,MLPES)combininganevolutionstrategyisproposedwhichimprovesthecommonlyusedones,back-propagation(BP)algorithms.Theprocessofthelearningalgorithmandthemodelparametersobtainedbyseveraltimesofcalculatingarealsopresented.Finally,theforecastingresultisanalyzedandcomparedwiththatofaBPalgorithm.Keywords:enterprisemanagement;supplychainmanagement;demandforecasting;evolutionstrategy;decouplingpoint;multilayerperceptron(MLP)0(),,,,,,:-,:[1],,[2],;[3],,[4~6],,,,[7],,,!,,,,,(MultiLayerPerceptronbasedonEvolutionStrategy,MLPES),1,,,,,,,,;;,,,,,,,,,,[8],,,;,,(),,,,,,,[9],,[10]21,::/():GDP:():3MLPES(EvolutionStrategies,ES),6运筹与管理2005年第14卷1,,ES,,,,,,(MultipleLayerPerceptronMLP),BP(BackPropagationAlgorithm,BP)[11],,,MLP,x1(t),∀,xs(t),s,y^(t)sigmoidsquashing(S),f(x)=11+e-x(1)MLP,,(),ESES,3.1T,y(t)t,ee=#T-1t=2(y(t+1)-y^(t+1))2(2),,,,[12]MLP,W,WMLP,W(),eMLP,e,F,e,3.2W+1,WijMLPij,Wij∃W,Wij%,,e(gradientdecent),,Wij%=Wij+i(k)Yj(k)Nij(0,1)(3)7第3期王瑛:产品需求预测的演化神经网络算法MLPES1,Wij%=Wij+i(k)Yj(k)(4),Nij(0,1)N(0,1),Yj(k)kj,i(k)ki,i(k)=eUi(k)(5)iTi,L()ii(L)=(Ti-Yi(L))&f%(Ui(L))=(Ti-Yi(L))&Yi(L)(1-Yi(L))(6)1L-1i(k)=f%(Ui(L)#jj(k+1)Wji(k+1))=Yi(L)(1-Yi(L))#jj(k+1)Wji(k+1)k=1,2,∀,L-1(7)3.3,+1,∋1,(++1),+1,,3.422MLPES3.5MLPESMLP,[13],,,MLPES,,19,1,7,118运筹与管理2005年第14卷4,,19902001,12,11,,MLP,,(),MLPES,BP10000MLP0.08076,50000,0.04895MLPES,3300,0.0218512,10000BP219.84804,50000BP208.25345;MLPES194.30263,174.96748,25.6508%,19.0241%11.1056%BP,500003(a),MLPES,33003(b)35,,,MLPES,,,BP,,,MLPESBP,[1]NewboldP,BosT.Onexponentialsmoothingandtheassumptionofdeterministictrendpluswhitenoisedata-generatingmodels[J].InternationalJournalofForecasting,1989,5(4):523527.[2]GardnerES,McKenzieE.Forecastingtrendsintimeseries[J].ManagementScience,1985,31(10):12371246[3]CharlesW,ChaseJr.Theroleofthedemandplannerinsupplychainmanagement[J].TheJournalofBusinessForecastingMethods&Systems,1998,17(3):2325.[4]RaoR,ParikhJK.ForecastandanalysisofdemandforpetroleumproductsinIndia[J].EnergyPolicy,1996,24(6):583592.[5]ParigiG,SchiltzerG.PredictingconsumptionofItalianhouseholdsbymeansofsurveyindicators[J].InternationalJournalofForecasting,1997,13(2):197209.[6]TridimasG.TheanalysisofconsumerdemandinGreece.Modelselectionanddynamicspecification[J].EconomicModelling,2000,17(4):455471.(下转第59页)9第3期王瑛:产品需求预测的演化神经网络算法MLPES2-1-1251527120950-8-205-800-8-205-80-124-2152-1185225-2033522-3-422-10215-165201150492268110271101525011492268110271101525011-511-70110-10115-510111027-11027005-2097027-1185225-20335222271166135225001716135154271411311106110716117913900016349,X*=(0,0,35/27,50/27,14/27,0)T,Z*=4/9,[1].[M].:,1990.[2],,.[M].:,2000.[3].[M].:,1996.[4],.[M].:,1983.[5].[M].:,2001.[6].[M].:,1989.[7],,.[J].(),2004,3(1):1-3.[8].[J].,2004,13(1):7-10.[9].[J].,2004,13(1):11-16.[10],,.[J].,2004,13(1):63-67.(上接第9页)[7]JeongB,JungHS,ParkNK.Acomputerizedcausalforecastingsystemusinggeneticalgorithmsinsupplychainmanagement[J].TheJournalofSystemsandSoftware,2002,60(3):223237.[8]NaylorBenJ,NaimMohamedM,BerryDanny.Leagility:Integratingtheleanandagilemanufacturingparadigmsinthetotalsupplychain[J].InternationalJournalProductionEconomics,1999,62(12):107118.[9]SlatsPietA,BholaBis,EversJosephJM,DijkhuizenGert.Logisticschainmodeling[J].EuropeanJournalofOperationsResearch,1995,87(1):120.[10]vanderVlistPiet,HoppenbrouwersJurgenJEM,HeggeHermanMH.Extendingtheenterprisethroughmulti-levelsupplycontrol[J].InternationalJournalofProductionEconomics,1997,53(1):3542.[11]RumelhartDE,HintonGE,WilliamsRJ.Learninginternalrepresentationsbyerrorpropagation[A].ParallelDistributedProcessing,vol.1,DavidE.Rumelhart,JamesL.McClelland(Ed.)[C].Cambridge,Massachusetts:MITPress,1986:318362.[12]GreenwoodGW.Trainingmulti-layerperceptronstoRecognizeattractor[J].IEEETransactionsonEvolutionaryComputation,1997,1(4):244248.[13]BrentRichardP.FastTrainingAlgorithmsforMultilayerNeuralNets[J].IEEETransactionsonNeuralNetworks,1991,2(3):346354.59第3期唐建国:线性规划的符号跟踪算法