供应链预警平台推理模型的研究

上海交通大学硕士学位论文供应链预警平台推理模型的研究姓名：朱建勇申请学位级别：硕士专业：软件工程指导教师：王东200901011ABSTRACTIIResearchontheEarly-WarningintheSupply-ChainBasedonRatiocinativeModelABSTRACTWithalargenumberofinformationsystemsbeingdeployedinsupplychain,supplychainriskmanagementhasbeentransformingfrompastpassivemodetonowadaysactivemode.Thepassivemodeaimsatreducingthelostcausingbyrisk.Howeverthepassivemodeintendstoforecastandpreventrisk.Almostalltherelevantresearchesareinthemanagementarea.Discussesfocusonthedefinitionofthesupplychainrisk;theelementsproducingsupplychainrisks;theclassificationoftheseelementsorthecontentofthesupplychainriskmanagement.Whiletheseresearchescanonlyserviceforstrategylevel,theycan’tsupporttheoperationlevelactionsintherealsupplychainearly-warningapplication.Theyhardlymentionhowtoimplementintheinformationsystem.Atthebeginning,wescanthecontentofsupplychainandsupplychainriskinthemanagementfield.Thenweanalyzethealgorithmsbasedonthetraitsofthesupplychainapplication.Firstlyraiseanopenandmulti-algorithmsadaptivemethod.Throughourresearchtoachievesuchgoal:Inconditionthatthereisnochangeoftargetandplaninsupplychain,useminimalcostinoperationleveltoeffectivelypreventsupplychainriskandcontinuouslyimprovesupplychainmanagement,inordertostrengthencompetitivecapability.Inthisthesis,wechoosedecisiontree,ANN(ArtificialNeuralNetworks)andBayesmethodassamplestoprofoundlyanalyzetheclassifyalgorithms’adaptivecharacterinthesupplychainearly-warningapplication.Basedontheanalysis,weraisealayeredanddivided-conquermethod.Atthemacro-layer,wesettlethequestionhowtocalculatetheriskcorrespondingtothetopographyofsupplychain.Atthemicro-layer,weresolvetheproblemhowtoadoptdifferentratiocinationmodelstoABSTRACTIIIcalculatetherisksofvarioussupplychainnodes.Thecooperationoftwolayersisalsotakenintoaccount.Weprovesuchmethodisfeasiblebysimplerealization.Attheendofthesis,wewillgiveaconclusionandanexpectation.Keywords:supplychain,early-warning,ratiocinativeModel,topography11.1.1.1.1.20%10%421.1.2.31.2.80Stevens19899020902090AminAmid[1]~[7][8]~[12]41.3.1.1.3.51.1.4.1.4.672.1.2.1.1.1)[13]2000;20022)82.1.2.1)Christopher[14]WilliamsT.M[15]Mi11er[16]Jaafari//[17]WilliamsC.A[18][19])[20]2)Deloitte20049[21][22][23]2.1.3.1)[24]12)10[25]3)4)112.1.4.1)2)3)[26][27]4)5)[28][29]12[30]2.1.5.[31][32]2.2.2.2.1.2:GINI13ID3C4.5,CARTSLIQSPRINT1)InformationGain[33]=10L10L&=100L100L&=1000L1000Lentropy(2-1)(2-2)SmSASAvAAvASvSAA14(2-3)jA(2-4)AA2)GiniindexGiniCARTGiniD(2-5)DmGiniAvDAAAAAvADDGini(2-6)Gini15A(2-7)2.2.2.ArtificialNeuralNetworksANNNNConnectionistModelNaturalNeuralNetworkBPHopfield1)TLU1-1(2-8)(weight)16∑w1w0w2wn2-1Figure2-1TLUillustrationn1-1ANDOR2)delta(2-9)(2-10)to17deltadeltadeltadeltadelta(2-11)(2-12)DddEE(2-13)(2-14)id183)sigmoidSigmoid2-2∑w1w2wn2-2SigmoidFigure2-2Sigmoidunitillustration(2-15)(2-16)sigmoidsigmoid0119(2-17)2-3sigmoid2-3[34]Figure2-3threetiersnetillustration(2-18)outputsdk20a)ub)k(2-19)sigmoidc)h(2-20)sigmoidhkd)(2-21)(2-22)ij,ij[36]212.2.3.1)[36]1763205020DeFinettiFisherJeffreysJeffreysGoodSavage1950Wald2070RubinLindleyMinmax2)DHDH22(2-23)h(priorprobability)DhDh(posteriorprobability)DhD3)NaïveBayesianclassifierNaïveBayesianassumptionDnnnDmXXX23(2-24)(2-25)P(X)(2-26)D(2-27)X24(2-28)[37]4)[38](BayesianNetwork,BN)BNBNBN(DirectedAcyclicGraph,DAG)25(ConditionalProbabilityTables,CPT)CPTBNBN(2-29)X=XCPTBNBNBNCPTBN263.1.12ID3AAcc27AA28ANNsigmoid50ANNANNANNANNANN29ANNHPDGoogleNotificationPlatformE-maliFoxmail30EMk3-1()3-1Table3-1classifyalgorithmapplicability31ANN3.2.32n+vout/inBOM50%10.53.2.1.1)3-133A300100BCA0.750.25BC3-1Figure3-1weightindication3-2sonyA100100BC1003-2Figure3-2directionalgraphwithcycleinoutinout3-3inoutA100100BC200100100InOut3-3inoutFigure3-3directionalgraphwithabstractnodes“in”and“out”342)iRiRAA0R1nRA1()1(1)niiRAR==−−∏(3-1)PAAPA=1-RA0P1PtotalRtotalRtotal=1-Ptotal353.2.2.1)3-4AP=0.9BP=0.82003-4Figure3-4thesimplesampleA0.9B0.8PA=1-RAPR0.283-4inoutA1PAP=0.9A0.93-5A0.9B0.81InOut113-5inoutFigure3-5thesimplesamplewith“in”and“out”3-5in1out01AB013-2360001AR=0.1P=0.9R=0.1P=0.9BR=0.2R=0.2P=0.8P=0.8totalProbability0.020.180.080.723-2Table3-2thecalculationofsimplesamplein1out0.72P(total)inoutin1outP(total)P(total)=out/in3-6A0.9C0.80.4InOut11B0.90.60.6D0.70.413-6Figure3-6thecomplicatedsample3-2Out10.760.60.360.40.400AP=0.9P=0.9P=0.9P=0.9P=0.9P=0.9P=0.9P=0.9BP=0.9P=0.9P=0.9P=0.9R=0.1R=0.1R=0.1R=0.1CP=0.8P=0.8R=0.2R=0.2P=0.8P=0.8R=0.2R=0.2DP=0.7R=0.3P=0.7R=0.3P=0.7R=0.3P=0.7R=0.3totalProbability0.45360.19440.11340.04860.05040.02160.01260.005437Out00000000AR=0.1R=0.1R=0.1R=0.1R=0.1R=0.1R=0.1R=0.1BP=0.9P=0.9P=0.9P=0.9R=0.1R=0.1R=0.1R=0.1CP=0.8P=0.8R=0.2R=0.2P=0.8P=0.8R=0.2R=0.2DP=0.7R=0.3P=0.7R=0.3P=0.7R=0.3P=0.7R=0.3totalProbability0.05040.02160.01260.00540.00560.00240.00140.00063-3Table3-3thecalculationofcomplicatedsampleP(total)=0.715684P(total)nvoutnout2n2noutnvn+v2nn+vn2)3-6inP=1A3-7A0.90.410.63-7Figure3-7thematerialflowsofnodeA10.9AEA_out=0.9A0.60.40.90.6=0.540.90.4=0.3638E(aedge)=E(totalinput)P(stable)Proportion(aedge)(3-2)7E3-8:A0.90.4E=0.361E=10.6E=0.54