电子信息行业上市公司财务报表智能分析系统设计

太原理工大学硕士学位论文电子信息行业上市公司财务报表智能分析系统设计姓名：秦静申请学位级别：硕士专业：@指导教师：段富20090301I,12,IIIIITHEFINANCIALINTELLIGENTANALYSISSYSTEMDESIGNOFELECTRONICINFORMATIONINDUSTRYCORPORATIONABSTRACTIntheinformationage,takefulluseofthecorporate‘shistoricaldata,isakeyfactortowinthecompetitiveadvantage.DataMiningisanimportanttechnologytosupporttheBusinessIntelligenceinformationsystems.traditionalanalyticalmethodscannotdeepintominingdata’spotentialinformation.Financialanalysiswhichuseofdataminingtechnologycanimprovetheavailabilityofinformationgreatly.decision-makersinsupportofalargenumberofhistoricaldataunderthemulti-anglecomparison,analysis,cometothescientificresults.thetrainedminingmodelcanalsobechangedflexible,ithasbetterlearningandmemoryabilities,canmeetthedemandforintelligentinformationsystem.thestudyoffinancialanalysishasgreattheoreticalvalue,italsohaveimportantpracticalsignificanceforevaluationofbusinessindustry,smallandmediuminvestors’ecision-making.Thisarticlewillfocusonseveraldataminingtechniquesapplytotheelectronicinformationindustryfield,usedecisiontreeandneuralnetworktoanalysiscompanies’financialstatusinthisfield.thecontentsasfollows:1Firstly,setupthedecisiontreemodel,thenodesofdecisiontreecancreatedinteractive,soitcangeneratethemostinfluencedcorrelationrulesofelectronicinformationindustry.Itisalsoimportanttobeabletousetheserulestojudgeroughlythesituationofthecompany'sfinancialstability.Inthedecisiontreemodel,thehigh-frequencyvariablswillbeasaneuralnetworkinput,sothetheresultswillbeaccurateandtheerrorswillbesmaller.IV2Inthispaper,inordersolvethenetwork’sinitialweightandthresholdissue,weusethegeneticalgorithm’sglobalsearchabilitytodeterminetheinitialthresholdvalue,andalsoputforwardageneticalgorithmforneuralnetworkstructureandweights’evolution.itcanbeseenfromtheoutputresults,theneuralnetwork’spredictiveabilityisbetterthandecisiontree.KEYWORDSdataminingfinancialanalysisdecisiontree,neuralnetworkimmunearithmetic11.1CRMERPOA””AutoTrader.comSASReuteresReuteresSPSSSPSS/ClementineBassExport80223000BassExportIBMIneelligentMinerGartnerGroup“”1.2KDD1989KDD81999PAKDD158IEEEKnowledgeandDataEngineering1993KDDITSASEnterpriseMinerspssISLClementineYahooHyperParallAspenNeuralWorksOracleSASEnterpriseMinerIBMIntelligentMinerSGISetMinerSPSSClementineSybaseWarehouseStudioRuleQuest3ResearchSee5CoverStoryEXPLORAKnowledgeDiscoveryWorkbenchDBMinerQuestMicrosoftSQLServer2005SQLServerAnalysisServerReportingServices1993Web1.31.4:412BP3BP4SASEM52.1[33]2-12-1Fig2-1financialanalysisobjectsandthecontentanalyzed2-1620,2.27171=--=+8==292+/2+/2+/3100100=-10=100%=100%=100%==-100%=-100%=-100%2.3WEB11SASSEMMA2.3.1SEMMA1.Sample──2.Explore──SASSAS/INSIGHT3.Modify──SampleexploreSAS4.Model──SAS125.Assess──SAS2-1SampleExploreModifyModelAssess2-1SEMMAFig2-1SEMMAMiningprocess2.3.2[21]13CART[41]IBM,SAS,SPSS,HNC,Angoss,RightPoint,ThinkingMachinesNeoVistaHNCFalcon[21]Falcon14N2.4SASEM,PROCNERUAL2.5SASSEMMA15,163.13.1.1McCllochPitts[28]201943(BulletinofMathematicalBiophysics),3-13-1McClloch-PittsFig3-1McClloch-Pittsartificialnervecellstructure[35]:(1)(2)(3)(4)(5)(6)17(ANN,artificialneuralnetwork)1957FrankRosenblattperceptron2080,,3-2[34]3-2Fig3-2commonlynervecellmodelX1,X2,…,XnNnW1,W2,…,WnN∑WiXiOθ(θ)θf(θ)Sigmoid()iiYfWIθ=−∑(3.1)f(·)-W---(weightvector)W=[W1,W2,…,Wn]TI---(inputvector)X=[X1,X2,…,Xn]T18()TfWX(activationfunction).net=WTX()f(net)n-1X1,X2,…,Xn-1Xn=-1Wn=θθ(refractoryperiod),(0.5-2.0),BPARTHopfieldBoltzmannBAMSOMBPBPBP3.1.2BPHopfieldHebb3.1.3BP1986Rumelhard,HintonWilliamsANN(ErrorBack-PropagationTrainingBP)19,BPBP(1)(2)(3)(4),BPBP33-3xywBP21()2piiiEyt=−∑(3.2)PkiykiitiBPBPS()1/1xfxe−=+20x1x2x3xny1y2y3y43-3BPFig3-3BPneuralnetworkmodel3-30-10expectedtargetvaluespredictedvalues01regressioncoefficients1050MLPMLP21[34]BP12BP112{,,...,}kkkknXxxx=(k=1,2,…,p)kN12{,,...,}kkkkmYyyy=m1nkkjjiiiinetwxθ==+∑(j=1,2,,l)(3.3)jiwiθ22lSSBP1()1kjkkjjnetofnete−==+(3.4)1()()lkkktttjjtjyfnetfTOθ∧===+∑(t=1,2,…,m)(3.5)tjTtθK211()2mkkkttiEyy∧==−∑(3.6)tjTjiwtδjδ(1)()()kktjtjtjtjtjThThTThOηδ+=+∆=+(3.7)(1)()()kkjijijijijiwhwhwwhxηδ+=+∆=+(3.8)(1)()()kttttthhhθθθθηδ+=+∆=+(3.9)23(1)()()kjjjjjhhhθθθθηδ+=+∆=+(3.10)ηhBP21111()2ppmkkkttkktEEyy∧=====−∑∑∑(3.11)3.2BP3.2.1BP1.{0,1}2.3.244.5.6.7.nε(ε)253.2.2BP1N12{,,...,}TNXXXX=12{,,...,}inXxxx=11212nrsssss=×+×++(3.1)r1s2s2BP:1()1fxE=+(3.2)E211()2NkkkEyy∧==−∑(3.3)(,);(1,2,...)kkxykN=ky∧263GAGAGAsp1(1)spsptp=−−(3.4)()1(1)Nkksptp−=−k=1,2,…,p(3.5)()Nkk1kkjjqp==∑k=1,2,…,p(3.6)[0,1]r4GAcpcn2ccpNn=(3.7)27[]iXjXl'iX'jX'()0()()(1)()iiijXkklXkXkXklknλλ≤=+−≤(3.8)'()0()()(1)()jjjiXkklXkXkXklknλλ≤=+−≤(3.9)1,2,...,cln=λ[01]λcn5mpbmn[]mmnpN=(3.10)lX,(1,2,...,mln=)[0,1]mλ()1bclmmgdXygλ=−(3.11)cgmg01llbXsignyXbsign−==−=(3.12)sign01'lX28'()0()1lllllXdXsignXXdXsign+==−=(3.13)mnnn3.2.3BPGALNcp,mpBPBP3input-hiddenZ1N=602,fitness()1/iFiconstE=+iEconst0F(i)3Z()Piαβ01MaxFitnessC()Pi0.6-0.9C=/()()()(1)FiFiMaxFitnessMaxFitnessPiCαβ=−+(3.14)()Pi290.6-0.9L22()PiL[1L-1]()(0,1)WaFiµ∆=a-11(0,1)µ()Pi0.01-0.23.3BP9620BP17BP3-120001S0-1input[32][10]101-9