支持向量机在水利水电工程中的应用

河北工程大学硕士学位论文支持向量机在水利水电工程中的应用姓名：王海燕申请学位级别：硕士专业：水利水电工程指导教师：赵林明20070401II20041kW1/41/56.94kW5.42kW4.02kW17500kW·h19591990AbstractIIIAbstractHydropowerinstalledcapacityhadexpended100,000MWsince2004,accountedaquarterofthewholenationalpowerinstalledcapacity,providedonefifthofthecountrywideelectricpowerrequirements.Inrecentyears,China’spowerconsumptionhasbeinggrownrapidly,andmanypartsofthenationareshortforfossilpowersupplying;whileChinaisaffluentinhydropowerresources,it’stheoreticalpotentialofhydropowerresourcesis694,000MW,technicalexploitablecapacityis542,000MW,economicexploitablecapacityis402,000MWandeconomicexploitableannualpoweroutputis1,750billionKW·h-1.Inordertoalleviatethecontradictionofelectricitysupplyanddemand,thenationhasincreasedtheinvestmenttothewater&electricityinvestments.ThegeneralpolicyofChinesepowersystemdevelopmentwhichis“powertransmissionfromwesttoeast,mutualsupplybetweensouthandnorth,nationwideinterconnection”,offersnewopportunitytoChina’shydro-energyproduction.Whileinthecourseofhydro-energyproduction,therearemanyproblemsneedtobestudied,andforthedrawbacksoftraditionalalgorithms.ThispaperfocusesontheapplicationofSupportVectorMachine(SVM)incorrelationareaofhydroelectricpowergeneration.ThemaincontainsofthepaperareFirstly,thepaperdescribesthenecessity,availability,anddevelopingtendencyoftheapplicationofSupportVectorMachineinrunnerforecasting,vibrationpatternrecognitionandcharacteristicfittingofhydraulicelectricgeneratingsets;Secondly,thepaperfocusesonthebasictheoriesofSupportVectorMachine,CellularAutomaton(CA),andsoon;Thirdly,takingtheaverageflowfrom1959to1990bymonthsofTianqiaoduanhydraulicstationupstreamofZhangRiverHandanasoriginaldata,thepaperbuildsthehelicalcellularautomatamodeloftherunoffforecastbasedonthebasictheoryofcellularautomaton,andusesSupportVectorRegressiontodotherunnerforecast;Fourthly,thepaperanalyzestheproductionmechanismandformationcharacteristicsofthepressurepulsationinturbinedraft-tubesanduseSupportVectorMachinetoanalyzethepressurepulsationconditionsinclusters,basedonabstractingthevibrationcharacteristicsofdrafttubeinwaveletpacketmethod;Fifthly,becausethecorrespondingcurvewhichexpressedtheturbinerunnerAbstractIVcavitationsandpressurefluctuationcharacteristicsaretoocomplextobeanalyzed,thepaperusessupportvectormachinetoestablishthemodelwhichcanexpresscavitationsandpressurefluctuationcharacteristicsatthesametimeandappliesittotheestablishmentofthetwooutputmodelofhydraulicturbinecavitationsandpressurefluctuationcharacteristicsinZipingpuhydro-plantSichuanprovince.Thispaperfocusesonwithdrawingtheactualproblemcorrelationdataofwaterconservancyandhydroelectricityengineering,buildingthemodelbasedonthecoherenttheory,andthenusingsupportvectormachinetofitoranalyzethemodel.ThemainachievementsareasfollowingFirstly,thepaperbuildsthehelicalcellularautomatamodel;Secondly,thepaperusesSupportvectormachinetodotherunoffforecast;Thirdly,thepapercreatesthetwooutputmodelofcavitationsandpressurefluctuationcharacteristicsofhydraulicturbine.Fourthly,thepaperestablishesthepressurepulsationconditionsclusteranalysismodel.KeywordsSupportVectorMachine(SVM);CellularAutomaton(CA);WaveletPacket;RunoffForecastI11.1[1]2004930[2]2070Box[3](autoregressivemoving-average,ARMA)90(ANN)[45](supportvectormachines,SVM)(structuralriskminimizationprinciple,SRM)VC[6][7]SVMLiong[8]SVM1989~19902SVM(FeatureSpace)BPBP1.21.2.17010338195919927[9-13]490[14]HsuBP(LLSSIM)BP[15][16]BPBP(TolerantCapacity)[17]1.2.2(FFT)FFT.(×15×2)FFT[18][19][20]ARARMAFFT[21][22](WP)(MESE)(WP-MESE)1.2.311nα11Qη11M6nm10%(srad2ω)[23]1.371.4SVM1-11-1Fig.1-1Thepaperframe82.1.SupportVectorMachineSVMSVMVapnikCOLT-9219921995VapinkTheNatureofStatisticalLearningTheoryRBFCellularAutomataCACACA10208092.2.2.2.1[24]Vapnik[2526]202-12-1Fig2-1Thecomparisonbetweentwokindsofclassificationlines2-1ab10[27][28]ChunkingalgorithmSequentialMinimalOptimization(SMO)SVMWeston1998()yxK,MercerMercer[29]2-1()yxK,)(xϕ∞∫dxx)(2ϕ0)()(),(∫∫dxdyyxyxKϕϕpyxyxK]1,[),(+=)exp(),(22σyxyxK−−=)exp(),(2σyxyxK−−=],tanh[),(cyxvyxK+=11Fourier))(21sin())(21sin(),(yxyxNyxK−−+−=B()yxByxKN−=+12),(AmariWu[30]2.2.2[31]SVMV.Vapnik601971V.VapnikA.ChervonenkisTheNecessaryandSufficientConditionsfortheUniformsConvergenceofAveragestoExpectedValuesSVMVC1982EstimationofDependenceBasedonEmpiricalDataV.VapnikSVM1992BoserGuyonandV.VapnikATrainingAlgorathmforOptimalMarginClassifiers1993CortesandV.VapnikTheSoftMarginClassifier1997V.VapnikS.GokowwichandA.SmolaSupportVectorMethodforFunctionApproximationRegressionEstimationandSignalProcessingSVMSVMSVM1998SmolaSVMSVMSVM[32][33][34][35][3637]SVM12[3839]DeKruif[40]Suykens[41][42-44][45]Chen[46]DS-CDMACao[47]SVM2.2.3liiiyx1)},{(=ixiiy1+=iy1−=iy0=+⋅bxωxωb2-2HH1H2H1H2(margin)(0)(OptimalHyper-plane)132-2Fig.2-2Sketchmapoftheoptimalhyperplane01)(1111≥−+⋅⇔−=−≤+⋅+=+≥+⋅bxyyforbxyforbxiiiiiiωωω(2-1)ωωωωω2maxmin11=+⋅−+⋅−=+=bxbxiyxiyxiiii(2-1)ω22ω(2-1))(2121)(2ωωωωΦ⋅==(primalproblem))(ωΦωωLargrangeLargrange∑=−+⋅−=liiiibxybL12]1)([21),,(ωαωαω0iαLargrangeLargrange),,(αωbLωbα),,(αωbL