支持向量機及其在汽輪機組性能監(jiān)測和故障診斷中的應用研究

1、東南大學博士學位論文支持向量機及其在汽輪機組性能監(jiān)測和故障診斷中的應用研究姓名：王雷申請學位級別：博士專業(yè)：熱能工程指導教師：徐治皋20070601AbstractAbstractSafeandeconomicaloperationofturbineunitsisalwaysconcernedPerformancemonitoringandfaultdiagnosistechnologyprovideprerequisiteforsaf

2、eandeconomicaloperationofturbineunitsSoitisveryimportanttostudyperformancemonitoringandfaultdiagnosismethodofturbineunit。Supportvectormachine(SVM)isamachinelearningmethodsbasedonthestatisticallearningtheory(SLT)，whichhas

3、manyadvantagesinpaUernrecognitionsuchassuperiorityinsmallsample，nonlinear，overfiRingandhigh—dimensionproblemsTheknowledgeimplicatedinthedatacanbescoopedoutunderthe1imitedcharacteristicinformmionconditionbythegreatestexte

4、ntFromthepointofviewforapplication，SVMisamethodthatismoreapplicabletostudytheperformancemonitoringandfaultdiagnosisofturbineunitinthepowetplantthanothersBasedonthis，SVMisappliedinperformancemonitoringandfaultdiagnosisoft

5、urbineunitTheestablishmentandrealizationofSupportvectorregression(SVR)model，MulticlassSVMmodelandtheselectionoftheparam咖rsmdiscussedinthepaper111emaincontentsofthethesisinclude：Accordingtodatacharacteristicofperformancep

6、arameterinpowetplant，performanceparametersandcalculationofNormVacuumaboutunitarestudiedbyusingregressionmethodSincesupportvectorregression(SVR)isadeptindealingwithsmallsampleproblem，anewonlinecalculatingmethodofperforman

7、ceparametersisraisedbasedonSVRFeasibilityandsuperiorityaboutmodelingmethodbasedonSVRarcanalyzedintheparametersmonitorandbasicprocessandparametersselectionofSVRmodelarediscussedmainlyinthedissertationAtthesarnetime，module

8、saboutNormVacuumofcondenserandheatrateareestablishedTheforecastmodelofrunningparametersbasedontimeseriesispresentedTocountertheproblemofrunningparameterschangingpedodicitywhenunitisoperating，viewofpointoncombiningsupport

9、vectorregressionmodelwithAutoregressive似R)modelisputforwardThentheSupportVectorAutoregressive(SVAR)modelappliedinpracticeproblemisestablishedbywhichnonlinearsamplecanbcdiscussedThatlaysthefoundationondiscussingfuturetren

10、dofoperatingparameters，Counteringtheoutlierproblemofthereal—timedatainthepowerplant，amethodofoutlierdetectionisproposedbasedonthecharacterofstructureriskfunctionandKKTconditionofsuppoMvectorregressionThemethodofdealingwi

11、thoutlieranditscharacteristicssuperiortoothermethodsarediscussedModelingtheorymodelingprocessandengineeringapplicabilityareintroducedparticularlyTosolvetheproblemexistedintraditionalpaRernrecognitionmethodwhenthesampleis

12、finiteinfailuremodeclassification；MulticlassSVMisappliedtofaultdiagnosisHowSVRisappliedtofaultdiagnosisisdiscussed，andthemainstepandalgorithmprocessaregivenThenmulti—classfaultmodelaboutcondenserplantofpowergenerationtur

13、bineisputforwardusingDAGSVMalgorithmwhichisgained蚵combiningoneagainstonealgorithmwithdecisiontreeAtlastthefactorsaffectingfaultcategoryperformancearediscussedin—de：pthKeywords：powerplant，turbineuni‘supportvectormachine，p