支持向量机及其在汽轮机组性能监测和故障诊断中的应用研究(I)

《支持向量机及其在汽轮机组性能监测和故障诊断中的应用研究(I)》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、东南大学博士学位论文支持向量机及其在汽轮机组性能监测和故障诊断中的应用研究姓名：王雷申请学位级别：博士专业：热能工程指导教师：徐治皋20070601AbstractSafeandeconomicaloperationofturbineunitsisalwaysconcerned．Performancemonitoringandfaultdiagnosistechnologyprovideprerequisiteforsafeandeconomicaloperationofturbineunits．Soitisveryimportanttostudyperformancemonitoringa

2、ndfaultdiagnosismethodofturbineunit。Supportvectormachine(SVM)isamachinelearningmethodsbasedonthestatisticallearningtheory(SLT)，whichhasmanyadvantagesinpaUernrecognition,suchassuperiorityinsmall·sample，nonlinear，over-fiRingandhigh—dimensionproblems．Theknowledgeimplicatedinthedatacanbescoopedoutunder

3、the1imitedcharacteristicinformmionconditionbythegreatestextent．Fromthepointofviewforapplication，SVMisamethodthatismoreapplicabletostudytheperformancemonitoringandfaultdiagnosisofturbineunitinthepowetplantthanothers．Basedonthis，SVMisappliedinperformancemonitoringandfaultdiagnosisofturbineunit．Theest

4、ablishmentandrealizationofSupportvectorregression(SVR)model，Multi-classSVMmodelandtheselectionoftheparam咖rsmdiscussedinthepaper．111emaincontentsofthethesisinclude：Accordingtodatacharacteristicofperformanceparameterinpowetplant，performanceparametersandcalculationofNormVacuumaboutunitarestudiedbyusin

5、gregressionmethod．Sincesupportvectorregression(SVR)isadeptindealingwithsmall-sampleproblem，anewon-linecalculatingmethodofperformanceparametersisraisedbasedonSVR．FeasibilityandsuperiorityaboutmodelingmethodbasedonSVRarcanalyzedintheparametersmonitor,andbasicprocessandparametersselectionofSVRmodelare

6、discussedmainlyinthedissertation．Atthesarnetime，modulesaboutNormVacuumofcondenserandheatrateareestablished．Theforecastmodelofrunningparametersbasedontimeseriesispresented．Tocountertheproblemofrunningparameterschangingpedodicitywhenunitisoperating，viewofpointoncombiningsupportvectorregressionmodelwi

7、thAutoregressive似R)modelisputforward．Then,theSupportVectorAutoregressive(SVAR)modelappliedinpracticeproblemisestablished,bywhichnon-linearsamplecanbcdiscussed．Thatlaysthefoundationondiscussingfuturetrendofo