1、中圖分類號：TPl83密級：單位代號學號1190302720913上海大學⑧碩士學位論文SHANGHAIUNIVERSITYMASTER’STHESIS憎人工神經網(wǎng)絡技術在狀態(tài)監(jiān)測和漂移預測中的應用作者學科專業(yè)沈文煒電路與系統(tǒng)導師施惠昌研究員完成日期2005年2月海人學顧I‘學位論文摘要AbstractArtificialNeuralNetworks(ANN)hasbeencatchingtheworld’sattentionsince

2、itcameintobeinginthefortiesofthetwentiethcenturyStudiesshowthatneurafnetworksarethemostpowerfultoolstoapproximatealmosteverynonlinearsystemandhavereliableaccuracyNowithasbeendevelopedtobeausefulnon—linearprocessingtool，a

3、ndisusedwithgreatsuccessinmanyfieldsDynamicPredictionandStatusMonitoring＆FaultDiagnosesistwoofimportantapplicationfieldsofANNThefoIlowingarewhatthispaperhasdoneinthefieldsofDriftPredictionandStatusMonitoring＆FaultDiagnos

4、esTobeginwiththeStatusMonitoring＆FaultDiagnosesusingneuralnetworksisdiscussedinthispaperWeinvestigatetheworkingstatusandfaultofrotorandstatoringeneratorRadialBasisFunctionNeuralNetworks(RBFNN)andInformationNeuralNetworks

5、(INN)modelsandlearningalgonthmsareshowedindetailThenaneuralnetworkwithnewstructure，RBINNisdesignedforfaultdiagnosisofgeneratorThekindofneuralnetworksintegratesthecharacteristicsofRBFNNandINNDiagnosingresultsshowthatthene

6、uralnetworkswithnewstructurecancorrectlyidentifythefaultofthestatusandfaultofrotorandstatoringenerator。Next，DynamicpredictionisalsodiscussedinthepaperTimeseriespredictionisanimportantaspectofdynamicdataanalysisandprocess

7、ingThereexiststhedemandofpredictingthefuturebaseduponhistoricaldataintheapplicationsofscience，economicsandengineeringInthefaceofthenonlinearandnon—stationarytimeserieslargelyexistedinnaturalandsocialeconomicalphenomena，t

8、hetraditionalmethodofstatisticalanalysisperformedbadlySince1987，LapedesandFarberfirstlyappliedneuralnetworkstotimeseriespredictionThensuchapredictingmethodwithneuralnetworkshaswidelybeenrecognizedPresentlymanydifferentty