1、重慶大學(xué)碩士學(xué)位論文基于偏最小二乘的BP網(wǎng)絡(luò)模型及其應(yīng)用姓名：張艷粉申請(qǐng)學(xué)位級(jí)別：碩士專(zhuān)業(yè)：概率論與數(shù)理統(tǒng)計(jì)指導(dǎo)教師：劉瓊蓀20070420重慶大學(xué)碩七學(xué)位論文英文摘要ABSTRACTThepartialleastsquaremethodisamultivariatestatisticalanalysismethodT11ismethodhasitspopularutilityintermsofchemicalplant，financi

2、alanalysis，marketanalysis，medicamentanalysis，computationchemical，industrydesignetcThismethodsettlespreferablytheproblemwhichdoesn’tbeensettlebythecommonmultivariatelinearregressionThismethodappliesmultivariateregressionc

3、anonicalcorrelationalanalysisandprincipalcomponentanalysistodataanalysisItrealizesmanifolddataanalysismethodThebackpropagationalgorithmhasbeenwidelyrecognizedasalleffectivemethodfortrainingfeedforwardneuralnetworksThisme

4、thodhasitspopularutilityintermsofpaRemrecognition，controlengineering，signalprocessingandeconomicpredictionetcHowever，thereexistsseveralessentialdefectsinback—propagationalgorithmasfollows：slowconvergence，trapinlocalminim

5、aworsetolerantcapacityandthereisnottheuniformtheorywhichtheweightsinitializationofinputfloorandoutputfloorareset，thememberofhiddennodesaresetandthememberofhiddenfloorsaresetTllispaperimproveonthedefectwhichthereisnottheu

6、niformtheorywhichtheweightsinitializationofinputfloorandoutputfloorareset，thememberofhiddennodesaresetandthememberofhiddenfloorsalesetinBPnetworkFirstly，wepropoundanovelBPnetworkmodelbasedonnonlineariterativepartialleast

7、squaresalgorithmwhichCanfitnonlineardataafterweexpoundpartialleast—squaresalgorithmsystematicallyinthispaperandwevalidatethisnewmodelandcomparetotheBPPLSmodelsforinstance111eresultshowedthatthenewmodelhadbeRerfittingandf

8、orecastthanBP，PLSmodelsThus，thedefectwhichthereisnottheuniformtheorywhichtheweightsinitializationofinputfloorandoutputfloorareset，thememberofhiddennodesaresetandthememberofhiddenfloorsaresetinBPnetworkissettledpartlyThis

9、modelCanreduceiterativestepnumberandadvancelearningefficiencyacertainextentalsoadvanceavailabilitytheperformanceofback—propagationalgorithmSecondlywebaseonthesalneideaWeusetheorthogonaisignalcorrectionmethodtosettlethede