基于小波神經(jīng)網(wǎng)絡的高爐爐溫預報模型研究.pdf

1、浙江大學碩士學位論文基于小波神經(jīng)網(wǎng)絡的高爐爐溫預報模型研究姓名：楊可蘋申請學位級別：碩士專業(yè)：運籌學與控制論指導教師：劉祥官20060501浙江大學碩士學位論文ABSTRACTAsthemainupperprocedureofmetallurgicalindustryBlastFurnace(BF)ironmakingisanimportantcomponentofthepillarindustryinnationaleconomywh

2、ichplaysasignificantroleinenergysavingandtechnicaldevelopmentofthewholeindustryTheprocessofironmakingishighlycomplicated，whoseoperatingmechanismischaracteristicofnonlinearitytimelag，highdimension，bignoiseanddistributionp

3、arametersetcTomaintaintheappropriatetemperatureinBFiscrucialforasmoothironmakingprocessDuetothehigbcomplexityitisalmostimpossibletomeasuretheexacttemperatureofhotmetalinBFThusthedifficultyofironmakingprocessautomationlie

4、sintheconstructionofaneffectivepredictioncontrolmodeltomakeforecastoftemperatureofhotmetalinBFSiliconcontentinhotmetalisnotonlyanimportantindextoevaluatethestatusofironmakingprocessandqualityofhotmetal，butalsothemostimpo

5、rtantparametertoreflecttemperatureofhotmetalinBFToeffectivelycontroltheBFironmakingprocess，predictionofthesiliconcontentbecomesanjmportantquestionindiscussionThemethodofwaveletnetworksisrelativelynewintimeseriesanalysis，

6、whichinheritsboththevirtueofwaveletinregionalizingoftimefrequencyfieldandselflearningabilityofneuralnetworksandhasstrongcapabilityinfunctionapproximationanderrortoleranceTherearetwocombinationalmethodsforwaveletandneural

7、networksThefirstonecalledwaveletnetworkswithrelaxedstructureistodeformtheinputseriesintodifferentlevelsofresolutionscalesbythewavelettransform，andforeachlevelofcorrespondinginputparameters，weuseneuralnetworkstogettheoutp

8、utvaluesAreconstructivetransformationisappliedtotheoutputvaluesofneuralnetworksandtheforecastsaregeneratedTheothercalledwaveletnetworkswithcompactstructurejstosubstitutethebasefunctionioneuralnetworkswithwaveletfunctiona

9、ndthewaveletfunctionisadjustedadaptivelytodowavelettransformandnewnetworksareenabledThispaperUSeSdatacollectedfromIntelligentControlExpertSysteminHandanSteelCorporationaSsampledatatoimplementthealgorithmsThesamplecontain

10、s500datapairsWeapplythethreemodelsbasedonthetheoryofwaveletanalysisandneuralnetworkstodoforecastsandtheresultsareanalyzedndetailItisshownthatthemethodofneuralnetworkswithcompactstructureCangivebetterhitratesofpredictionf