RBFNN研究及其在鐵水脫硫靜態(tài)模型中的應(yīng)用.pdf

1、重慶大學(xué)碩士學(xué)位論文RBFNN研究及其在鐵水脫硫靜態(tài)模型中的應(yīng)用姓名：陳才申請學(xué)位級別：碩士專業(yè)：控制理論與控制工程指導(dǎo)教師：曹長修20030508重慶大學(xué)碩士學(xué)位論文英文摘要ABSTRACTSulfurisaharmfulelementinmostkindsofsteel，SOjtiSnecessarytemoveitfromtheironwaterbeforesteelmakingThepresentdesulfurationtec

2、hniqueisstillbasedonmanualcontrolthatdependsontheexperiencesofworkers，SOitlowerstheautomationlevelofthefactoryandthestabilityofsteelqualityBecausemostofthepopulardesulfurationcontrolmodelsabroadarebasedondesulfurationmec

3、hanism，whichrequiretheverystableproductionsituation，itisdifficulttotransplantthemtothedifferentfactoriesAnalyzingthetechniquesandfactorsofdusufurationthisdissertationconstructsanitonwaterdesulfurationstaticintelligentpre

4、dictionmodelbasedonRBFneuralnetworkFirsfly，thedissertationcarefullyintroducesthebasictheoryofRBFneuralnetwork，andthenconductsadeepresearchinthenetworkstructurelearning：thecenternumbers，spreadandweightThemostimportantfact

5、orofRBFneuralnetworkisthecenterselectionwhichinfluencestheperformanceofRBFgreatlyAkindofrivalpenalizedcompetitiveleaming(RPCUtechniqueisimplementedtodeddethecentersofRBFbywhichthecenternumbersareacquiredautomaticallyInor

6、dertooptimizethecentersofRBFmoreeffectivelythisdissertationintroducestheinput—outputclusteringmethodandcombinesitwitllRPCLalgorithminclusteringThespreadofRBFaffectsthegeneralizationoftheneuralnetworkdirectlyLargespreadwi

7、llleadinaccuracyonthecontrarysmallspreadwillharmthegeneralizationTherefore，thechaossearchalgorithmisadoptedtooptimizethespreadofRBFTosolvetheproblemofthedeterminationoftheregularizationcoefficientintheweightlearningofreg

8、ulariZtionRBFthedissertationprovidesEMalgorithmtolearntheweightwithoutdeterminingthevalueofregularizafioncoefficientinadvanceFinallyallironwaterdesulfurationstaticpredictionmodelisconstructedbytheoptimizedRBFneuralnetwor

9、k111einputparametersincludetheweightofironwaterthequantityofsulfurbeforedesulfurationandthequantityofsulfurafterdesulfurationTheoutputparameteristheweightofdusulfuretionagentTheoffiinesimulationisconductedandtheresultssh

10、owthefollowingconclusion：1111eimprovedRPCLalgorithmcadnotonlydeterminethecenternumbersofRBFbutalsooptimizethelocationofthesecenters，whichimprovedtheaccuracyofthemodel；2Thechaossearchalgorithmcanfindthesuitablespread，whic