一種新的神經(jīng)模糊系統(tǒng)及其對(duì)“當(dāng)?shù)睾?jiǎn)單遠(yuǎn)程復(fù)雜”控制原理的實(shí)現(xiàn).pdf

1、中國(guó)科學(xué)院自動(dòng)化研究所碩士學(xué)位論文一種新的神經(jīng)模糊系統(tǒng)及其對(duì)“當(dāng)?shù)睾?jiǎn)單遠(yuǎn)程復(fù)雜”控制原理的實(shí)現(xiàn)姓名：陳龍申請(qǐng)學(xué)位級(jí)別：碩士專(zhuān)業(yè)：控制理論與控制工程指導(dǎo)教師：王飛躍20030629AbstractBecauseofitswellcapabilityinexpressingtheexperts’knowledgeandsimulatingtheinferenceprocessofhumanbeing，fuzzyinferencesystem

2、isappliedwidelyHowever，peoplefoundsoonthatfuzzyinferencesystemiSnotgoodatlearningfromdataThenthelearnableneuralnetworkiScombinedwiththefuzzyinferencesystemtobuildthe‘NeuEo—fuzzySystem”Therearesomedisadvantagesinthetradit

3、ionalneuro—fuzzysystemsandtocopwiththosedisadvantagesthisPaFlerprovidesanewneuro—fuzzysystemThenewsystemusingMamdartitypefuzzyrulesavoidsthedisadvantagesofbeinghardt0incorporateexperts’knowledgeandhardt0understandInthese

4、condplace，thenewsystemusesdiscretemembershipfunctionsattheconsequentpartsoftheIF—THENrulesSoanyoriginalmembershipfunctioncarlbeapproximatedbyadiscretemembershipfunctionandtileshapeofthemembershipfunctioncanbeadjustedintr

5、ainingprocedureintumThenewneurofuzzysystemincludesafuzzyinferencesystemanditsonetoodemappingneuralnetwork，andthisPaperalsoprovidestheRBFneuralnetworkbasedstructurallearningalgorithmandthegradient—basedparameterslearninga

6、lgorithmforthenewsystemThisPaperalsosuccessfullyappliedtheflewneuro—fuzzysysteminseveralirrlportantproblemsofIntelligentTransportationS，stemsWiththedevelopmentofcontroltheory，theimplementationofcontrolalgorithmsmoreandmo

7、redependsonthepowerfulcomputationalsystems，Ontheotherhand，pe。p／eareaskingforsmallerandsimplercontroller。Thentheconflictofcomplexcomputationandsimpleimplementationappears，T0solvetheconflict，“LocalSimpleRemoteComplex”contr

8、olprincipleiSaoodchoiceTWOcorrespondentpartsatlocalandremotefie[dsareintegrantOartstoimplementthe“LocalSimpleRemoteComplex”controlprincipleThe10calpartshouldbesimpleandnoneedtobelearnablebuttheremotepartshouldhavewclIcap

9、abilityoflearffmgandoptimizingparametersusedatthelocalpartThispaperutilizestheonetOonemappingcharacteristicofthenewneuro—fuzzysystemandputsthesimplefuzzyinfererlcesystematthelocalfieldwhileputSthelearnablecomplexneuralne

10、tworkattheremotefieldRemoteneuralnetworkgetsthetrainingdatafromthelocalfieIdandoptimizesthepararoeters，whichinturnalesentbacktOthelocaIfuzzyinferencetooptimizeitsperformanceInthelongrun，the“LocalSimpleRemoteComplex”contr