搜索
目录Part1PIDtypefuzzycontrollerandparametersadaptivemethod.........1Part2Applicationofselfadaptationfuzzy-PIDcontrolformainsteamtemperaturecontrolsysteminpowerstation错误!未定义书签。Part3Neuro-fuzzygeneralizedpredictivecontrolofboilersteamtemperature.....................................................................………13Part4为Part3译文：锅炉蒸汽温度模糊神经网络的广义预测控制211Part1PIDtypefuzzycontrollerandParametersadaptivemethodWuzhiQIAO,MasaharuMizumotoAbstract:Theauthorsofthispapertrytoanalyzethedynamicbehavioroftheproduct-sumcrisptypefuzzycontroller,revealingthatthistypeoffuzzycontrollerbehavesapproximatelylikeaPDcontrollerthatmayyieldsteady-stateerrorforthecontrolsystem.ByrelatingtotheconventionalPIDcontroltheory,weproposeanewfuzzycontrollerstructure,namelyPIDtypefuzzycontrollerwhichretainsthecharacteristicssimilartotheconventionalPIDcontroller.Inordertoimprovefurthertheperformanceofthefuzzycontroller,weworkoutamethodtotunetheparametersofthePIDtypefuzzycontrolleronline,producingaparameteradaptivefuzzycontroller.Simulationexperimentsaremadetodemonstratethefineperformanceofthesenovelfuzzycontrollerstructures.Keywords:Fuzzycontroller;PIDcontrol;Adaptivecontrol1.IntroductionAmongvariousinferencemethodsusedinthefuzzycontrollerfoundinliteratures,themostwidelyusedonesinpracticearetheMamdanimethodproposedbyMamdaniandhisassociateswhoadoptedtheMin-maxcompositionalruleofinferencebasedonaninterpretationofacontrolruleasaconjunctionoftheantecedentandconsequent,andtheproduct-summethodproposedbyMizumotowhosuggestedtointroducetheproductandarithmeticmeanaggregationoperatorstoreplacethelogicalAND(minimum)andOR(maximum)calculationsintheMin-maxcompositionalruleofinference.Inthealgorithmofafuzzycontroller,thefuzzyfunctioncalculationisalsoacomplicatedandtimeconsumingtask.TagagiandSugenoproposedacrisptypemodelinwhichtheconsequentpartsofthefuzzycontrolrulesarecrispfunctionalrepresentationorcrisprealnumbersinthesimplifiedcaseinsteadoffuzzysets.Withthismodelofcrisprealnumberoutput,thefuzzysetoftheinferenceconsequencewill2beadiscretefuzzysetwithafinitenumberofpoints,thiscangreatlysimplifythefuzzyfunctionalgorithm.BoththeMin-maxmethodandtheproduct-summethodareoftenappliedwiththecrispoutputmodelinamixedmanner.Especiallythemixedproduct-sumcrispmodelhasafineperformanceandthesimplestalgorithmthatisveryeasytobeimplementedinhardwaresystemandconvertedintoafuzzyneuralnetworkmodel.Inthispaper,wewilltakeaccountoftheproduct-sumcrisptypefuzzycontroller.2.PIDtypefuzzycontrollerstructureAsillustratedinprevioussections,thePDfunctionapproximatelybehaveslikeaparametertime-varyingPDcontroller.Sincethemathematicalmodelsofmostindustrialprocesssystemsareoftype,obviouslytherewouldexistansteady-stateerroriftheyarecontrolledbythiskindoffuzzycontroller.ThischaracteristichasbeenstatedinthebriefreviewofthePIDcontrollerintheprevioussection.Ifwewanttoeliminatethesteady-stateerrorofthecontrolsystem,wecanimaginetosubstitutetheinput(thechangerateoferrororthederivativeoferror)ofthefuzzycontrollerwiththeintegrationoferror.Thiswillresultthefuzzycontrollerbehavinglikeaparametertime-varyingPIcontroller,thusthesteady-stateerrorisexpelledbytheintegrationaction.However,aPItypefuzzycontrollerwillhaveaslowrisetimeifthePparametersarechosensmall,andhavealargeovershootifthePorIparametersarechosenlarge.Sotheremaybethetimewhenonewantstointroducenotonlytheintegrationcontrolbutthederivativecontroltothefuzzycontrolsystem,becausethederivativecontrolcanreducetheovershootofthesystem'sresponsesoastoimprovethecontrolperformance.Ofcoursethiscanberealizedbydesigningafuzzycontrollerwiththreeinputs,error,thechangerateoferrorandtheintegrationoferror.However,thesemethodswillbehardtoimplementinpracticebecauseofthedifficultyinconstructingfuzzycontrolrules.Usuallyfuzzycontrolrulesareconstructedbysummarizingthemanualcontrolexperienceofanoperatorwhohasbeencontrollingtheindustrialprocessskillfullyandsuccessfully.Theoperatorintuitivelyregulatestheexecutortocontroltheprocessbywatchingthe3errorandthechangerateoftheerrorbetweenthesystem'soutputandtheset-pointvalue.Itisnotthepracticefortheoperatortoobservetheintegrationoferror.Moreover,addingoneinputvariablewillgreatlyincreasethenumberofcontrolrules,theconstructingoffuzzycontrolrulesareevenmoredifficulttaskanditneedsmorecomputationefforts.HencewemaywanttodesignafuzzycontrollerthatpossessesthefinecharacteristicsofthePIDcontrollerbyusingonlytheerrorandthechangerateoferrorasitsinputs.OnewayistohaveanintegratorseriallyconnectedtotheoutputofthefuzzycontrollerasshowninFig.1.InFig.1,1Kand2Karescalingfactorsforeand~respectively,andflistheintegralconstant.Intheproceedingtext,forconvenience,wedidnotconsiderthescalingfactors.HereinFig.2,whenwelookattheneighborhoodofNODEpointinthee-~plane,itfollowsfrom(1)thatthecontrolinputtotheplantcanbeapproximatedby(1)Hencethefuzzycontrollerbecomesaparametertime-varyingPIcontroller,itsequivalentproportionalcontrolandintegralcontrolcomponentsareBK2DandilK1Prespectively.WecallthisfuzzycontrollerasthePItypefuzzycontroller(PIfc).WecanhopethatinaPItypefuzzyco