InternationalTechnologyandInnovationConference2007THEDESIGNOFFUZZYPIDMULTI-CHANNELTEMPERATURECONTROLSYSTEMBASEDONNEURALNETWORKZHANGYongjun*,WANGZhixing*,WANGLilit*HarbinInstituteofTechnology,Harbin150006,China,zhangyongjun916@sohu.comtHarbinUniversityofScienceandTechnology,Harbin150080,China,gaisidiwenwll@yahoo.com.cnKeywords:FuzzyPIDcontrol,neuralnetwork,controlsystem,temperaturemeasurement.AbstractBasedonfuzzyneuralnetworkPIDcontroltechnology,thepaperdesignsamulti-channeltemperaturesystemusedinpetrochemicalindustry.ByadjustingPIDcontrolparametersthroughfuzzyneuralnetwork,thesystemperfectstheauto-adaptingperformanceofPIDcontrol.Themaincomponentoftemperaturemonitorandcontrolhardwaresystemismicro-controllerC8051F020.Thishardwaresystemcouldrealizetemperaturedetectionandcontrolonsixteenmonitorpoints,andtransmitsthetemperaturesignalthroughCANbustouppercomputerreal-timely.Theresultsofexperimentsprovethatthissystemsatisfiesoperationrequirements,andhasnicerdynamicstate,staticstateandauto-adaptabilitystate.experienceofexperts,whicharefixedinthelanguageruleandinferenceprocess,anditissupposedthatcontrolprocesscannotmakeremarkablechangebeyondtheseexperiencescopes,sotheparametershavecertainlimitation.Neuralnetworkcouldobtaintherulesdirectlyfromthedatasamplewithoutusingdomainknowledge,sothefuzzyneuralnetworkhasfuzzyinferenceability,self-study,self-adaptability,faulttoleranceandparallelism-in-handlingability.ThepaperusestheneuralnetworktoadjustPIDparameterstoperfecttheself-adaptabilityofPIDcontrol,anddesignsahardwarecontrolsystemmulti-channeltemperaturecontrolcombiningwithmicro-controllerC8051F020.Theresultsofapplicationprovethatthehardwaresystemhasgoodrobustness,dynamictrackingqualityandstablestateprecision.2ManuscriptpreparationControlalgorithmofsequencePIDis:1tde(t)u(t)=Kp[e(t)+-foe(t)dt+Td-]11dt1IntroductionInpetrochemicalindustry,inordertoensuretheequipmentstorunstablyandensuretheproductquality,severalrequirementsofworkingenvironmentandworkingtemperaturetotheequipmentshasstricthavebeenlistedstrictly.Buttheprocessofthermoregulationisthetypicalnon-linearity,strongcoupling,timevarying,andtimelagprocess,whichisabigdeferredprocess.Owingtothebigdeferring,thesystemhasseriousnonlinearityandtimevaryingcharacteristic,andtherearemanydisturbancefactorsaffectingthetemperatureandenvironmentchangingoftheequipments,whichmakethethermoregulationverydifficult.Whereassomeroutinecontrolplansusinginthermoregulationeffectsofbigpetrochemicalequipmentisnotidealenough,soitisverysignificanttodesignthenewcontroltacticsofthepetrochemicaltemperaturecontrolsystem.ThefuzzyPIDcontrol[3]isanamazingcontrolmethodinthedomainofcontrol,andusesthefuzzycontroltechnologyindependentofcontroltarget'smathematicalmodel.Throughtheprocessingoffuzzyinformation,thefuzzyPIDcontrolcanwellcontrolthecomplexobject,andhashighprecisionofstablestate.ThefuzzyPIDcontrolhasupperagility,adaptabilityandcontrolprecision,andobtainsmoreandmoreapplicationinmanydomains.AsthecontrolregulationparametersoffuzzyPIDcontrolleraresummedupviathe(1)InthedesignofpositiontypePIDcontroller,theincreasetypePIDisusuallyused.Thediscreteformis:u(t)=u(t-l)+Kp[e(t)-e(t-l)]+Kie(t)+Kd[e(t)-2e(t-l)+e(t-2)](2)WhereKp,K.,KIdaretheproportional,integral,anddifferentialcoefficientsrespectively.AndTissamplingcycle,u(t)isthecontrollingquantityofcurrentsamplingtime.Itissupposedthate(t)and6e(t)aretheoutputerroranderrorchangerate,thenadjustingPIDparametersusingfuzzyneuralnetworkbasedonconventionalPIDcontrolleristoestablishthecontinuousfunctionrelationshipbetweenthecontrolparametersKp,K.,KIdandoutputerrorabsolutevalue16e(t)I.ItissupposedIe(t)I,errorchangerateabsolutevaluethatthefunctionrelationshipbetweencontrolparametersanderror,errorchangerateistheformula(3).Accordingtodifferentinputoferroranderrorchangerate,PIDcontrolparameterscouldbecreatedreal-timelyonline,whichmakesthecontrollerperformancebest.-1940-Section2AutomationTechnology&RoboticsKp=.Ii(Ie(t)I,I~e(t)I)(3)K;=12(Ie(t)I,I~e(t)I)Kd=13(Ie(t)I,I~e(t)I)Accordingtotheexperts'experienceandactualsituation,thefollowingrulecouldbesummedup:whensystemerrore(t)isbig,inordertomakethesystemhasbetterfast-trackingperformanceandinordertoavoidthesystemresponseandlesserappearsbigovershooting[1],thebigKpKdshouldbeusedandtheintegralactionshouldbelimited.Figure1:structureofneuralnetworkfuzzyPIDcontrolsystem.Temperatureerrore(t)andtemperatureerrorchangerateae(t).InthefuzzyPIDcontrollerdesignedbasedonneuralnetwork,thereareseveninputlanguagevalues(NB,NM,NS,O,PS,PM,PB),whicharecorrespondingtouniverseofdiscourseoffuzzyset(-6,-5,-4,-3,-2,-1,0,1,2,3,4,5,6).Themembershipfunctionofeachfuzzylanguagevariableistanglefunction,whichisshowninfigure2andfigure3.Inordertoenhancetherobustnessofsystemandresolutionofmembershipfunction,theformoffunctionnearthevalueshouldbealittlesteep.Whenthesystemerrore(t)isinmediumsize,inordertomakethesystemappearlesserovershooting,Kpshouldbelesser.Inthiscase,thevalueofKdhastheinfluenceonsystemresponse.Whenthesystemerrore(t)islesser,inordertomakethesystemhavebetterstablestateperformance,KandK.shouldbebig.p1°3StructureandprincipleoffuzzyneuralnetworkPIDcontroller4.2NeuralnetworkadjustsfuzzyrulemoduleNeuralnetworkstructure[4]isshowninfigure4.ItusesthreelayersBPneuralnetworks,whichcompriseofinputlayer,concealinglayerandoutputlayer.Inputlayerise(t)andae(t)correspondingtotwoinputnodes,concealinglayertakingchargefuzzyandfuzzyreasoninghaseightnodes,outputlayerhasthreenodescorrespondingtothreecontrolparametersK,K.andK.Asthethreecontrolp1ThestructureofneuralnetworkfuzzyPIDcontrollerisshowninfigure1.Thecontrolleriscomposedofthreeparts:(1)thetraditionalPIDcontroller:whichcarriesclosedloopcontroloverthecontrolledobjectandprocessdirectly,andtheparametersKp,K.,Kareself-turningonline.(2)fuzzy1dmodule:whichcarriesthehandlingoffuzzinessandnormalizeddifferenceonstatevariable.(3)neuralnetwork:whichmeansfuzzyrule.ThroughstudyingNN,theruleisintheformofweightcoefficient,andthecreatingoftheruleistransformedtoconfirmandamendtheweightcoefficientinitialvalue.Accordingtotherunningstate,adjustingPIDcontrolparametersmakestheperformanceofthecontrolsystematitsbest.Theconcretemethodis:theoutputstateofneuralunitshouldbecorrespondingtotheadjustedparametersofPIDcontroller.ThePIDcontrolparametersaresettomakethecontrolsystembestviaadjustingthecontroller'weightcoefficient.dparameterscannotbenegative,theneuralunitactivationfunctionofoutputlayerisassumedasnon-negativeSigmoidfunction.Butactivation.functionofconcealinglayerneuralunitisassumedaspositiveandnegativesymmetricalSigmoidfunction.UNBr---_NIvINSO__PS__PIvI__-,PB10.80.60.44RealizingoffuzzyneuralnetworkPIDcontrolalgorithm4.1Filing-upfuzzymoduleItissupposedthattherearetwofuzzyvariables:0.20-6-4-2o246e(f)Figure2:membershipfunctionoftemperatureerrorchangee(t).Theinputofneuralnetworkise(t)andae(t),theinputandoutputofconcealinglayeris:2ne/)(k)=I(2)_OJ(k)-f[]=0Aj,.£..J(O~~)o(l)(k)lJ](2)._netj(k)]
