外文翻译--PLC变频调速的网络反馈系统的实现.doc
英文原文:RealizationofNeuralNetworkInverseSystemwithPLCinVariableFrequencySpeed-RegulatingSystemAbstract.Thevariablefrequencyspeed-regulatingsystemwhichconsistsofaninductionmotorandageneralinverter,andcontrolledbyPLCiswidelyusedinindustrialfield.However,forthemultivariable,nonlinearandstronglycoupledinductionmotor,thecontrolperformanceisnotgoodenoughtomeettheneedsofspeed-regulating.Themathematicmodelofthevariablefrequencyspeed-regulatingsysteminvectorcontrolmodeispresentedanditsreversibilityhasbeenproved.Byconstructinganeuralnetworkinversesystemandcombiningitwiththevariablefrequencyspeed-regulatingsystem,apseudo-linearsystemiscompleted,andthenalinearclose-loopisdesignedtogethighperformance.UsingPLC,aneuralnetworkinversesystemcanberealizedinsystem.Theresultsofexperimentshaveshownthattheperformancesofvariablefrequencyspeed-regulatingsystemcanbeimprovedgreatlyandthepracticabilityofneuralnetworkinversecontrolwastestified.1.IntroductionInrecentyears,withpowerelectronictechnology,microelectronictechnologyandmoderncontroltheoryinfiltratingintoACelectricdrivingsystem,invertershavebeenwidelyusedinspeed-regulatingofACmotor.Thevariablefrequencyspeed-regulatingsystemwhichconsistsofaninductionmotorandageneralinverterisusedtotaketheplaceofDCspeed-regulatingsystem.Becauseofterribleenvironmentandseveredisturbanceinindustrialfield,thechoiceofcontrollerisanimportantproblem.Inreference123,Neuralnetworkinversecontrolwasrealizedbyusingindustrialcontrolcomputerandseveraldataacquisitioncards.Theadvantagesofindustrialcontrolcomputerarehighcomputationspeed,greatmemorycapacityandgoodcompatibilitywithothersoftwareetc.Butindustrialcontrolcomputeralsohassomedisadvantagesinindustrialapplicationsuchasinstabilityandfallibilityandworsecommunicationability.PLCcontrolsystemisspecialdesignedforindustrialenvironmentapplication,anditsstabilityandreliabilityaregood.PLCcontrolsystemcanbeeasilyintegratedintofieldbuscontrolsystemwiththehighabilityofcommunicationconfiguration,soitiswildlyusedinrecentyears,anddeeplywelcomed.Sincethesystemcomposedofnormalinverterandinductionmotorisacomplicatednonlinearsystem,traditionalPIDcontrolstrategycouldnotmeettherequirementforfurthercontrol.Therefore,howtoenhancecontrolperformanceofthissystemisveryurgent.Theneuralnetworkinversesystem45isanovelcontrolmethodinrecentyears.Thebasicideaisthat:foragivensystem,aninversesystemoftheoriginalsystemiscreatedbyadynamicneuralnetwork,andthecombinationsystemofinverseandobjectistransformedintoakindofdecouplingstandardizedsystemwithlinearrelationship.Subsequently,alinearclose-loopregulatorcanbedesignedtoachievehighcontrolperformance.Theadvantageofthismethodiseasilytoberealizedinengineering.Thelinearizationanddecouplingcontrolofnormalsystemcanrealizeusingthismethod.CombiningtheneuralnetworkinverseintoPLCcaneasilymakeuptheinsufficiencyofsolvingtheproblemsofnonlinearandcouplinginPLCcontrolsystem.Thiscombinationcanpromotetheapplicationofneuralnetworkinverseintopracticetoachieveitsfulleconomic.Inthispaper,firstlytheneuralnetworkinversesystemmethodisintroduced,andmathematicmodelofthevariablefrequencyspeed-regulatingsysteminvectorcontrolmodeispresented.Thenareversibleanalysisofthesystemisperformed,andthemethodsandstepsaregiveninconstructingNN-inversesystemwithPLCcontrolsystem.Finally,themethodisverifiedintraditionalPIcontrolandNN-inversecontrol.2.NeuralNetworkInverseSystemControlMethodThebasicideaofinversecontrolmethod6isthat:foragivensystem,an-thintegralinversesystemoftheoriginalsystemiscreatedbyfeedbackmethod,andcombiningtheinversesystemwithoriginalsystem,akindofdecouplingstandardizedsystemwithlinearrelationshipisobtained,whichisnamedasapseudolinearsystemasshowninFig.1.Subsequently,alinearclose-loopregulatorwillbedesignedtoachievehighcontrolperformance.Inversesystemcontrolmethodwiththefeaturesofdirect,simpleandeasytounderstanddoesnotlikedifferentialgeometrymethod7,whichisdiscussestheproblemsin"geometrydomain".Themainproblemistheacquisitionoftheinversemodelintheapplications.Sincenon-linearsystemisacomplexsystem,anddesiredstrictinverseisverydifficulttoobtain,evenimpossible.Theengineeringapplicationofinversesystemcontroldontmeettheexpectations.Asneuralnetworkhasnon-linearapproximateability,especiallyfornonlinearthepowerfultooltosolvetheproblem.athNNinversesystemintegratedinversesystemwithnon-linearabilityoftheneuralnetworkcanavoidthetroublesofinversesystemmethod.Thenitispossibletoapplyinversecontrolmethodtoacomplicatednon-linearsystem.athNNinversesystemmethodneedslesssysteminformationsuchastherelativeorderofsystem,anditiseasytoobtaintheinversemodelbyneuralnetworktraining.CascadingtheNNinversesystemwiththeoriginalsystem,apseudo-linearsystemiscompleted.Subsequently,alinearclose-loopregulatorwillbedesigned.3.MathematicModelofInductionMotorVariableFrequencySpeed-RegulatingSystemandItsReversibilityInductionmotorvariablefrequencyspeed-regulatingsystemsuppliedbytheinverteroftrackingcurrentSPWMcanbeexpressedby5thordernonlinearmodelind-qtwo-phaserotatingcoordinate.Themodelwassimplifiedasa3-ordernonlinearmodel.Ifthedelayofinverterisneglected,themodelisexpressedasfollows:(1)wheredenotessynchronousanglefrequency,andisrotatespeed.arestatorscurrent,andarerotorsfluxlinkagein(d,q)axis.isnumberofpoles.ismutualinductance,andisrotorsinductance.Jismomentofinertia.isrotorstimeconstant,andisloadtorque.Invectormode,thenSubstituteditintoformula(1),then(2)Takingreversibilityanalysesofforum(2),thenThestatevariablesarechosenasfollowsInputvariablesareTakingthederivativeonoutputinformula(4),then(5)(6)ThentheJacobimatrixisRealizationofNeuralNetworkInverseSystemwithPLC(7)(8)Assoandsystemisreversible.Relative-orderofsystemisWhentheinverterisrunninginvectormode,thevariabilityoffluxlinkagecanbeneglected(consideringthefluxlinkagetobeinvariablenessandequaltotherating).Theoriginalsystemwassimplifiedasaninputandanoutputsystemconcludedbyforum(2).Accordingtoimplicitfunctionontologytheorem,inversesystemofformula(3)canbeexpressedas(9)Whentheinversesystemisconnectedtotheoriginalsysteminseries,thepseudolinearcompoundsystemcanbebuiltasthetypeof4.RealizationStepsofNeuralNetworkInverseSystem4.1AcquisitionoftheInputandOutputTrainingSamplesTrainingsamplesareextremelyimportantinthereconstructionofneuralnetworkinversesystem.Itisnotonlyneedtoobtainthedynamicdataoftheoriginalsystem,butalsoneedtoobtainthestaticdate.Referencesignalshouldincludealltheworkregionoforiginalsystem,whichcanbeensuretheapproximateability.Firstlythestepofactuatingsignalisgivencorrespondingevery10HZform0HZto50HZ,andtheresponsesofopenloopareobtain.Secondlyarandomtanglesignalisinput,whichisarandomsignalcascadingonthestepofactuatingsignalevery10seconds,andthecloseloopresponsesisobtained.Basedontheseinputs,1600groupstrainingsamplesaregotten.4.2TheConstructionofNeuralNetworkAstaticneuralnetworkandadynamicneuralnetworkcomposedofintegralisusedtoconstructtheinversesystem.Thestructureofstaticneuralnetworkis2neuronsininputlayer,3neuronsinoutputlayer,and12neuronsinhiddenlayer.Theexcitationfunctionofhiddenneuronismonotonicsmoothhyperbolictangentfunction.Theoutputlayeriscomposedofneuronwithlinearthresholdexcitationfunction.Thetrainingdatumarethecorrespondingspeedofopen-loop,close-loop,firstorderderivativeofthesespeed,andsettingreferencespeed.After50timestraining,thetrainingerrorof