IntelligentControlLecturer：王學(xué)泰GroupMembers：王學(xué)泰李媛張玉磊2011-04-021太原理工大學(xué)謝剛Dr.GangXie,TUT1INTRODUCTIONWithincreasingdemandsforhighprecisionautonomouscontroloverwideoperatingenvelopes,conventionalcontrol（傳統(tǒng)控制）

engineeringapproachesareunabletoadequatelydealwithsystemcomplexity（復(fù)雜性）,nonlinearities（非線性）,spatial（多維）andtemporalparametervariations（多參數(shù)）,andwithuncertainty（不確定）.

IntelligentControlorself-organising/learningcontrolisanewemergingdisciplinethatisdesignedtodealwithproblems.Ratherthanbeingmodelbased,itisexperientialbased（基于經(jīng)驗而不是模型）.

IntelligentControlistheamalgamofthedisciplinesofArtificialIntelligence人工智能,SystemsTheory系統(tǒng)論andOperationsResearch運籌學(xué)Intelligentcontroldescribesthedisciplinewherecontrolmethodsaredevelopedthatattempt

toemulate（模擬

）importantcharacteristicsofhumanintelligence.Thesecharacteristicsincludeadaptationandlearning,planningunderlargeuncertaintyandcopingwithlargeamountsof

data.CharacteristicsofIntelligentSystemsAdaptation自適應(yīng)andLearning再學(xué)習(xí):

Theabilitytoadapttochangingconditionsisecessaryinan

intelligentsystem.Althoughadaptationdoesnotnecessarilyrequiretheabilitytolearn,for

systemstobeabletoadapttoawidevarietyofunexpectedchangeslearningisessential.So

theabilitytolearnisanimportantcharacteristicof(highly)intelligentsystems.Autonomy自治性

andIntelligence智能性:

Autonomyinsettingandachievinggoalsisanimportan

characteristicofintelligentcontrolsystems.Whenasystemhastheabilitytoac

appropriatelyinanuncertainenvironmentforextendedperiodsoftimewithoutexternal

interventionitisconsideredtobehighlyautonomous.StructuresandHierarchies分級遞階:

Inordertocopewithcomplexity,anintelligentsystemmust

haveanappropriatefunctionalarchitectureorstructureforefficientanalysisandevaluation

ofcontrolstrategies.Fuzzy(logic)Control模糊（邏輯）控制ANN-BasedControl(ArtificialNeuralNetwork）基于人工神經(jīng)網(wǎng)絡(luò)的控制ExpertControl專家控制GeneticAlgorithm遺傳算法Intelligentcontrolcanbedividedintothefollowingmajorsub-domains:summary：Thecollectionofmethodologiescomprisingsoftcomputingincludee.g.fuzzylogic,neuralnetworks(neurocomputing)andgeneticalgorithms.Theseareoftencomplementaryratherthancompetingmethodologiesandcanoftenbecombinedinordertocreateintelligentsystems.Thestrengthsofthedifferentmethodologiesaresummarizedinthetablebelow.2FuzzyControlFuzzytheorybeganwithapaperon“fuzzysets（模糊集合）”,writtenbyProf.L.A.Zadehin1965.Fuzzysetsarethosesetswhoseboundaryisnotclear.Fuzzylogicsarecalculationproceduresonfuzzysets.Atechnologyinwhichthewholesystemcanberoughlydefined,thatis“fuzzytheory”wasproposed.Afuzzycontrolsystemisacontrolsystembasedonfuzzylogic（模糊邏輯）—amathematicalsystemthatanalyzesanaloginputvaluesintermsoflogicalvariablesthattakeoncontinuousvaluesbetween0and1,incontrasttoclassicalordigitallogic,whichoperatesondiscretevaluesofeither0or1(trueorfalse).2.1FuzzyLogicNormal“Crisp”logic

whereeverythingmustbeeither

TrueorFalse。Fuzzylogicisaformofmany-valuedlogicderivedfromfuzzysettheory（模糊集合理論）todealwithreasoningthatisfluidorapproximateFuzzylogicacknowledgesandexploitsthetoleranceforuncertaintyandimprecision.Thesentenceontheotherside

ofthelineisfalseThesentenceontheotherside

ofthelineisfalselinguisticvariables（語言變量）takeonlinguisticvalues（語言值）whicharewords(linguisticterms)withassociateddegreesof

membership（隸屬度）intheset。

2.2FuzzyControlDesign?INFORM1990-1998 Slide17Fuzzycontrolisamethodologytorepresentandimplementa(smart)human’sknowledgeabout

howtocontrolasystem.AfuzzycontrollerisshowninFigure1.?INFORM1990-1998 Slide18Thefuzzycontrollerhasseveral

components:?Therule-baseisasetofrulesabouthowtocontrol.?Fuzzi?cation（模糊化）istheprocessoftransformingthenumericinputsintoaformthatcanbeused

bytheinferencemechanism.?Theinferencemechanism（推理機）usesinformationaboutthecurrentinputs(formedbyfuzzi?cation),decideswhichrulesapplyinthecurrentsituation,andformsconclusionsaboutwhattheplant

inputshouldbe.?Defuzzi?cation（去模糊化）convertstheconclusionsreachedbytheinferencemechanismintoanumeric

inputfortheplant.MaptoFuzzySetsFuzzyRules

IFAANDBTHENL

*

*DefuzzificationInputsOutputget_inputs();fire_rules();find_output();TermDefinitions:Distance :={far,medium,close,zero,neg_close}Angle :={pos_big,pos_small,zero,neg_small,neg_big}Power :={pos_high,pos_medium,zero,neg_medium,neg_high}2.2.1Fuzzification:

-LinguisticVariables-?INFORM1990-1998 Slide21MembershipFunctionDefinition:TheLinguisticVariablesArethe“Vocabulary”ofaFuzzyLogicSystem!Computationofthe“IF-THEN”-Rules:#1:IFDistance=mediumANDAngle=pos_smallTHENPower=pos_medium#2:IFDistance=mediumANDAngle=zeroTHENPower=zero#3:IFDistance=farANDAngle=zeroTHENPower=pos_medium2.2.2Fuzzy-Inference:

-“IF-THEN”-Rules-?INFORM1990-1998 Slide22Aggregation: Computingthe“IF”-PartComposition: Computingthe“THEN”-PartTheRulesoftheFuzzyLogicSystemsArethe“Laws”ItExecutes!

2.2.3Defuzzification?INFORM1990-1998 Slide23

Defuzzificationistheprocessofproducingaquantifiableresultinfuzzylogic,givenfuzzysetsandcorrespondingmembershipdegrees。?INFORM1990-1998 Slide24summary：Fuzzycontrolistypicallyusedwhentheexplicitsystemanalyticalmodelis

notavailable.

Fuzzycontrolisintuitivetounderstandandeasytodesignfor

engineerswhoareunfamiliarwithclassicalcontroltheory.Afuzzycontroller

canbedesignedbasedone.g.ahumanoperatorsexperience.Fuzzycontrol

consistsofselectingandusing1.acollectionofrulesthatdescribethecontrolstrategy描述控制策略的規(guī)則集合2.membershipfunctionsforthelinguisticvariablesintherules在上述規(guī)則下，表示語言變量的隸屬度函數(shù)3.logicalconnectionsforfuzzyrelations模糊關(guān)系之間的邏輯關(guān)系4.adefuzzicationmethod

去模糊化的方法

3Arti?cialneuralnetwork?INFORM1990-1998 Slide253.1BiologicalapproachtoAIArti?cialneuralnetworksarecircuits,computeralgorithms（計算方法）,ormathematicalrepresentationsloosely

inspired（激勵）bythemassivelyconnectedsetofneuronsthatformbiologicalneuralnetworks.

Arti?cial

neuralnetworksareanalternativecomputingtechnologythathaveprovenusefulinavarietyof

patternrecognition（模式識別）,signalprocessing（信號處理）,estimation,andcontrolproblems.?INFORM1990-1998 Slide26?INFORM1990-1998 Slide27ArtificialneuronsNeuronsworkbyprocessinginformation.Theyreceiveandprovideinformationinformofspikes.?INFORM1990-1998 Slide28ArtificialneuralnetworksAnartificialneuralnetworkiscomposedofmanyartificialneuronsthatarelinkedtogetheraccordingtoaspecificnetworkarchitecture.Theobjectiveoftheneuralnetworkistotransformtheinputsintomeaningfuloutputs.?INFORM1990-1998 Slide293.2LearningProcessesLearning=learningbyadaptationTheyounganimallearnsthatthegreenfruitsaresour,whiletheyellowish/reddishonesaresweet.Thelearninghappensbyadaptingthefruitpickingbehavior.?INFORM1990-1998 Slide301.Learningwithateacher有教師學(xué)習(xí)

(Supervisedlearning)2.Learningwithoutateacher無教師學(xué)習(xí)（UnsupervisedLearning）3.ReinforcemengtLearning再勵學(xué)習(xí)?INFORM1990-1998 Slide31a.Learningwithaperceptron（感知機）Perceptron:Data:Error:Learning:?INFORM1990-1998 Slide32Perceptron:Data:Error:Learning:?INFORM1990-1998 Slide33b.LearningwithRBF（徑向基函數(shù)）neuralnetworksOnlythesynapticweightsoftheoutputneuronaremodified.AnRBFneuralnetworklearnsanonlinearfunction.?INFORM1990-1998 Slide34c.LearningwithMLP（多層感知機）neuralnetworks?INFORM1990-1998 Slide35Data:Error:xyout12…p-1p?INFORM1990-1998 Slide36d.Learningwithbackpropagation（BP反向傳播神經(jīng)網(wǎng)絡(luò)）Solutionofthecomplicatedlearning:calculatefirstthechangesforthesynapticweightsoftheoutputneuron;calculatethechangesbackwardstartingfromlayerp-1,andpropagatebackwardthelocalerrorterms.?INFORM1990-1998 Slide37e.LearningwithFeed-Forward(前饋)NNx1=x2=x3=Inputsandoutputsarenumeric.?INFORM1990-1998 Slide38

Learningtasksofartificialneuralnetworkscanbereformulatedasfunctionapproximation(函數(shù)逼近)tasks.

Neuralnetworkscanbeconsideredasnonlinearfunctionapproximating（非線性逼近）tools(i.e.,linearcombinationsofnonlinearbasisfunctions),wheretheparametersofthenetworksshouldbefoundbyapplyingoptimisationmethods（最優(yōu)化方法）.Theoptimisation（最優(yōu)化）isdonewithrespecttotheapproximationerrormeasure.

Ingeneralitisenoughtohaveasinglehiddenlayer（單隱含層）neuralnetwork(MLP,RBForother)tolearntheapproximationofanonlinearfunction.Insuchcasesgeneraloptimisationcanbeappliedtofindthechangerulesforthesynapticweights.3.3

Training（訓(xùn)練）NeuralNetworks?INFORM1990-1998 Slide39Howdoweconstructaneuralnetwork?Wetrainitwithexamples.Regardlessofthetypeof

network,wewillrefertoitaswhereθisthevectorofparametersthatwetunetoshapethenonlinearityitimplements(Fcould

beafuzzysystemtoointhediscussionbelow).Foraneuralnetworkθwouldbeavectorofthe

weightsandbiases.SometimeswewillcallFan“approximatorstructure.”Supposethatwegather

input-outputtrainingdatafromafunctiony=g(x)thatwedonothaveananalyticalexpression

for(e.g.,itcouldbeaphysicalprocess).?INFORM1990-1998 Slide40TrainingoftheANN(ArtificialNeuralNet)iseffectedby:Startingwithartibrarywieghts（任意權(quán)值）Presentingthedata,instancebyinstance

adaptingtheweightsaccordingtheerrorforeachinstance.Repeatinguntilconvergence（收斂）.?INFORM1990-1998 Slide41summaryANNisamassivelyparalleldistributedprocessor（分布式并行處理）.1.Simpleprocessingunitsthatcanstoreexperience2.Alearningprocess4ExpertControlSystemsAnexpertcontrolsystemisacomputerprogramthatisdesignedtoholdtheaccumulatedknowledge（累計知識）

ofoneormoredomainexperts4.1ComponentsofanExpertSystemTheknowledgebaseisthecollectionoffactsandruleswhichdescribealltheknowledgeabouttheproblemdomainTheinferenceengineisthepartofthesystemthatchooseswhichfactsandrulestoapplywhentryingtosolvetheuser’squeryTheuserinterfaceisthepartofthesystemwhichtakesintheuser’squeryinareadableformandpassesittotheinferenceengine.Itthendisplaystheresultstotheuser.InterpreterInferenceEngineKnowledgeBasedRulesDatabaseContextSetoffactsNaturalLanguageInterfaceExpertUserExpertSystemStructure（1）KnowledgeBase（知識庫）Representsallthedataandinformationimputedbyexpertsinthefield.Storesthedataasasetofrulesthatthesystemmustfollowtomakedecisions.KnowledgeAcquisitionExpertSystemKnowledgeEngineerHumanExpert

（2）InferenceEngine（推理機）Askstheuserquestionsaboutwhattheyarelookingfor.Appliestheknowledgeandtherulesheldintheknowledgebase.Appropriatelyusesthisinformationtoarriveatadecision.（3）UserInterface（人機界面）Allowstheexpertsystemandtheusertocommunicate.Findsoutwhatitisthatthesystemneedstoanswer.Sendstheuserquestionsoranswersandreceivestheirresponse.4.2CharacteristicswithExpertSystemsThereisnoexpertinthefieldTheexpertisunabletocommunicatehis/herideasTheexpertisunwillingtocommunicatehis/herideasTheexpertisnotavailableMusthaveallinformationonasubjectCanallthetestingbeaccomplished?UseracceptanceAdvantagesofExpertSystemsCanbesimpletouseEfficientresultsAccurateresultsAdaptationandadjustmentstochangingconditionsCosteffectiveProblemswithExpertSystemsLimiteddomainSystemsarenotalwaysuptodate,anddon’tlearnNo“commonsense”Expertsneededtosetupandmaintainsystem5GeneticAlgorithmsAnalgorithmisasetofinstructionsthatisrepeatedtosolveaproblem.Ageneticalgorithmconceptuallyfollowsstepsinspiredbythebiologicalprocessesofevolution.GeneticAlgorithmsfollowtheideaofSURVIVALOFTHEFITTEST-Better（適者生存）andbettersolutionsevolvefrompreviousgenerationsuntilanearoptimalsolutionisobtained.Alsoknownasevolutionaryalgorithms,geneticalgorithmsdemonstrateselforganizationandadaptationsimilartothewaythatthefittestbiologicalorganismsurviveandreproduce.Ageneticalgorithmisaniterativeprocedure（迭次求近）thatrepresentsitscandidatesolutionsasstringsofgenescalledchromosomes（染色體）.5.1Individual

Representing(個體表現(xiàn))

Anindividualisdatastructurerepresenting

the“geneticstructure（基因組合）”ofapossiblesolution.Geneticstructureconsistsofanalphabet(usually0,1)BinaryEncoding二進(jìn)制編碼MostCommon–stringofbits,0or1. Chrom:A=1011001011Chrom:B=1111110000GivesyoumanypossibilitiesExampleProblem:KnapsackproblemTheproblem:therearethingswithgivenvalueandsize.Theknapsackhasgivencapacity.Selectthingstomaximizethevalues.Encoding:Eachbitsays,ifthecorrespondingthingisintheknapsackPermutationEncoding排列編碼Usedin“orderingproblems”Everychromosomeisastringofnumbers,whichrepresentsnumberisasequence.ChromA:153264798ChromB:857723149Example:TravellingsalesmanproblemTheproblem:citiesthatmustbevisited.Encodingsaysorderofcitiesinwhichsalesmanwilllvisit.ValueEncoding值編碼Usedforcomplicatedvalues(realnumbers)andwhenbinarycodingwouldbedifficultEachchromosomeisastringofsomevalues.ChromA:1.23235.32430.4556ChromB:abcdjeifjdhdierjfdChromC:(back),(back),(right),(forward),(left)Example:Findingweightsforneuralnets.Theproblem:findweightsfornetworkEncoding:RealvaluesthatrepresentweightsRulebasesystem規(guī)則庫Givenarule(ifcolor=redandsize=smallandshape=roundthenobject=apple.Assumethateachfeaturehasfinitesetofvalues(e.g.,size=small,large)Representthevalueasasubstringoflengthequltothenumberofpossiblevalues.Forexample,small=10,large=01.Theentirerulewouldbe10010010100–setofrulesconcatenatingthevaluestogether.5.2SelectionCriteria（選擇標(biāo)準(zhǔn)）Fitnessproportionateselection,rankselectionmethods.

Fitnessproportionate（適應(yīng)度比例法）–eachindividual,I,hastheprobabilityfitness(I)/sum_over_all_individual_jFitness(j),whereFitness(I)isthefitnessfunctionvalueforindividualI.

Rankselection（

隨機抽樣法）–sortsindividualbyfitnessandtheprobabilitythatanindividualwillbeselectedisproportionaltoitsrankinthissortedlist.FitnessFunction（適應(yīng)度函數(shù)）Representsarankofthe“representation”Itisusuallyarealnumber.Thefunctionusuallyhasavaluebetween0and1.Onecanhaveasubjectivejudgment(e.g.1-5forrecipe2-1-4.)Similarlythelengthoftherouteinthetravelingsalespersonproblemisagoodmeasure,becausetheshortertheroute,thebetterthesolution.

5.3Reproduction（復(fù)制）Reproduction-Throughreproduction,geneticalgorithmsproducenewgenerationsofimprovedsolutionsbyselectingparentswithhigher

fitnessratings（適應(yīng)度）Crossover（交叉）-Manygeneticalgorithmsusestringsofbinarysymbolsforchromosomes,asinourKnapsackexample,torepresentsolutions.Crossovermeanschoosingarandompositioninthestring(say,after2digits)andexchangingthesegmentseithertotherightortotheleftofthispointwithanotherstringpartitionedsimilarlytoproducetwonewoffspring.CrossoverExampleParentA011011ParentB101100“bysplittingeachnumberasshownbetweenthesecondandthirddigits(positionisrandomlyselected)01*1011 10*1100Mutation（變異）-Mutationisanarbitrarychangeinasituation.Sometimesitisusedtopreventthealgorithmfromgettingstuck.Theprocedurechangesa1toa0,or0toa1.Thischangeoccurswithaverylowprobability(say1in1000)10101111100011Parent1Parent210100111100110Child1Child2MutationMutationandCrossoverCrossoverOperators

Single-pointcrossover:ParentA:10010|11101ParentB:01011|10110ChildAB:1001010110ChildBA:0101111101

Two-pointcrossover:ParentA:1001|011|101

ParentB:0101|110|110ChildAB:1001110101

ChildBA:0101011110UniformCrossoverandMutationUniformcrossover:ParentA:1001011101ParentB:0101110110ChildAB:1101111101

ChildBA:0001010110Mutation:randomlytoggleonebitIndividualA:1001011101IndividualA':1000011101Crossover–PermutationEncodingSinglepointcrossover-onecrossoverpointisselected,tillthispointthepermutationiscopiedfromthefirstparent,thenthesecondparentisscannedandifthenumberisnotyetintheoffspringitisadded

(123456789)+(453689721)=(123456897)Mutation

Orderchanging-twonumbersareselectedandexchanged(123456897)=>(18345629

7)

Crossover–ValueEncodingCrossover

AllcrossoversfrombinaryencodingcanbeusedMutation

Addingasmallnumber(forrealvalueencoding)-toselectedvaluesisadded(or