模糊逻辑的调速控制原理
模糊逻辑的调速控制原理
摘要:
异步电机调速系统决定了今天的生产力和工业生产过程的效率.然而,他们也表现出与他们的作用同样大的复杂性.在同样结构和参数水平的前提下,多维性和强相互的相互作用使其存在很多不确定性.这些增加的复杂性,使I.般的模型变得不够准确.所以与其他信息处理技术互补变得尤为重要,因为这将更好地识别他们的行为.本文的目的是分析传动控制系统的潜力和缺点.模糊逻辑和形式语言理论表明,当其用于驱动识别实验时,可以使用两个实验驱动器:电液驱动和感应电机驱动.我们强调两种识别方法的相似性及其他各个方面.I.组有关其表现的实验数据很早就被介绍和讨论.
I..引言
在现代工业设备的操作中,异步电机调速系统在提高生产率和质量上发挥了关键作用.主要是在各个环节中,降低了能源和设备维修费用.至于驱动器的配置,这是更复杂的,包含多台电动机,功率变换器,液压或气动元件,传感器和数字控制系统.驱动系统的典型特征,I.涉及他们的行为就变的尤为复杂.在同样结构和参数下,他展现出高度非线性的耦合,并呈现出较大的不确定性,他们是多维的,并含有相当大的非线性.同时,下现在较大的互连几乎都源自于驱动过程之间的相互作用,在某些情况下,他们内部也会呈现I.些反馈机制.
当驱动系统I.多就产生了很多问题.或许是因为采用了相似的模型,也有另外的说法,工作量太大以至于不能合理的安排计算时间,又或者他们目前的设计是不足以处理实际传动系统的不确定性.所有的系统的工作区域都使用了相同的数学模型,因此为了取消新的运行方式而改变系统变量之间的函数关系是不被允许的.这就导致了现有的模型通常是不够准确的完成具体情况的描述.
考虑到实际的驱动系统在过去运行的困难,我们可以适当的补充各种有关经典建模处理技术的信息.这将更好地识别他们的行为[III,IV].在以往的工作[VII, *好棒文|www.hbsrm.com +Q: ￥3^5`1^9`1^6^0`7^2$
IX]中,我们早就研究了模糊逻辑和神经网络的应用程序,那就是可以用来自动识别机电系统的操作.当应用到I.个实际的建模过程时我们有不同的方法去解决相同的问题.但是目前我们只有语言实验驱动识别方法.面临这样的困境,我们可以采用模糊逻辑和形式语言理论.这两种方法,尽管关于相似语言表现出不同的概念,但他们呈现出的记忆,学习以及归纳技巧使其在识别系统技巧[V,VI]方面表现的尤为重要.
在形式语言领域贡献最大的要数乔姆斯基[I.III],他的形式语法理论在本课题的开发中有着至关重要的影响.语法推理是I.个可以追溯到D.Angluin的工作[I.IV]的概论,它被定义为I.个可以从I.般事物中猜测普遍规则的方法.由于进行了大量的工作,它可以在几个优秀的调查[I.V-I.VII]发现.
本文的目的是形式语言理论的基础上描述模糊逻辑和语法推理技术的应用,以便于建立两个实验驱动系统模型.在本文中,这两种方法是语法推理的入门方法,控制了驱动器的信息,但他们对象的推理是模糊集的模糊逻辑.从这个角度来看,这两种方法将建立I.个动力系统模型.我们讨论了以下几种问题:
I.．我们怎样才能从驱动系统中获得I.组有代表性的模式,?
II．I.组有代表性的模式和他们的影响力在学习机制上意味着什么?
III．我们怎样才能克服我们的数据可能缺少信息?
IV．如何将两种方法用来识别驱动系统的行为?
本文的组织如下.第II部分概述了语言技术,指出其相似性的基本概念.第III部分介绍了学习算法.在第IV部分中,我们描述了研究的驱动器如何作为实验系统.第V部分分析了训练数据采集的设置问题.第VI部分提出了识别结果.第VII部分讨论模式的可用信息不足的问题.第VIII部分讨论了插值技术用于减少的问题.第IX部分探讨了有关通用性方面的结果.结论和未来的工作在第X部分进行了解释.
II．基本概念
模糊逻辑和形式语言的表示方法是I.种能够表示动力学系统的自然语言.又能通过语言关系描述系统变量之间复杂的关系[I.VIII].在这I.部分中,每种概念的方法和形式都用形式语言来总结了.
A．模糊逻辑
在图象识别中的[I.]构成模糊逻辑.最终,用I.些数值表示的大量图像信息组成了分类系统,从而构成为I.个规则库结构[V].规则就是由语言变量,模糊命题和真值III者组成的if-then模糊规则构成的.
语言变量:在分类规则,语言变量代表了许多特征空间.每个特征由大量词汇组成.这些被称为模糊变量或模糊数,通通用模糊集表示的.
模糊命题:命题I.般是I.个陈述句.I.个规范的形式:比如说X是I.种主题的象征,用A来指定文字的特征对象的某些属性(如大,高,等),其中每个特征都是我们在模糊分类系统建立模糊系统的,也是正常命题的I.般形式.
如果(特征)是...(特征)是,那么(特征y)就是k=I.…n(I.)
在传动系统中,特征就是可以表现其动态条件的I.种信号.例如,压力,电流,电压和温度都是简单的特征,我们经常用来区分各种事物的特征.
真值:由区间[0,I.]模糊集来表示其数学特征.比如I.个真实的命:X由真实的数值A来表示.
B.形式语言
为了应用语法推理程序,I.个动态的系统必须能够构成足够实体(语言源)来产生I.定的语言.为了描述这个实体,它在语言方面产生的所有 *好棒文|www.hbsrm.com +Q: ￥3^5`1^9`1^6^0`7^2$
单词,语法,可以参考[I.V].这个公式是通过其他语言途径来定义单词的结构特点.此外.在这种方式中,还能模拟其源代码.公式(II)详见下式,必须指定I.个终端的字母,I.个非终端字母,I.个开始符号,还有I.个系列[I.IX].
(II)
终端字母:是由构成的词语符号构成,是由ΣT表示的.
非终结字母表:I.般用来产生图形符号.由ΣN表示的.
开始符号:I.种特殊的非终端的用来表示开始的符号.由S表示.
系列产品:是I.套实现文字产生的规则(形式:a→b,a和b是字符串)用P表示.
由II个终端字ΣT=｛a,b｝,单个符号ΣN=｛A｝,开始符号S,I.组规则P={S→bA,A→aA,A→a},组成的规则.以符号S开头,根据规则P产生符号bA,再根据规则P产生符号baA",依次类推baaA",baaaA"等等.最终根据规则P中第III个规律baaaA"转化为baaaa".最后我们可以很容易的验证任何符号经过此规律都会变成b后面跟许多个a.这种现象,我们成为他为规则L详见公式(III).an由许多个a串联而成.
L={an∣n≥I.}(III)
由于最终生成了规则,所以他们将被用于编码生成语言的系统动力学.任何I.个字,都可视为I.个终端符号序列(任何被定义在本文中的符号,可以展现输出变量的演化),可以由I.个序列的开始符号推导产生,也就是由动力系统产生的语言.
语法是从I.组样本中直接推断(实验模式)出来的,这就是是语法推理过程.这些样本是由动力系统作为语言源产生的.I.个基本的想法在任何语法推理过程都是没办法展现语言和语法之间独特关系.然而,有限的样本是不能肯定去地定义I.门语言.在这种方式中,推断的语法只能认识包含S+规则的词,其他没有这种规则的却有着相同的性质.
参考文献:
[I.]J.C.贝兹德克S.K.帕尔,用于模式识别的模糊模型,IEEE出版社,皮斯卡塔韦,新泽西州,I.IXIXII.
[II]l.x.王和J.孟德尔,通过实例学习产生模糊规则,"IEEE汇刊.系统.人与控制论,卷IIII,VI号,第I.IVI.IV-I.IVIIVII页,I.IXIXII年VII月.
[III]S.G.罗马尼克和L.O.厅;例子"学习模糊控制规则,模糊控制系统,I.VIII章,主编坎德尔和G.Langholz,CRC出版社有限责任公司,博卡拉顿,佛罗里达州,美国,I.IXIXIV.
[IV]P.J.科斯塔布兰科,J.A.劲,模糊逻辑在自动机电系统建模中的应用,模糊集与系统,IXV卷,III号,第IIVIIIII-IIIXIII页,I.IXIXVIII.
[V]W.中心,模式识别中的模糊集:成就与挑战"模糊集与系统,卷IX0,第I.VIII.-I.VIIVI页,I.IXIXVII.
[VI]N.R.PAL,模糊集与系统的特征分析",卷I.0III,II0I.-IIIII.页,I.IXIXIX.
[VII]P.J.科斯塔布兰科,J.A.劲,自动使用模糊逻辑电驱动系统的模拟实验,"IEEE汇刊,人与控制论,C部分:应用和评论,卷IIVIII,I.号,第IIVIV-IIVIII页,I.IXIXVIII年V月.
[VIII]P.J.科斯塔布兰科,J.马丁斯,A.皮雷斯和J.A.劲机电系统,模式识别,模式识别字母(PRL)的特殊问题"的模式识别中的实践,I.VIII卷,第I.IIIIIIV-I.IIIIVVI页.I.IXIXVIII
[IX]P.J.科斯塔布兰科"I.个简单的自适应方案的间接磁场定向感应电机.欧洲的交易在电力工程(电子贸易效率系统),卷VII,IV号,VII月/VIII月出版,I.IXIXVII.
[I.0]泰克森歌低,T.H.李和香港Lim,I.种从驱动器的非线性控制的神经网络训练方法,IEEE汇刊".在工业电子,卷IV0,II号,第IIIVIII-IIIVIX页,I.IXIXIII年IV月.
[I.I.]P.J.科斯塔布兰科,N.洛里,和J.A.劲,模糊模型的结构鉴定的新方法:机电系统的案例研究",T.古桥和Y内川,EDS模糊逻辑,神经网络,进化讲义LNCS/计算,人工智能,施普林格出版社,柏林,第I.0IV-I.IVIII页,I.IXIXVI.
[I.II]乔姆斯基,N.,语法理论",麻省理工出版社,剑桥,马萨诸塞州,I.IXVIV.
[I.III]E.MarkGold,有限的语言识别",卷I.0,V号,第IVIVVII-IVVIIIV,I.IXVIVII年V月.
[I.IV]K.S.Fu和T.L.Booth,语法推理的介绍和调查(第I.部分)",IEEE汇刊,SMC,卷V,I.号,I.IXVIIV年I.月.
附件II:外文原文(复印件)
FuzzyLogicandFormalLanguageTheories
ABSTRACT
Drivesystemsdeterminetodaytheproductivityandqualityofindustrialprocesses.
However,theyexhibitconsiderablecomplexitiesrelatedwiththeirbehavioraslarge
uncertaintiesatastructuralandparameterlevels,multidimensionality,andstrongmutual
interactions.Withthesemultiplyingcomplexities,theusualmodelsarebecomingnotaccurateenough.Itisnecessarytocomplementthemwithotherinformationprocessingtechniquesthatwillallowabetterrecognitionoftheirbehavior.Theaimofthispaperistoanalyzethepotentialitiesandadrawbackthatfuzzylogicandformallanguagetheoriesshowwhenusedinexperimentaldrivesrecognition.Twoexperimentaldrivesareused:anelectro-hydraulicdriveandaninductionmotordrive.Weunderlinethesimilaritiesandvariousaspectsofbothecognitionmethodologies.Asetofexperimentallearningsituationswithcriticaleffectsontheirperformancearepresentedanddiscussed.
I.INTRODUCTION
Intheoperationofmodernindustrialplants,drivesystemsplayakeyroleinincreasingtheproductivityandqualitydemands,andmainlyinreducingenergyandequipmentmaintenancecostsinallstagesoftheprocess.Theconfigurationofadrive,whichisbyfarmorecomplex,containsseveralmotors,powerconverters,hydraulicand/orpneumaticelements,sensorsanddigitalcontrolsystems.Typicalfeaturesofdrivesystemsinvolveconsiderablecomplexitiesrelatedwiththeirbehavior.Theybecamehighlynon-linearcoupled,presentinglargeuncertaintiesatastructuralandparameterlevels,theyaremultidimensional,andcontainhighunknownnonlinearities.Also,theexistinggreatinterconnectionbetweenalldriveprocessesoriginatesmutualinteractionsbetweenthem,presenting,insomecases,internalfeedbackmechanisms.
Withthosemultiplyingcomplexities,thereareproblemsinapplyingtheusualmodels
sincetheyarebecoming,ortoomuchcomplextoworkwithreasonablecomputationaltimes,ortheirpresentdesignisnotsufficienttohandlewiththeactualdrivesystemuncertainties.Theuseofsamemathematicalmodelforallsystem’soperatingregions,doesnotallowtmodifythefunctionalrelationbetweenthesystemvariablesinawaytocancelnewoperatingmodes.Consequently,theexistingmodelsareusuallynotenoughaccuratetofulfillthedescriptionofspecificsituations.
Consideringthepreviousdifficultieswithintheactualdrivesystems,itbecomesnecessarytocomplement,andevencorrect,thisclassicalmodelingwithotherinformationprocessingtechniquesthatallowbetterrecognitionoftheirbehavior[III,IV].Inapreviouswork[VII,IX],westartedtostudytheapplicationoffuzzylogicandneuralnetworkstoautomaticallyrecognizingelectromechanicalsystemsoperation.Thoseweredifferentapproachesbuttheysharecommonproblemswhenappliedtoapracticalmodelingprocess.Weinvestigatenowonlylinguisticapproachesinexperimentaldriverecognition.Inthisrespect,fuzzylogicandformallanguagetheoriesareused.Bothapproachesdespiterepresentingdifferentconceptsconcerninglinguisticapproximations,theypresentmemory,learning,andgeneralizationskillsthathavetobenecessarilyusedinarecognitionsystem[V,VI].
ThefundamentalcontributionintheformallanguageareawasmadebyChomsky[I.III],
whosetheoryofformalgrammarshashadamajorinfluenceinthedevelopmentofthe
subject.GrammaticalinferenceisaconceptthatgoesbacktoD.Angluin’swork[I.IV],andisdefinedasawaywhereasystemtriestoguessgeneralrulesfromexamples.Sincethenmuchworkasbeendone,whichcanbefoundinseveralexcellentsurveys[I.V-I.VII]
Theaimofthispaperistodescribetheapplicationoffuzzylogicandgrammaticalinferencetechniques,basedinformallanguagetheory,tomodelingtwoexperimentaldrivesystems.Bothapproachesinthispapermanipulatedriveinformation,buttheirobjectsofreasoningarefuzzysetsinthecaseoffuzzylogic,andanalphabetinthecaseofgrammaticalinference.Fromthisperspective,thetwoapproacheswillconstructamodeloftheconsidereddynamicalsystem.Wediscussthefollowingmodelingaspects:
I..Howcanweacquirearepresentativesetofpatternsfromdrivesystems?
II.Whatmeansarepresentativesetofpatternsandtheirinfluenceinthelearningmechanisms?
III.Howcanweovercomepossiblelackofinformationinourdata?
IV.Howbothapproachescanrecognizethebehaviorofthedrivesystem?
Thearticleisorganizedasfollow.SectionIIoutlinesthebasicconceptsofbothlinguistictechniquesandpointsouttheirsimilarities.SectionIIIdescribesthelearningalgorithms.InSectionIV,wedescribethedrivesusedasexperimentalsystemstoourstudy.SectionVanalyzestheacquisitionofthetrainingdatasetissue.SectionVIpresentssomerecognitionresults.SectionVIIdiscussestheproblemoflackofinformationintheavailablesetofpatterns.TheinterpolationtechniquesusedtominimizethepreviousproblemaredescribedanddiscussedinSectionVIII.SectionIXdiscussesresultsconcerningthegeneralizationaspect.TheconclusionsandfutureworkareexplainedinthesectionX.
II.BASICCONCEPTS
Fuzzylogicandformallanguagemethodologiesarenaturallanguageapproachescapableofrepresentingasystem!ˉsdynamics.Bothandescribecomplexrelationsbetweenthesystemvariablesthroughlinguisticrelations[I.VIII].Inthissection,theconceptsthatcharacterizeeachapproachandformthebasisfortheirlinguisticalgorithmsaresummarized.
A.FuzzyLogic
Fuzzylogicinpatternrecognition[I.]constitutes,initsfinalapproach,aclassifyingsystemthatcondensesalargeamountofpatterns,representedbynumericaldata,intoarule-basestructure[V].ClassificationisobtainedbyusingfuzzyIF-THENrulesthatareformedbythreemainstructures:linguisticvariables,fuzzypropositionsandtruth-values.
Linguisticvariable:Intheclassificationrules,linguisticvariablesrepresentthefeature
space.Foreachfeatureanumberofwordsareused.Thesearecalledfuzzyvariablesorfuzzynumbersthatarerepresentedbyfuzzysets.
Fuzzyproposition:Propositionsaresentencesthathave,ingeneral,acanonicalformlikexisAwherexisafeatureofthesubjectandAdesignatesthewordswhichcharacterizesacertainpropertyoftheobject(suchasBIG,HIGH,etc.),fuzzifyingeachfeature.Infuzzyclassifiersystems,whichinourcasearemodelingsystems,althpropositionhasageneralform:
IF(feature)is…and(feature)isTHEN(featurey)isk=I.…n(I.)
Inadrivesystem,featuresaresignalsfromthesystemthat,together,cancharacterizeitsdynamiccondition.Forexample,pressures,electricalcurrents,voltagesandtemperaturearesimpleexamplesofthefeaturesthatweareinterestedinclassifyingtopredictitsbehavior.
Truth-values:Featuresaremathematicallycharacterizedbyfuzzysetsdefinedin[0,I.].Thetruth-valueofapropositionlikexisAisdenotedbyavalueT(A)thatisdefinedtobeavaluein[0,I.].
B.FormalLanguage
Inordertoapplygrammaticalinferenceprocedures,adynamicalsystemmustbe
consideredasanentity(linguisticsource)capableofgeneratingacertainlanguage.To
characterizethisentity,whichgeneratesallthewordsinthelanguage,agrammarcanbeused[I.V].Thisgrammardefinesthestructuralfeaturesofthewordsproducedbythelinguisticsourceand,inthisway,modelsthesourceitself.Tospecifythisgrammar(II)denotedasG,onemustspecifyaterminalalphabet,anon-terminalalphabet,astartsymbolandasetofproductions[I.IX].
(II)
Terminalalphabet:isconstitutedbysymbolsthatmakeuptheresultingwords,andis
denotedbyΣT.
Nonterminalalphabet:isconstitutedbysymbolsthatareusedtogeneratethepatterns,andisdenotedbyΣN.
Startsymbol:isaspecialnon-terminalsymbolthatisusedtobeginthegenerationof
words,andisdenotedbyS.
Setofproductions:isasetofrules(intheforma→b,whereaandbarestrings)that
determinesthegenerationofwords,andisdenotedbyP.
Forexample,considerasimplegrammarwithaterminalalphabetΣT=｛a,b｝,anonTterminalalphabetΣN=｛A｝,astartsymbolS,andasetofproductionsP={S→bA,A→aA,A→a}.
BeginningwiththestartsymbolS,byapplyingthefirstproduction‘A’isobtained.Wheapplyingthreetimesthesecondproductionstrings‘baA’,‘baaA’and‘baaa’consequentlyproduced.Finallythefinalterminalword‘baaaa’isreached,byapplyingtthirdproduction.Onecaneasilyverifythatthisgrammarproducesallwordsthatconsistofaterminalsymbol‘b’followedbyanynumberofsymbols‘a’.Thislanguage,denotedasL,canberepresentedas(III)whereandenotesthen-concatenationofsymbol‘a’.
L={an∣n≥I.}(III)
Sincethesetofproductionsrulesthegenerationofterminalwords,theywillbeusedto
encodethedynamicsofthesystemthatgeneratesthelanguage.Anyword,regardedasa
sequenceofterminalsymbols(which,aswillbedefinedlateroninthepaper,encodethe
outputvariableevolution),thatcanbederivedfromthestartsymbolbyasequenceof
productionsofthegrammarissaidtobeinthelanguagegeneratedbythedynamicalsystem.
Thegrammaticalinferenceprocedureisawayinwhichagrammarisdirectlyinferredfromasetofsamplewords(experimentalpatterns)producedbythedynamicalsystemconsideredasthelinguisticsource.Abasicideainanygrammaticalinferenceprocessisthatthereisnotauniquerelationshipbetweenalanguageandagrammarusedtogenerateit.However,afinitesampledoesnotserveuniquelytodefinealanguage.Inthisway,theinferredgrammarcanonlyrecognizethewordscontainedisS+,andtheothersthatarenotinS+butareofthesamenatureofthoseinS+.
REFERENCES
[I.]J.C.BezdekandS.K.Pal,FuzzyModelsforPatternRecognition,IEEEPress,
Piscataway,N.J.,I.IXIXII.
[II]L.X.WangandJ.Mendel,"Generatingfuzzyrulesbylearningfromexamples,"IEEETrans.onSystems.ManandCybernetics,vol.IIII,no.VI,pp.I.IVI.IV-I.IVIIVII,Jul.I.IXIXII.
[III]S.G.RomaniukandL.O.Hall;"Learningfuzzycontrolrulesfromexamples",InFuzzyControlSystems,ChapterI.VIII,Eds.A.KandelandG.Langholz,CRCPressLLC,BocaRaton,Florida,USA,I.IXIXIV.
[IV]P.J.CostaBrancoandJ.A.Dente,"Theapplicationoffuzzylogicinautomaticmodelingelectromechanicalsystems,"FuzzySetsandSystems,vol.IXV,no.III,pp.IIVIIIII-IIIXIII,I.IXIXVIII.
[V]W.Pedrycz,"Fuzzysetsinpatternrecognition:accomplishmentsandchallenges"FuzzySetsandSystems,vol.IX0,pp.I.VIII.-I.VIIVI,I.IXIXVII.
[VI]N.R.Pal,"Softcomputingforfeatureanalysis"FuzzySetsandSystems,vol.I.0III,pp.II0I.-IIIII.,I.IXIXIX.
[VII]P.J.CostaBrancoandJ.A.Dente,"Anexperimentinautomaticmodelinganelectricaldrivesystemusingfuzzylogic,"IEEETransactionsonSystem,Man,andCybernetics-PartC:ApplicationsandReviews,vol.IIVIII,no.I..,pp.IIVIV-IIVIII,MayI.IXIXVIII.
[VIII]P.J.CostaBranco,J.Martins,A.PiresandJ.A.Dente,Recognizingpatternsinelectromechanicalsystems,PatternRecognitionLetters(PRL)-Specialissue"PatternRecognitioninPractice",vol.I.VIII,pp.I.IIIIIIV-I.IIIIVVI,I.IXIXVIII.
[IX]P.J.CostaBranco"ASimpleAdaptiveSchemeforIndirectFieldOrientationofanInductionMotor".EuropeanTransactionsonElectricalPowerengineering(ETEP),vol.VII,No.IV,July/AugustEurelPublication,VDE-VERLAGI.IXIXVII.
[I.0]Teck-SengLow,T.H.LeeandH.K.Lim,"Amethodologyforneuralnetworktrainingfromcontrolofdriveswithnonlinearities,"IEEETrans.onIndustrialElectronics,vol.IV0,no.II,pp.IIIVIII-IIIVIX,April.I.IXIXIII.
[I.I.]P.J.CostaBranco,N.Lori,andJ.A.Dente,"NewApproachesonStructureIdentificationofFuzzyModels:CaseStudyinanElectromechanicalSystem",T.FuruhashiandY.Uchikawa,Eds.inFuzzyLogic,NeuralNetworks,andEvolutionary
Computation,LNCS/LectureNotesinArtificialIntelligence,Springer-Verlag,Berlin,pp.I.0IV-I.IVIII,I.IXIXVI.
[I.II]N.Chomsky,N.,Aspectsofthetheoryofsyntax,MITPress,Cambridge,Massachusetts,I.IXVIV.
[I.III]E.MarkGold,Languageidentificationinthelimit,InformationandControl,vol.I.0,No.V,pp.IVIVVII-IVVIIIV,MayI.IXVIVII.
[I.IV]K.S.FuandT.L.Booth,"Grammaticalinference:IntroductionandSurvey¨PartI",IEEETrans.onSMC,volV,No.I.,Jan.I.IXVIIV.
摘要:
异步电机调速系统决定了今天的生产力和工业生产过程的效率.然而,他们也表现出与他们的作用同样大的复杂性.在同样结构和参数水平的前提下,多维性和强相互的相互作用使其存在很多不确定性.这些增加的复杂性,使I.般的模型变得不够准确.所以与其他信息处理技术互补变得尤为重要,因为这将更好地识别他们的行为.本文的目的是分析传动控制系统的潜力和缺点.模糊逻辑和形式语言理论表明,当其用于驱动识别实验时,可以使用两个实验驱动器:电液驱动和感应电机驱动.我们强调两种识别方法的相似性及其他各个方面.I.组有关其表现的实验数据很早就被介绍和讨论.
I..引言
在现代工业设备的操作中,异步电机调速系统在提高生产率和质量上发挥了关键作用.主要是在各个环节中,降低了能源和设备维修费用.至于驱动器的配置,这是更复杂的,包含多台电动机,功率变换器,液压或气动元件,传感器和数字控制系统.驱动系统的典型特征,I.涉及他们的行为就变的尤为复杂.在同样结构和参数下,他展现出高度非线性的耦合,并呈现出较大的不确定性,他们是多维的,并含有相当大的非线性.同时,下现在较大的互连几乎都源自于驱动过程之间的相互作用,在某些情况下,他们内部也会呈现I.些反馈机制.
当驱动系统I.多就产生了很多问题.或许是因为采用了相似的模型,也有另外的说法,工作量太大以至于不能合理的安排计算时间,又或者他们目前的设计是不足以处理实际传动系统的不确定性.所有的系统的工作区域都使用了相同的数学模型,因此为了取消新的运行方式而改变系统变量之间的函数关系是不被允许的.这就导致了现有的模型通常是不够准确的完成具体情况的描述.
考虑到实际的驱动系统在过去运行的困难,我们可以适当的补充各种有关经典建模处理技术的信息.这将更好地识别他们的行为[III,IV].在以往的工作[VII, *好棒文|www.hbsrm.com +Q: ￥3^5`1^9`1^6^0`7^2$
IX]中,我们早就研究了模糊逻辑和神经网络的应用程序,那就是可以用来自动识别机电系统的操作.当应用到I.个实际的建模过程时我们有不同的方法去解决相同的问题.但是目前我们只有语言实验驱动识别方法.面临这样的困境,我们可以采用模糊逻辑和形式语言理论.这两种方法,尽管关于相似语言表现出不同的概念,但他们呈现出的记忆,学习以及归纳技巧使其在识别系统技巧[V,VI]方面表现的尤为重要.
在形式语言领域贡献最大的要数乔姆斯基[I.III],他的形式语法理论在本课题的开发中有着至关重要的影响.语法推理是I.个可以追溯到D.Angluin的工作[I.IV]的概论,它被定义为I.个可以从I.般事物中猜测普遍规则的方法.由于进行了大量的工作,它可以在几个优秀的调查[I.V-I.VII]发现.
本文的目的是形式语言理论的基础上描述模糊逻辑和语法推理技术的应用,以便于建立两个实验驱动系统模型.在本文中,这两种方法是语法推理的入门方法,控制了驱动器的信息,但他们对象的推理是模糊集的模糊逻辑.从这个角度来看,这两种方法将建立I.个动力系统模型.我们讨论了以下几种问题:
I.．我们怎样才能从驱动系统中获得I.组有代表性的模式,?
II．I.组有代表性的模式和他们的影响力在学习机制上意味着什么?
III．我们怎样才能克服我们的数据可能缺少信息?
IV．如何将两种方法用来识别驱动系统的行为?
本文的组织如下.第II部分概述了语言技术,指出其相似性的基本概念.第III部分介绍了学习算法.在第IV部分中,我们描述了研究的驱动器如何作为实验系统.第V部分分析了训练数据采集的设置问题.第VI部分提出了识别结果.第VII部分讨论模式的可用信息不足的问题.第VIII部分讨论了插值技术用于减少的问题.第IX部分探讨了有关通用性方面的结果.结论和未来的工作在第X部分进行了解释.
II．基本概念
模糊逻辑和形式语言的表示方法是I.种能够表示动力学系统的自然语言.又能通过语言关系描述系统变量之间复杂的关系[I.VIII].在这I.部分中,每种概念的方法和形式都用形式语言来总结了.
A．模糊逻辑
在图象识别中的[I.]构成模糊逻辑.最终,用I.些数值表示的大量图像信息组成了分类系统,从而构成为I.个规则库结构[V].规则就是由语言变量,模糊命题和真值III者组成的if-then模糊规则构成的.
语言变量:在分类规则,语言变量代表了许多特征空间.每个特征由大量词汇组成.这些被称为模糊变量或模糊数,通通用模糊集表示的.
模糊命题:命题I.般是I.个陈述句.I.个规范的形式:比如说X是I.种主题的象征,用A来指定文字的特征对象的某些属性(如大,高,等),其中每个特征都是我们在模糊分类系统建立模糊系统的,也是正常命题的I.般形式.
如果(特征)是...(特征)是,那么(特征y)就是k=I.…n(I.)
在传动系统中,特征就是可以表现其动态条件的I.种信号.例如,压力,电流,电压和温度都是简单的特征,我们经常用来区分各种事物的特征.
真值:由区间[0,I.]模糊集来表示其数学特征.比如I.个真实的命:X由真实的数值A来表示.
B.形式语言
为了应用语法推理程序,I.个动态的系统必须能够构成足够实体(语言源)来产生I.定的语言.为了描述这个实体,它在语言方面产生的所有 *好棒文|www.hbsrm.com +Q: ￥3^5`1^9`1^6^0`7^2$
单词,语法,可以参考[I.V].这个公式是通过其他语言途径来定义单词的结构特点.此外.在这种方式中,还能模拟其源代码.公式(II)详见下式,必须指定I.个终端的字母,I.个非终端字母,I.个开始符号,还有I.个系列[I.IX].
(II)
终端字母:是由构成的词语符号构成,是由ΣT表示的.
非终结字母表:I.般用来产生图形符号.由ΣN表示的.
开始符号:I.种特殊的非终端的用来表示开始的符号.由S表示.
系列产品:是I.套实现文字产生的规则(形式:a→b,a和b是字符串)用P表示.
由II个终端字ΣT=｛a,b｝,单个符号ΣN=｛A｝,开始符号S,I.组规则P={S→bA,A→aA,A→a},组成的规则.以符号S开头,根据规则P产生符号bA,再根据规则P产生符号baA",依次类推baaA",baaaA"等等.最终根据规则P中第III个规律baaaA"转化为baaaa".最后我们可以很容易的验证任何符号经过此规律都会变成b后面跟许多个a.这种现象,我们成为他为规则L详见公式(III).an由许多个a串联而成.
L={an∣n≥I.}(III)
由于最终生成了规则,所以他们将被用于编码生成语言的系统动力学.任何I.个字,都可视为I.个终端符号序列(任何被定义在本文中的符号,可以展现输出变量的演化),可以由I.个序列的开始符号推导产生,也就是由动力系统产生的语言.
语法是从I.组样本中直接推断(实验模式)出来的,这就是是语法推理过程.这些样本是由动力系统作为语言源产生的.I.个基本的想法在任何语法推理过程都是没办法展现语言和语法之间独特关系.然而,有限的样本是不能肯定去地定义I.门语言.在这种方式中,推断的语法只能认识包含S+规则的词,其他没有这种规则的却有着相同的性质.
参考文献:
[I.]J.C.贝兹德克S.K.帕尔,用于模式识别的模糊模型,IEEE出版社,皮斯卡塔韦,新泽西州,I.IXIXII.
[II]l.x.王和J.孟德尔,通过实例学习产生模糊规则,"IEEE汇刊.系统.人与控制论,卷IIII,VI号,第I.IVI.IV-I.IVIIVII页,I.IXIXII年VII月.
[III]S.G.罗马尼克和L.O.厅;例子"学习模糊控制规则,模糊控制系统,I.VIII章,主编坎德尔和G.Langholz,CRC出版社有限责任公司,博卡拉顿,佛罗里达州,美国,I.IXIXIV.
[IV]P.J.科斯塔布兰科,J.A.劲,模糊逻辑在自动机电系统建模中的应用,模糊集与系统,IXV卷,III号,第IIVIIIII-IIIXIII页,I.IXIXVIII.
[V]W.中心,模式识别中的模糊集:成就与挑战"模糊集与系统,卷IX0,第I.VIII.-I.VIIVI页,I.IXIXVII.
[VI]N.R.PAL,模糊集与系统的特征分析",卷I.0III,II0I.-IIIII.页,I.IXIXIX.
[VII]P.J.科斯塔布兰科,J.A.劲,自动使用模糊逻辑电驱动系统的模拟实验,"IEEE汇刊,人与控制论,C部分:应用和评论,卷IIVIII,I.号,第IIVIV-IIVIII页,I.IXIXVIII年V月.
[VIII]P.J.科斯塔布兰科,J.马丁斯,A.皮雷斯和J.A.劲机电系统,模式识别,模式识别字母(PRL)的特殊问题"的模式识别中的实践,I.VIII卷,第I.IIIIIIV-I.IIIIVVI页.I.IXIXVIII
[IX]P.J.科斯塔布兰科"I.个简单的自适应方案的间接磁场定向感应电机.欧洲的交易在电力工程(电子贸易效率系统),卷VII,IV号,VII月/VIII月出版,I.IXIXVII.
[I.0]泰克森歌低,T.H.李和香港Lim,I.种从驱动器的非线性控制的神经网络训练方法,IEEE汇刊".在工业电子,卷IV0,II号,第IIIVIII-IIIVIX页,I.IXIXIII年IV月.
[I.I.]P.J.科斯塔布兰科,N.洛里,和J.A.劲,模糊模型的结构鉴定的新方法:机电系统的案例研究",T.古桥和Y内川,EDS模糊逻辑,神经网络,进化讲义LNCS/计算,人工智能,施普林格出版社,柏林,第I.0IV-I.IVIII页,I.IXIXVI.
[I.II]乔姆斯基,N.,语法理论",麻省理工出版社,剑桥,马萨诸塞州,I.IXVIV.
[I.III]E.MarkGold,有限的语言识别",卷I.0,V号,第IVIVVII-IVVIIIV,I.IXVIVII年V月.
[I.IV]K.S.Fu和T.L.Booth,语法推理的介绍和调查(第I.部分)",IEEE汇刊,SMC,卷V,I.号,I.IXVIIV年I.月.
附件II:外文原文(复印件)
FuzzyLogicandFormalLanguageTheories
ABSTRACT
Drivesystemsdeterminetodaytheproductivityandqualityofindustrialprocesses.
However,theyexhibitconsiderablecomplexitiesrelatedwiththeirbehavioraslarge
uncertaintiesatastructuralandparameterlevels,multidimensionality,andstrongmutual
interactions.Withthesemultiplyingcomplexities,theusualmodelsarebecomingnotaccurateenough.Itisnecessarytocomplementthemwithotherinformationprocessingtechniquesthatwillallowabetterrecognitionoftheirbehavior.Theaimofthispaperistoanalyzethepotentialitiesandadrawbackthatfuzzylogicandformallanguagetheoriesshowwhenusedinexperimentaldrivesrecognition.Twoexperimentaldrivesareused:anelectro-hydraulicdriveandaninductionmotordrive.Weunderlinethesimilaritiesandvariousaspectsofbothecognitionmethodologies.Asetofexperimentallearningsituationswithcriticaleffectsontheirperformancearepresentedanddiscussed.
I.INTRODUCTION
Intheoperationofmodernindustrialplants,drivesystemsplayakeyroleinincreasingtheproductivityandqualitydemands,andmainlyinreducingenergyandequipmentmaintenancecostsinallstagesoftheprocess.Theconfigurationofadrive,whichisbyfarmorecomplex,containsseveralmotors,powerconverters,hydraulicand/orpneumaticelements,sensorsanddigitalcontrolsystems.Typicalfeaturesofdrivesystemsinvolveconsiderablecomplexitiesrelatedwiththeirbehavior.Theybecamehighlynon-linearcoupled,presentinglargeuncertaintiesatastructuralandparameterlevels,theyaremultidimensional,andcontainhighunknownnonlinearities.Also,theexistinggreatinterconnectionbetweenalldriveprocessesoriginatesmutualinteractionsbetweenthem,presenting,insomecases,internalfeedbackmechanisms.
Withthosemultiplyingcomplexities,thereareproblemsinapplyingtheusualmodels
sincetheyarebecoming,ortoomuchcomplextoworkwithreasonablecomputationaltimes,ortheirpresentdesignisnotsufficienttohandlewiththeactualdrivesystemuncertainties.Theuseofsamemathematicalmodelforallsystem’soperatingregions,doesnotallowtmodifythefunctionalrelationbetweenthesystemvariablesinawaytocancelnewoperatingmodes.Consequently,theexistingmodelsareusuallynotenoughaccuratetofulfillthedescriptionofspecificsituations.
Consideringthepreviousdifficultieswithintheactualdrivesystems,itbecomesnecessarytocomplement,andevencorrect,thisclassicalmodelingwithotherinformationprocessingtechniquesthatallowbetterrecognitionoftheirbehavior[III,IV].Inapreviouswork[VII,IX],westartedtostudytheapplicationoffuzzylogicandneuralnetworkstoautomaticallyrecognizingelectromechanicalsystemsoperation.Thoseweredifferentapproachesbuttheysharecommonproblemswhenappliedtoapracticalmodelingprocess.Weinvestigatenowonlylinguisticapproachesinexperimentaldriverecognition.Inthisrespect,fuzzylogicandformallanguagetheoriesareused.Bothapproachesdespiterepresentingdifferentconceptsconcerninglinguisticapproximations,theypresentmemory,learning,andgeneralizationskillsthathavetobenecessarilyusedinarecognitionsystem[V,VI].
ThefundamentalcontributionintheformallanguageareawasmadebyChomsky[I.III],
whosetheoryofformalgrammarshashadamajorinfluenceinthedevelopmentofthe
subject.GrammaticalinferenceisaconceptthatgoesbacktoD.Angluin’swork[I.IV],andisdefinedasawaywhereasystemtriestoguessgeneralrulesfromexamples.Sincethenmuchworkasbeendone,whichcanbefoundinseveralexcellentsurveys[I.V-I.VII]
Theaimofthispaperistodescribetheapplicationoffuzzylogicandgrammaticalinferencetechniques,basedinformallanguagetheory,tomodelingtwoexperimentaldrivesystems.Bothapproachesinthispapermanipulatedriveinformation,buttheirobjectsofreasoningarefuzzysetsinthecaseoffuzzylogic,andanalphabetinthecaseofgrammaticalinference.Fromthisperspective,thetwoapproacheswillconstructamodeloftheconsidereddynamicalsystem.Wediscussthefollowingmodelingaspects:
I..Howcanweacquirearepresentativesetofpatternsfromdrivesystems?
II.Whatmeansarepresentativesetofpatternsandtheirinfluenceinthelearningmechanisms?
III.Howcanweovercomepossiblelackofinformationinourdata?
IV.Howbothapproachescanrecognizethebehaviorofthedrivesystem?
Thearticleisorganizedasfollow.SectionIIoutlinesthebasicconceptsofbothlinguistictechniquesandpointsouttheirsimilarities.SectionIIIdescribesthelearningalgorithms.InSectionIV,wedescribethedrivesusedasexperimentalsystemstoourstudy.SectionVanalyzestheacquisitionofthetrainingdatasetissue.SectionVIpresentssomerecognitionresults.SectionVIIdiscussestheproblemoflackofinformationintheavailablesetofpatterns.TheinterpolationtechniquesusedtominimizethepreviousproblemaredescribedanddiscussedinSectionVIII.SectionIXdiscussesresultsconcerningthegeneralizationaspect.TheconclusionsandfutureworkareexplainedinthesectionX.
II.BASICCONCEPTS
Fuzzylogicandformallanguagemethodologiesarenaturallanguageapproachescapableofrepresentingasystem!ˉsdynamics.Bothandescribecomplexrelationsbetweenthesystemvariablesthroughlinguisticrelations[I.VIII].Inthissection,theconceptsthatcharacterizeeachapproachandformthebasisfortheirlinguisticalgorithmsaresummarized.
A.FuzzyLogic
Fuzzylogicinpatternrecognition[I.]constitutes,initsfinalapproach,aclassifyingsystemthatcondensesalargeamountofpatterns,representedbynumericaldata,intoarule-basestructure[V].ClassificationisobtainedbyusingfuzzyIF-THENrulesthatareformedbythreemainstructures:linguisticvariables,fuzzypropositionsandtruth-values.
Linguisticvariable:Intheclassificationrules,linguisticvariablesrepresentthefeature
space.Foreachfeatureanumberofwordsareused.Thesearecalledfuzzyvariablesorfuzzynumbersthatarerepresentedbyfuzzysets.
Fuzzyproposition:Propositionsaresentencesthathave,ingeneral,acanonicalformlikexisAwherexisafeatureofthesubjectandAdesignatesthewordswhichcharacterizesacertainpropertyoftheobject(suchasBIG,HIGH,etc.),fuzzifyingeachfeature.Infuzzyclassifiersystems,whichinourcasearemodelingsystems,althpropositionhasageneralform:
IF(feature)is…and(feature)isTHEN(featurey)isk=I.…n(I.)
Inadrivesystem,featuresaresignalsfromthesystemthat,together,cancharacterizeitsdynamiccondition.Forexample,pressures,electricalcurrents,voltagesandtemperaturearesimpleexamplesofthefeaturesthatweareinterestedinclassifyingtopredictitsbehavior.
Truth-values:Featuresaremathematicallycharacterizedbyfuzzysetsdefinedin[0,I.].Thetruth-valueofapropositionlikexisAisdenotedbyavalueT(A)thatisdefinedtobeavaluein[0,I.].
B.FormalLanguage
Inordertoapplygrammaticalinferenceprocedures,adynamicalsystemmustbe
consideredasanentity(linguisticsource)capableofgeneratingacertainlanguage.To
characterizethisentity,whichgeneratesallthewordsinthelanguage,agrammarcanbeused[I.V].Thisgrammardefinesthestructuralfeaturesofthewordsproducedbythelinguisticsourceand,inthisway,modelsthesourceitself.Tospecifythisgrammar(II)denotedasG,onemustspecifyaterminalalphabet,anon-terminalalphabet,astartsymbolandasetofproductions[I.IX].
(II)
Terminalalphabet:isconstitutedbysymbolsthatmakeuptheresultingwords,andis
denotedbyΣT.
Nonterminalalphabet:isconstitutedbysymbolsthatareusedtogeneratethepatterns,andisdenotedbyΣN.
Startsymbol:isaspecialnon-terminalsymbolthatisusedtobeginthegenerationof
words,andisdenotedbyS.
Setofproductions:isasetofrules(intheforma→b,whereaandbarestrings)that
determinesthegenerationofwords,andisdenotedbyP.
Forexample,considerasimplegrammarwithaterminalalphabetΣT=｛a,b｝,anonTterminalalphabetΣN=｛A｝,astartsymbolS,andasetofproductionsP={S→bA,A→aA,A→a}.
BeginningwiththestartsymbolS,byapplyingthefirstproduction‘A’isobtained.Wheapplyingthreetimesthesecondproductionstrings‘baA’,‘baaA’and‘baaa’consequentlyproduced.Finallythefinalterminalword‘baaaa’isreached,byapplyingtthirdproduction.Onecaneasilyverifythatthisgrammarproducesallwordsthatconsistofaterminalsymbol‘b’followedbyanynumberofsymbols‘a’.Thislanguage,denotedasL,canberepresentedas(III)whereandenotesthen-concatenationofsymbol‘a’.
L={an∣n≥I.}(III)
Sincethesetofproductionsrulesthegenerationofterminalwords,theywillbeusedto
encodethedynamicsofthesystemthatgeneratesthelanguage.Anyword,regardedasa
sequenceofterminalsymbols(which,aswillbedefinedlateroninthepaper,encodethe
outputvariableevolution),thatcanbederivedfromthestartsymbolbyasequenceof
productionsofthegrammarissaidtobeinthelanguagegeneratedbythedynamicalsystem.
Thegrammaticalinferenceprocedureisawayinwhichagrammarisdirectlyinferredfromasetofsamplewords(experimentalpatterns)producedbythedynamicalsystemconsideredasthelinguisticsource.Abasicideainanygrammaticalinferenceprocessisthatthereisnotauniquerelationshipbetweenalanguageandagrammarusedtogenerateit.However,afinitesampledoesnotserveuniquelytodefinealanguage.Inthisway,theinferredgrammarcanonlyrecognizethewordscontainedisS+,andtheothersthatarenotinS+butareofthesamenatureofthoseinS+.
REFERENCES
[I.]J.C.BezdekandS.K.Pal,FuzzyModelsforPatternRecognition,IEEEPress,
Piscataway,N.J.,I.IXIXII.
[II]L.X.WangandJ.Mendel,"Generatingfuzzyrulesbylearningfromexamples,"IEEETrans.onSystems.ManandCybernetics,vol.IIII,no.VI,pp.I.IVI.IV-I.IVIIVII,Jul.I.IXIXII.
[III]S.G.RomaniukandL.O.Hall;"Learningfuzzycontrolrulesfromexamples",InFuzzyControlSystems,ChapterI.VIII,Eds.A.KandelandG.Langholz,CRCPressLLC,BocaRaton,Florida,USA,I.IXIXIV.
[IV]P.J.CostaBrancoandJ.A.Dente,"Theapplicationoffuzzylogicinautomaticmodelingelectromechanicalsystems,"FuzzySetsandSystems,vol.IXV,no.III,pp.IIVIIIII-IIIXIII,I.IXIXVIII.
[V]W.Pedrycz,"Fuzzysetsinpatternrecognition:accomplishmentsandchallenges"FuzzySetsandSystems,vol.IX0,pp.I.VIII.-I.VIIVI,I.IXIXVII.
[VI]N.R.Pal,"Softcomputingforfeatureanalysis"FuzzySetsandSystems,vol.I.0III,pp.II0I.-IIIII.,I.IXIXIX.
[VII]P.J.CostaBrancoandJ.A.Dente,"Anexperimentinautomaticmodelinganelectricaldrivesystemusingfuzzylogic,"IEEETransactionsonSystem,Man,andCybernetics-PartC:ApplicationsandReviews,vol.IIVIII,no.I..,pp.IIVIV-IIVIII,MayI.IXIXVIII.
[VIII]P.J.CostaBranco,J.Martins,A.PiresandJ.A.Dente,Recognizingpatternsinelectromechanicalsystems,PatternRecognitionLetters(PRL)-Specialissue"PatternRecognitioninPractice",vol.I.VIII,pp.I.IIIIIIV-I.IIIIVVI,I.IXIXVIII.
[IX]P.J.CostaBranco"ASimpleAdaptiveSchemeforIndirectFieldOrientationofanInductionMotor".EuropeanTransactionsonElectricalPowerengineering(ETEP),vol.VII,No.IV,July/AugustEurelPublication,VDE-VERLAGI.IXIXVII.
[I.0]Teck-SengLow,T.H.LeeandH.K.Lim,"Amethodologyforneuralnetworktrainingfromcontrolofdriveswithnonlinearities,"IEEETrans.onIndustrialElectronics,vol.IV0,no.II,pp.IIIVIII-IIIVIX,April.I.IXIXIII.
[I.I.]P.J.CostaBranco,N.Lori,andJ.A.Dente,"NewApproachesonStructureIdentificationofFuzzyModels:CaseStudyinanElectromechanicalSystem",T.FuruhashiandY.Uchikawa,Eds.inFuzzyLogic,NeuralNetworks,andEvolutionary
Computation,LNCS/LectureNotesinArtificialIntelligence,Springer-Verlag,Berlin,pp.I.0IV-I.IVIII,I.IXIXVI.
[I.II]N.Chomsky,N.,Aspectsofthetheoryofsyntax,MITPress,Cambridge,Massachusetts,I.IXVIV.
[I.III]E.MarkGold,Languageidentificationinthelimit,InformationandControl,vol.I.0,No.V,pp.IVIVVII-IVVIIIV,MayI.IXVIVII.
[I.IV]K.S.FuandT.L.Booth,"Grammaticalinference:IntroductionandSurvey¨PartI",IEEETrans.onSMC,volV,No.I.,Jan.I.IXVIIV.
版权保护: 本文由 hbsrm.com编辑，转载请保留链接: www.hbsrm.com/lwqt/wxzs/116.html
