UNIT1415Word格式文档下载.docx

UNIT1415Word格式文档下载.docx

《UNIT1415Word格式文档下载.docx》由会员分享，可在线阅读，更多相关《UNIT1415Word格式文档下载.docx（22页珍藏版）》请在冰豆网上搜索。

Thedevelopmentofaneffectivemethodologyforthedesignofintelligentcontrolsystemsundoubtedlyrequiresthesynthesisofmanyconceptsfromartificialintelligence,realtimecomputing,andcontrolengineering.Threeapproachesthathavethepotentialforintelligentcontrolareexpertsystemsasadaptiveelementsinacontrolsystem;

fuzzycalculationsasdecision-producingelementsinacontrolsystem;

andneuralnetworksascompensationelementsincontrolsystems.

Expertsystems,orrule-basedsystems,embodyarule-basedsolvingparadigmbuiltaroundif-thenrules.Whentheprocedureworksforwardfromasequenceofifconditionstoasequenceofthenactions,itiscalledforwardchaining.Whiletheapproachmaybeapplicabletothedecision-makingneedsofacontrolsystem,itisgenerallytooslowforhigh-speedsystemsandithaslimitedlearningcapability.Backwardchainingstartswithadesiredsequenceofthenactionsandworksbackwardtodeterminewhethertheappropriateifconditionsaremet.Itmaybemoreappropriateforcontrolsystemapplicationsbecauseafeedbackloopisgeneratedwhenanerrordetectionandcorrectioncapabilityisaddedtoabackwardchainingsystem.

Infuzzysettheory,themembershipfunctionassignedtotheelementsiscontinuousandliesbetweenzeroandunity,whichisdifferentwiththestandardsettheory.Ifthemembershipfunctiontakesononlytheextremevaluesofzeroandunity,thenfuzzysetsreducetostandardsettheory.Fuzzysettheoryisusefulinthosesituationsinwhichdataandrelationshipscannotbewrittenincrispmathematicalterms.Forexample,anaircraftpilotwhostatesthattherudderisnotworkingcorrectlyisprovidingsignificantfuzzyinformationthatshouldbeintegratedintothedecision-makingprocessoftheoutercontrolloopeventhoughitcannotbecharacterizedbystandardmathematicalmodels.[3]

Neuralnetworksprovideapowerfulapproachfordevelopingempiricalnonlinearmodelsforawidevarietyofphysicalphenomena.Withneuralnetworks,theproblemofcontrolcanbeconsideredasapatternrecognitionproblem,wherethepatternstoberecognizedarechangesignalsthatmapintoactionsignalforspecifiedsystemperformance.Theintelligentcontrollershouldrecognizeandisolatepatternsofchangeinrealtimeandlearnfromexperiencetorecognizechangemorequickly,evenwithincompletedata.

Weareconvincedthatthesetechniquesarecertainlyusefulandanimportantsteptowardintelligentcontrol,butarenottheanswertointelligentcontrolofcomplexandlarge-scalesystems.Inrecentyearsmechanical,electrical,andcomputerengineeringhavefusedtothefieldofmechatronics.Intelligentdevicesincludingsensor,actuator,andsoftwarearebuiltintointelligentmechatroniccomponentsintegratingallthreeaspectsnaturallyinminimumspace.Thecomplexityofamechatronicmoduleismainlyinthecontrolsoftwareoftenreferredtoastheintelligenceofthesystem.

Tosummarizethepropositionforanapproachofintelligentcontrol:

systemandtechnologyintegrationincludingmechatronics,computerscience,systemdesignoptimization,communication,andhumaninteractionmayyieldwhatweexpectfromIntelligentControl.TheviewofIntelligentControlisillustratedinFig.14-A-1.

Fig.14-A-1.KeytechnologiesforIntelligentControlsystem

Onereasontoincludehumaninteractionandtreatitasanessentialparttointelligentcontrolsystemisunfulfilledhighexpectationsofartificialintelligencetocreategoal-orientedintelligentbehavior.[4]

Systemdesignoptimizationandgenerationofoptimalreferencetrajectoriesforcomplexintelligentcontrolsystemsisanimportantissue.Therearevariouspowerfulnumericaloptimizationtoolsavailableandarecentdevelopmentinsemidefiniteprogramming,linearmatrixinequalities（LMI）andinteriorpointmethodshascreatedincreasinglyefficientalgorithms.Oneimportantkeyissueremainingistoimproveperformanceofintelligentcontrolsystemsiterativelywithoutimpairingalreadylearnedknowledge.Onemaycallthiscarefulexplorationtoimproveperformancefromtrialtotrial.Itseemsstilldifficulttoachievethiswithpresentmethodologyfromadaptivecontrolornewertechniqueslikeartificialneuralnetworkorfuzzyapproaches.

Ongoingresearchusingtheintelligentcontrolschemeispointinginadirectionwherethecontinuouscontrolschemeshavetobeaugmentedwithadiscreteeventmodelandsupervisorycontroller.[5]

NewWords

adaptive[ə'

dæ

ptiv]adj.适应的

biology[bai'

ɔlədʒi]n.生物学

dimensionality[di,menʃə'

næ

ləti]n.维度

distributed[di'

stribjutid]adj.分布式的，分散式的

criteria[krai'

tiəriə]n.标准，条件（criterion的复数）

qualitatively['

kwɔliteitivli]adv.定性地，从品质上讲

discipline['

disiplin]n.学科，纪律

imitate['

imiteit]vt.模仿，仿效，仿造，仿制

paradigm['

pæ

rədim]n.范例

empirical[em'

pirikəl]adj.根据经验的，经验主义的

mechatronic[,mekə'

trɔnik]adj.机电一体化的

mechatronics[,mekə'

trɔniks]n.机械电子学

methodology[,meθə'

dɔlədʒi]n.方法学，方法论

perceive[pə'

si:

v]vt.感知，察觉，感觉，理解，认知；

vi.感到，感知

proposition[,prɔpə'

ziʃən]n.命题，建议

psychology[sai'

kɔlədʒi]n.心理学

recognition[,rekəg'

niʃən]n.识别

spectrum[´

spektrəm]n.光谱，频谱，范围

stringent['

strindʒənt]adj.严厉的，迫切的

Phrases

Bayesianprobability贝叶斯统计

fuzzylogic模糊逻辑

machinelearning机器学习

evolutionarycomputation进化计算

geneticalgorithms遗传算法

decisionmaker决策者

decisionspace决策空间

noiselevel噪声电平

copewith处理，应付

neuralnetwork神经元网络

asequenceof一系列

feedbackloop反馈回路

membershipfunction隶属度函数

supervisorycontroller监督控制

Abbreviations

AI（ArtificialIntelligence）人工智能

ANN（ArtificialNeuralNetworks）人工神经网络

LMI（linearmatrixinequalities）线性矩阵不等式

Notes

1.Asaresult,overthepastfourtofivedecadesnumeroustermsborrowedfrompsychologyandbiology,suchaspatternrecognition,adaptation,learningandselforganization,havebeenintroducedintothesystemsliterature.

因此，在过去的四、五十年内，很多心理学、生物学等学科的概念，如模式识别、适应性、学习和自组织，被引入系统论中。

2.Agooddefinitionforintelligencecontrolisthedisciplineinwhichcontrolalgorithmsaredevelopedbyemulatingcertaincharacteristicsofintelligentbiologicalsystems.

智能控制的一个恰当的定义是一门学科，控制算法的发展是通过模拟智能生物系统的某些特点来实现的。

3.Forexample,anaircraftpilotwhostatesthattherudderisnotworkingcorrectlyisprovidingsignificantfuzzyinformationthatshouldbeintegratedintothedecision-makingprocessoftheoutercontrolloopeventhoughitcannotbecharacterizedbystandardmathematicalmodels.

例如，飞机员指出方向舵工作错误是给出的一个重要模糊信息，即使这一信息不能用经典数学模型来描述，也必须被加入外部控制回路的决策过程中。

4.Onereasontoincludehumaninteractionandtreatitasanessentialparttointelligentcontrolsystemisunfulfilledhighexpectationsofartificialintelligencetocreategoal-orientedintelligentbehavior.

将与人的交互作用包括在智能控制系统中，并将其看作一个重要的组成部分的一个原因是，人工智能难以实现目标导向的智能行为。

5.Ongoingresearchusingtheintelligentcontrolschemeispointinginadirectionwherethecontinuouscontrolschemeshavetobeaugmentedwithadiscreteeventmodelandsupervisorycontroller.

正在进行中的研究利用智能控制在一个方向上计划，在这个方向上，连续控制计划必须与离散事件模型和监控器相结合。

B：

ArtificialNeuralNetwork

Artificialneuralnetworks,alsocalledneuralnetworks,areinformationproces