UNIT1415.docx
- 文档编号:8176133
- 上传时间:2023-01-29
- 格式:DOCX
- 页数:22
- 大小:144.36KB
UNIT1415.docx
《UNIT1415.docx》由会员分享,可在线阅读,更多相关《UNIT1415.docx(22页珍藏版)》请在冰豆网上搜索。
UNIT1415
UNIT14
Intelligentcontrolisaclassofcontroltechniques,thatusevariousAI(ArtificialIntelligence)computingapproacheslikeANN(ArtificialNeuralNetworks),Bayesianprobability,fuzzylogic,machinelearning,evolutionarycomputationandgeneticalgorithms.Inthisunit,theoverviewofintelligentcontrolandtheNeuralNetworkispresented.
A:
IntelligentControl
Ascontrolmethodshavefoundtheirwayintostandardpractice,theyhaveopenedthedoortoawidespectrumofcomplexapplications.Suchcomplexsystemsarecharacterizedbypoormodels,highdimensionalityofthedecisionspace,distributedsensorsanddecisionmakers,highnoiselevels,multiplesubsystems,levels,time-scalesand/orperformancecriteria,complexinformationpatterns,overwhelmingamountofdataandstringentperformancerequirements.Sincetheabilityoflivingsystemstocopewithsimilarsituationsiswellknown,systemstheoristsattemptedtoincorporatesuchcharacteristicsinartificialsystems.
Asaresult,overthepastfourtofivedecadesnumeroustermsborrowedfrompsychologyandbiology,suchaspatternrecognition,adaptation,learningandselforganization,havebeenintroducedintothesystemsliterature.[1]Intelligentcontrolismerelythelatestinthisseriesofterms.Qualitatively,asystemwhichincludestheabilitytosenseitsenvironment,processtheinformationtoreduceuncertainty,plan,generateandexecutecontrolactioninthesituationsmentionedearlier,constitutesanintelligentcontrolsystem.Thegreatertheabilitytodealwiththeabovedifficulties,themoreintelligentisthecontrolsystem.
Agooddefinitionforintelligencecontrolisthedisciplineinwhichcontrolalgorithmsaredevelopedbyemulatingcertaincharacteristicsofintelligentbiologicalsystems.[2]Itisquicklyemergingasatechnologythatmayopenavenuesforsignificantadvancesinmanyareas.Infact,fueledbyadvancementincomputingtechnology,italreadyachievedsomeveryexcitingandpromisingresults.Currently,thereismuchinterestindesigningintelligentcontrolsystemsforsuchapplicationsasroboticdevicesworkinginunstructuredenvironments,highperformanceaircraftwithselfrepairingcapabilities,andcomplexindustrialplantssubjecttocatastrophicactuatororsensorfailures.
Oneofthekeygoalsinintelligentcontrolsystemdesignistoimitatehumanbehavior.Ifwewereabletogeneraterealartificialintelligence,theresultingbehaviorwouldbenaturallyperceivedasintelligent.Butsincethisseemstobethelong-termchallengeitismorereasonabletofollowthesubgoaltoimitatehumans.
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'dæptiv]adj.适应的
biology[bai'ɔlədʒi]n.生物学
dimensionality[di,menʃə'næləti]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['pærədim]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
- 配套讲稿:
如PPT文件的首页显示word图标,表示该PPT已包含配套word讲稿。双击word图标可打开word文档。
- 特殊限制:
部分文档作品中含有的国旗、国徽等图片,仅作为作品整体效果示例展示,禁止商用。设计者仅对作品中独创性部分享有著作权。
- 关 键 词:
- UNIT1415
![提示](https://static.bdocx.com/images/bang_tan.gif)