附件1智能控制課程試題A附件1題號一二三四五六七總分分數(shù)合分人：復(fù)查人：一、填空題〔每空1分，共20分〕分數(shù)評卷人1．智能控制系統(tǒng)的根本類型有、、、、和。2．智能控制具有2個不同于常規(guī)控制的本質(zhì)特點：和。3．一個理想的智能控制系統(tǒng)應(yīng)具備的性能是、、、、等。4.人工神經(jīng)網(wǎng)絡(luò)常見的輸出變換函數(shù)有：和。5.人工神經(jīng)網(wǎng)絡(luò)的學(xué)習(xí)規(guī)那么有：、和。6.在人工智能領(lǐng)域里知識表示可以分為和兩類。二、簡答題：〔每題5分，共30分〕分數(shù)評卷人1.智能控制系統(tǒng)應(yīng)具有的特點是什么？2.智能控制系統(tǒng)的結(jié)構(gòu)一般有哪幾局部組成，它們之間存在什么關(guān)系？3.比擬智能控制與傳統(tǒng)控制的特點。4．神經(jīng)元計算與人工智能傳統(tǒng)計算有什么不同？5．人工神經(jīng)元網(wǎng)絡(luò)的拓撲結(jié)構(gòu)主要有哪幾種？6．簡述專家系統(tǒng)與傳統(tǒng)程序的區(qū)別。三、作圖題：〔每圖4分，共20分〕分數(shù)評卷人1.畫出以下應(yīng)用場合下適當(dāng)?shù)碾`屬函數(shù)：〔a〕我們絕對相信附近的e(t)是“正小”，只有當(dāng)e(t)足夠遠離時，我們才失去e(t)是“正小”的信心；〔b〕我們相信附近的e(t)是“正大”，而對于遠離的e(t)我們很快失去信心；〔c〕隨著e(t)從向左移動，我們很快失去信心，而隨著e(t)從向右移動，我們較慢失去信心。2.畫出以下兩種情況的隸屬函數(shù)：〔a〕精確集合的隸屬函數(shù)；〔b〕寫出單一模糊〔singletonfuzzification〕隸屬函數(shù)的數(shù)學(xué)表達形式，并畫出隸屬函數(shù)圖。四、計算題：〔每題10分，共20分〕分數(shù)評卷人1.一個模糊系統(tǒng)的輸入和輸出的隸屬函數(shù)如圖1所示。試計算以下條件和規(guī)那么的隸屬函數(shù)：〔a〕規(guī)那么1：Iferroriszeroandchang-in-erroriszeroThenforceiszero。均使用最小化操作表示蘊含(usingminimumopertor)；〔b〕規(guī)那么2：Iferroriszeroandchang-in-errorispossmallThenforceisnegsmall。均使用乘積操作表示蘊含(usingproductopertor)；2.設(shè)論域，且試求〔補集〕，〔補集〕五、試論述對BP網(wǎng)絡(luò)算法的改良。〔共10分〕分數(shù)評卷人附件1智能控制課程試題B附件1題號一二三四五六七總分分數(shù)合分人：復(fù)查人：一、填空題〔每空1分，共20分〕分數(shù)評卷人1．智能控制的研究對象具備的特點有：、和。2．智能控制系統(tǒng)的主要類型有：、、、、和。3．確定隸屬函數(shù)的方法大致有、和。4.國內(nèi)外學(xué)者提出了許多面向?qū)ο蟮纳窠?jīng)網(wǎng)絡(luò)控制結(jié)構(gòu)和方法，從大類上看，較具代表性的有以下幾種：、和。5.在一個神經(jīng)網(wǎng)絡(luò)中，常常根據(jù)處理單元的不同處理功能，將處理單元分成有以下三種：、和。6.專家系統(tǒng)具有三個重要的特征是：、和。二、簡答題：〔每題5分，共30分〕分數(shù)評卷人智能控制有哪些應(yīng)用領(lǐng)域？試舉例說明其工作原理。試說明智能控制的三元結(jié)構(gòu)，并畫出展示它們之間關(guān)系的示意圖。模糊邏輯與隨機事件的聯(lián)系與區(qū)別。給出典型的神經(jīng)元模型。BP根本算法的優(yōu)缺點。專家系統(tǒng)的根本組成。三、作圖題：〔每圖4分，共20分〕分數(shù)評卷人1.畫出以下應(yīng)用場合下適當(dāng)?shù)碾`屬函數(shù)：〔a〕隨著e(t)從向左移動，我們很快失去信心，而隨著e(t)從向右移動，我們較慢失去信心。〔b〕我們相信附近的e(t)是“正大”，而對于遠離的e(t)我們很快失去信心；〔c〕我們絕對相信附近的e(t)是“正小”，只有當(dāng)e(t)足夠遠離時，我們才失去e(t)是“正小”的信心；2.畫出以下兩種情況的隸屬函數(shù)：〔a〕精確集合的隸屬函數(shù)；〔b〕寫出單一模糊〔singletonfuzzification〕隸屬函數(shù)的數(shù)學(xué)表達形式，并畫出隸屬函數(shù)圖。四、計算題：〔每題10分，共20分〕分數(shù)評卷人1.一個模糊系統(tǒng)的輸入和輸出的隸屬函數(shù)如圖1所示。試計算以下條件和規(guī)那么的隸屬函數(shù)：〔a〕規(guī)那么1：Iferroriszeroandchang-in-errorisnegsmallThenforceispossmall。均使用最小化操作表示蘊含(usingminimumopertor)；〔b〕規(guī)那么2：Iferroriszeroandchang-in-errorispossmallThenforceisnegsmall。均使用乘積操作表示蘊含(usingproductopertor)；2.設(shè)論域，且試求〔補集〕，〔補集〕五、試論述建立專家系統(tǒng)的步驟?！补?0分〕分數(shù)評卷人附件1智能控制課程試題C附件1題號一二三四五六七總分分數(shù)合分人：復(fù)查人：一、填空題〔每空1分，共20分〕分數(shù)評卷人1．智能控制是一門新興的學(xué)科，它具有非常廣泛的應(yīng)用領(lǐng)域，例如、、、和。2．傳統(tǒng)控制包括和。3．一個理想的智能控制系統(tǒng)應(yīng)具備的性能是、、、、等。4．學(xué)習(xí)系統(tǒng)的四個根本組成局部是、、、。5．專家系統(tǒng)的根本組成局部是、、。二、簡答題：〔每題5分，共30分〕分數(shù)評卷人智能控制系統(tǒng)的結(jié)構(gòu)一般有哪幾局部組成，它們之間存在什么關(guān)系？智能控制系統(tǒng)有哪些類型，各自的特點是什么？比擬智能控制與傳統(tǒng)控制的特點。4．根據(jù)外部環(huán)境所提供的知識信息與學(xué)習(xí)模塊之間的相互作用方式，機器學(xué)習(xí)可以劃分為哪幾種方式？5．建造專家控制系統(tǒng)大體需要哪五個步驟？6．為了把專家系統(tǒng)技術(shù)應(yīng)用于直接專家控制系統(tǒng)，在專家系統(tǒng)設(shè)計上必須遵循的原那么是什么？三、作圖題：〔每圖4分，共20分〕分數(shù)評卷人1.畫出以下應(yīng)用場合下適當(dāng)?shù)碾`屬函數(shù)：〔a〕我們絕對相信附近的e(t)是“正小”，只有當(dāng)e(t)足夠遠離時，我們才失去e(t)是“正小”的信心；〔b〕我們相信附近的e(t)是“正大”，而對于遠離的e(t)我們很快失去信心；〔c〕隨著e(t)從向左移動，我們很快失去信心，而隨著e(t)從向右移動，我們較慢失去信心。2.畫出以下兩種情況的隸屬函數(shù)：〔a〕精確集合的隸屬函數(shù)；〔b〕寫出單一模糊〔singletonfuzzification〕隸屬函數(shù)的數(shù)學(xué)表達形式，并畫出隸屬函數(shù)圖。四、計算題：〔每題10分，共20分〕分數(shù)評卷人1.一個模糊系統(tǒng)的輸入和輸出的隸屬函數(shù)如圖1所示。試計算以下條件和規(guī)那么的隸屬函數(shù)：〔a〕規(guī)那么1：Iferroriszeroandchang-in-erroriszeroThenforceiszero。均使用最小化操作表示蘊含(usingminimumopertor)；〔b〕規(guī)那么2：Iferroriszeroandchang-in-errorispossmallThenforceisnegsmall。均使用乘積操作表示蘊含(usingproductopertor)；2.設(shè)論域，且試求〔補集〕，〔補集〕五、畫出靜態(tài)多層前向人工神經(jīng)網(wǎng)絡(luò)〔BP網(wǎng)絡(luò)〕的結(jié)構(gòu)圖，并簡述BP神經(jīng)網(wǎng)絡(luò)的工作過程〔10分〕分數(shù)評卷人。附件1智能控制課程試題D附件1題號一二三四五六七總分分數(shù)合分人：復(fù)查人：一、填空題〔每空1分，共20分〕分數(shù)評卷人1．智能控制是一門新興的學(xué)科，它具有非常廣泛的應(yīng)用領(lǐng)域，例如、、、和。2．智能控制系統(tǒng)的主要類型有：、、、、和。3．一個理想的智能控制系統(tǒng)應(yīng)具備的性智能能是、、等。4．在設(shè)計知識表達方法時，必須從表達方法的、、這四個方面全面加以均衡考慮。5．在一個神經(jīng)網(wǎng)絡(luò)中，常常根據(jù)處理單元的不同處理功能，將處理單元分成輸入單元、輸出單元和三類。二、簡答題：〔每題5分，共30分〕分數(shù)評卷人智能控制系統(tǒng)的結(jié)構(gòu)一般有哪幾局部組成，它們之間存在什么關(guān)系？試說明智能控制的三元結(jié)構(gòu)，并畫出展示它們之間關(guān)系的示意圖。比擬智能控制與傳統(tǒng)控制的特點。4．神經(jīng)網(wǎng)絡(luò)應(yīng)具的四個根本屬性是什么？5.神經(jīng)網(wǎng)絡(luò)的學(xué)習(xí)方法有哪些？6.按照專家系統(tǒng)所求解問題的性質(zhì)，可分為哪幾種類型？三、作圖題：〔每圖4分，共20分〕分數(shù)評卷人1.畫出以下應(yīng)用場合下適當(dāng)?shù)碾`屬函數(shù)：〔a〕我們絕對相信附近的e(t)是“正小”，只有當(dāng)e(t)足夠遠離時，我們才失去e(t)是“正小”的信心；〔b〕我們相信附近的e(t)是“正大”，而對于遠離的e(t)我們很快失去信心；〔c〕隨著e(t)從向左移動，我們很快失去信心，而隨著e(t)從向右移動，我們較慢失去信心。2.畫出以下兩種情況的隸屬函數(shù)：〔a〕精確集合的隸屬函數(shù)；〔b〕寫出單一模糊〔singletonfuzzification〕隸屬函數(shù)的數(shù)學(xué)表達形式，并畫出隸屬函數(shù)圖。四、計算題：〔每題10分，共20分〕分數(shù)評卷人1.一個模糊系統(tǒng)的輸入和輸出的隸屬函數(shù)如圖1所示。試計算以下條件和規(guī)那么的隸屬函數(shù)：〔a〕規(guī)那么1：Iferroriszeroandchang-in-erroriszeroThenforceiszero。均使用最小化操作表示蘊含(usingminimumopertor)；〔b〕規(guī)那么2：Iferroriszeroandchang-in-errorispossmallThenforceisnegsmall。均使用乘積操作表示蘊含(usingproductopertor)；2.設(shè)論域，且試求〔補集〕，〔補集〕五、試述專家控制系統(tǒng)的工作原理〔共10分〕分數(shù)評卷人Fuzzycontrolofaball-balancingsystemⅠ.IntroductionTheball-balancingsystemconsistsofacartwithanarcmadeoftwoparallelpipesonwhichasteelballrolls.Thecartmovesonapairoftrackshorizontallymountedonaheavysupport(Fig.1).Thecontrolobjectiveistobalancetheballonthetopofthearcandatthesametimeplacethecartinadesiredposition.Itiseducational,becausethelaboratoryrigissufficientlyslowforvisualinspectionofdifferentcontrolstrategiesandthemathematicalmodelissufficientlycomplextobechallenging.Itisaclassicalpendulumproblem,liketheonesusedasabenchmarkproblemforfuzzyandneuralnetcontrollers,assalesmaterialforfuzzydesigntools.Initially,thecartisinthemiddleofthetrackandtheballisontheleftsideofthecurvedarc.Acontrollerpullsthecartlefttogettheballupnearthemiddle,thenthecontrolleradjuststhecartpositionverycarefully,withoutloosingtheball.Fuzzycontrolprovidesaformatmethodologyforrepresenting,manipulatingandimplementingahuman’sheuristicknowledgeabouthowtocontrolasystem[1-3].Here,thefuzzycontroldesignmethodwillbeusedtocontroltheball-balancingsystem.Fig.1Ball-balancinglaboratoryrigⅡ.Designobjectivea).Learningtheoperatingprincipleoftheball-balancingsystem;b).Masteringthefuzzycontrolprincipleanddesignprocedure;c).Enhancingtheprogrammingpowerusingmatlab.Ⅲ.Designrequirementsa).Balancingtheballonthetopofthearcandatthesametimeplacethecartinadesiredposition.b).Comparingthecontrolresultofthelinearcontrollerwiththatofthefuzzycontrollerandthinkingabouttheadvantageoffuzzycontroltoconventionalcontrol.Ⅳ.Designprinciple①Modeldescriptionoftheball-balancingsystemIntroducethestatevectorofstatevariables(representscartpositionandrepresentsballangulardeviation)Thenonlinearstate-spaceequations[5]aregivenasfollows:Whererepresentscartradiusofthearc,isthecartweight,representscartdrivingforce,istheballradius,istheballrollingradius,istheballweight,istheballmomentofinertiaandrepresentsgravity.Themodelcanbelinearisedaroundtheorigin.Theapproximationstothetrigonometricfunctionsareintroducedasfollowsandthelinearstate-spacemodelcanbeobtainedasfollowsMatricesaresimplyandgivenasfollowswith,Theactualvaluesoftheconstantsare.②FuzzycontrollerdesignTherearespecificcomponentscharactersticofafuzzycontrollertosupportadesignprocedure.IntheblockdiagraminFig.2,thefuzzycontrollerhasfourmaincomponents.Thefollowingexplainstheblockdiagram.Fig.2FuzzycontrollerarchitectureFuzzificationThefirstcomponentisfuzzification,whichconvertseachpieceofinputdatatodegreesofmembershipbyalookupinoneofseveralmembershipfunctions.Thefuzzificationblockthusmatchestheinputdatawiththeconditionsoftherulestodeterminehowwelltheconditionofeachrulematchesthatparticularinputinstance.RulebaseTherulebasecontainsafuzzylogicquantificationoftheexpert’slinguisticdescriptionofhowtoachievegoodcontrol.c.InferenceengineForeachrule,theinferenceenginelooksupthemembershipvaluesintheconditionoftherule.AggregationTheaggregationoperationisusedwhencalculatingthedegreeoffulfillmentorfiringstrengthoftheconditionofarule.Aggregationisequivalenttofuzzification,whenthereisonlyoneinputtothecontroller.Aggreagtionissometimesalsocalledfufilmentoftheruleorfiringstrength.ActivationTheactivationofaruleisthedeductionoftheconclusion,possiblyreducedbyitsfiringstrength.Arulecanbeweightedbyaprioribyaweightingfactor,whichisitsdegreeofconfidence.Thedegreeofconfidenceisdeterminedbythedesigner,oralearningprogramtryingtoadapttherulestosomeinput-outputrelationship.AccumulationAllactivatedconclusionsareaccumulatedusingthemaxoperation.d.DefuzzificationTheresultingfuzzysetmustbeconvertedtoanumberthatcanbesenttotheprocessesasacontrolsignal.Thisoperationiscalleddefuzzification.Theoutputsetscanbesingletons,buttheycanalsobelinearcombinationsoftheinputs,orevenafunctionoftheinputs.TheT-SfuzzymodelwasproposedbyTakagiandSugenoinanefforttodevelopasystematicapproachtogeneratingfuzzyrulesfromagiveninput-outputdataset[4].Itsrulestructurehasthefollowingform:Whereisafuzzyset,istheinput,isthenumberofinputs，istheoutputspecifiedbytherule，isthetruthvalueparameter.Usingfuzzyinferencebaseduponproduct-sum-gravityatagiveninput,，thefinaloutputofthefuzzymodel,isinferredbytakingtheweightedaverageofwhereisthenumberoffuzzyrules,theweight,impliestheoveralltruthvalueoftherulecalculatedbasedonthedegreesofmembershipvalues:③ComputersimulationThesimulationresultscanbeobtainedbythedesignedprogramusingmatlab.Initialconditionscanbechangedandcontrollergainscanbeadjusted.Thenthedesiredresultscanbeobtained.Ⅴ.Designprocedurea).Themodeloftheball-balancingsystemhasbeengiven;b).Fuzzycontrollerdesign;Fuzzycontroldesignessentiallyamountsto(1)choosingthefuzzycontrollerinputsandoutputs(2)choosingthepreprocessingthatisneededforthecontrollerinputsandpossiblypostprocessingthatisneededfortheoutputs,and(3)designingeachofthefourcomponentsofthefuzzycontrollershowninFig.2.c).Computersimulation.References[1].K.M.PassinoandS.Yurkovich(1997).Fuzzycontrol,1stedn,AddisionWesleyLongman,Colifornia.[2].CaiZixing.IntelligentControl:Principles,TechniquesandApplications.Singapore-NewJersey:WorldScientificPublishers,Dec.1997.[3].Pedrycz,W.(1993).Fuzzycontrolandfuzzysystems,secondedn,WileyandSons,NewYork.[4].Takagi,T.andSugno,M.(1985).Fuzzyidentificationofsystemsanditsapplicationstomodelingandcontrol,IEEETrans.Systems,Man&Cybernetics15(1):116-132.SpeedcontroldesignforavehiclesystemusingfuzzylogicⅠ.IntroductionEngineandotherautomobilesystemsareincreasinglycontrolledelectronically.Thishasledtoimprovedfueleconomy,reducedpollution,

improveddrivingsafetyandreducedmanufacturingcosts.Howevertheautomobile

isahostileenvironment:especiallyintheenginecompartment,wherehightemperature,humidity,vibration,electricalinterferenceandafinecocktailofpotentiallycorrosivepollutantsarepresent.Thesehostilefactorsmaycauseelectricalcontactstodeteriorate,surfaceresistancestofallandsensitiveelectronicsystemstofailinavarietyofmodes.Someofthesefailuremodeswillbebenign,whereasothersmaybedangerousandcauseaccidentsandendangertohumanlife.Acruisecontrolsystem,orvehiclespeedcontrolsystemcankeepavehicle'sspeedconstantonlongrunsandthereforemayhelppreventdriverfatigue[2-5].Ifthedriverhandsoverspeedcontroltoacruisecontrolsystem,thenthecapabilityofthesystemtocontrolspeedtothesetvalueisjustascriticaltosafetyasisthecapabilityofthedrivertocontrolspeedmanually.Sothecruisecontrolsystemdesignisimperativeandimportanttoanautomobile.Ⅱ.Designrequirementsa).Designingcontrollerusingfuzzylogic;b).Makingtheautomobile’sspeedkeepconstant.Ⅲ.ModeldescriptionoftheautomobileThedynamicsoftheautomobile[1]aregivenasfollowsWhereisthecontrolinput(representsathrottleinputandrepresentsabrakeinput),isthemassofthevehicle,isitsaerodynamicdrag,isaconstantfrictionalforce,isthedriving/brakingforce,andsecissaturatedat).Wecanusefuzzycontrolmethodtodesignacruisecontrolsystem.Obviously,thefuzzycruisecontroldesignobjectiveistodevelopafuzzycontrollerthatregulatesavehicle’sspeedtoadriver-specifiedvalue.Ⅳ.SpeedcontroldesignusingfuzzylogicFuzzycontrollogicandneuralnetworksareotherexamplesofmethodologiescontrolengineersareexaminingtoaddressthecontrolofverycomplexsystems.Agoodfuzzycontrollogicapplicationisincruisecontrolarea.1)DesignofPIfuzzycontrollerSupposethatwewishtobeabletotrackasteporrampchangeinthedriver-specifiedspeedvalueveryaccurately.A“PIfuzzycontroller”canbeusedasshowninFig.1.InFig.1,thefuzzycontrollerisdenotedby;andarescalinggains;andistheinputoftheintegrator.Fig.1SpeedcontrolsystemusingaPIfuzzycontrollerFindthedifferentialequationthatdescribestheclosed-loopsystem.Letthestatebeandfindasystemofthreefirst-orderordinarydifferentialequationsthatcanbeusedbytheRunge-Kuttamethodinthesimulationoftheclosed-loopsystem.isusedtorepresentthecontrollerinthedifferentialequations.Forthereferenceinput,threedifferenttestsignalscanbeusedasfollows:a:Testinput1makes=18m/sec(40.3mph)forand＝22m/sec(49.2mph)for.b:Testinput2makes=18m/sec(40.3mph)forandincreaseslinearly(aramp)from18to22by,andthenfor.c:Testinput3makes=22forandweuseastheinitialcondition(thisrepresentsstartingthevehicleatrestandsuddenlycommandingalargeincreasespeed).Usefortestinput1and2.Designthefuzzycontrollertogetlessthan2%overshoot,arise-timebetween5and7sec,andasettlingtimeoflessthan8sec(i.e.,reachtowithin2%ofthefinalvaluewithin8sec)forthejumpfrom18to22in“testinput1”thatisdefinedabove.Also,fortherampinput(“testinput2”above)itmusthavelessthan1mph(0.447)steady-stateerror(i.e.,attheendoftheramppartoftheinputhavelessthan1mpherror).Fullyspecifythecontroller(e.g.,themembershipfunctions,rule-basedefuzzification,etc.)andsimulatetheclosed-loopsystemtodemonstratethatitperformsproperly.Provideplotsofandonthesameaxisandonadifferentplot.Fortestinput3findtherise-time,overshoot,2%settlingtime,andsteady-stateerrorfortheclosed-loopsystemforthecontrollerthatyoudesignedtomeetthespecificationsfortestinput1and2.UsingtheRunge-Kuttamethodandintegrationstepsizeof0.01,thesimulationresultscanbeshownasfollows.①．Testinput1Fig.2Vehiclespeedsandtheoutputoffuzzycontrollerusingtestinput1②．Testinput2Fig.3Vehiclespeedsandtheoutputoffuzzycontrollerusingtestinput2③．Testinput3Fig.4Vehiclespeedsandtheoutputoffuzzycontrollerusingtestinput32)DesignofPDfuzzycontrollerSupposethatyouareconcernedwithtrackingastepchangeinaccuratelyandthatyouusethePDfuzzycontrollershowninFig.5.Torepresentthederivative,simplyuseabackwarddifferenceWhereistheintegrationstepsizeinyoursimulation(oritcouldbeyoursamplingperiodinanimplementation).Fig.5SpeedcontrolsystemusingaPDfuzzycontrollerDesignaPDfuzzycontrollertogetlessthan2%overshoot,arise-timebetween7and10sec.andasettlingtimeoflessthan10secfortestinput1definedina).Also,fortherampinput(testinput2in1))itmusthavelessthan1mphsteady-stateerrortotheramp(i.e.,attheendoftheramppartoftheinput,havelessthan1mpherror).Fullyspecifyyourcontrollerandsimulatetheclosed-loopsystemtodemonstratethatitperformsproperly.Provideplotsofandonthesameaxisandonadifferentplot.Inthesimulations,theRunge-Kuttamethodisusedandanintegrationstepsizeof0.01.Assumethatfortestinputs1and2(henceweignorethederivativeinputincomingupwiththestateequationsfortheclosed-loopsystemandsimplyusetheapproximationforc(t)thatisshownabovesothatwehaveatwo-statesystem).Asafinaltestletandusetestinput3definedin1).①．Testinput1Fig.6Vehiclespeedsandtheoutputoffuzzycontrollerusingtestinput1②．Testinput2Fig.7Vehiclespeedsandtheoutputoffuzzycontrollerusingtestinput2③．Testinput3Fig.8Vehiclespeedsandtheoutputoffuzzycontrollerusingtestinput3Ⅴ.SummaryTokeepanautomobile’sspeedconstant,aspeedcontroldesignmethodusingfuzzylogicispresented.PIfuzzycontrollerandPDfuzzycontrollerdesignschemesaregiventoregulateavehicle’sspeedtoadriver-specifiedvalue.Thesimulationresultsshowthevalidityandoftheproposedtechnique.Thecontroldesignprocedurecanbesummarizedasfollows:ModelingandperformanceobjectivesBasically,theroleofmodelingafuzzycontroldesignisquite