An introduction to data-driven control systems ali khaki-sedigh - Get instant access to the full ebo

Page 1


Instant digital products (PDF, ePub, MOBI) ready for you

Download now and discover formats that fit your needs...

Introduction to linear control systems Bavafa-Toosi

https://ebookmass.com/product/introduction-to-linear-control-systemsbavafa-toosi/ ebookmass.com

An Introduction to Data Science 1st Edition, (Ebook PDF)

https://ebookmass.com/product/an-introduction-to-data-science-1stedition-ebook-pdf/

ebookmass.com

Data-driven Solutions to Transportation Problems Yinhai Wang

https://ebookmass.com/product/data-driven-solutions-to-transportationproblems-yinhai-wang/ ebookmass.com

A Cowboy State of Mind Jennie Marts

https://ebookmass.com/product/a-cowboy-state-of-mind-jennie-marts-3/ ebookmass.com

Divine

Democracy: Political Theology after Carl Schmitt

https://ebookmass.com/product/divine-democracy-political-theologyafter-carl-schmitt-miguel-vatter/

ebookmass.com

To Die in Tuscany David P. Wagner

https://ebookmass.com/product/to-die-in-tuscany-david-p-wagner-2/

ebookmass.com

Pro Jakarta EE 10: Open Source Enterprise Java-based Cloud-native Applications Development Peter Späth

https://ebookmass.com/product/pro-jakarta-ee-10-open-sourceenterprise-java-based-cloud-native-applications-development-peterspath/

ebookmass.com

Thought and Play in Musical Rhythm: Asian, African, and Euro-American Perspectives Richard Wolf

https://ebookmass.com/product/thought-and-play-in-musical-rhythmasian-african-and-euro-american-perspectives-richard-wolf/

ebookmass.com

Diagnostic Imaging: Gynecology 3rd Edition Akram M. Shaaban

https://ebookmass.com/product/diagnostic-imaging-gynecology-3rdedition-akram-m-shaaban/

ebookmass.com

Design Thinking and Innovation Metrics: Powerful Tools to Manage Creativity, OKRs, Product, and Business Success

https://ebookmass.com/product/design-thinking-and-innovation-metricspowerful-tools-to-manage-creativity-okrs-product-and-business-successmichael-lewrick/

ebookmass.com

AnIntroductiontoData-DrivenControlSystems

IEEEPress

445HoesLane

Piscataway,NJ08854

IEEEPressEditorialBoard

SarahSpurgeon, EditorinChief

JónAtliBenediktsson

AnjanBose

JamesDuncan

AminMoeness

DesineniSubbaramNaidu

BehzadRazavi

JimLyke

HaiLi

BrianJohnson

JeffreyReed

DiomidisSpinellis

AdamDrobot

TomRobertazzi

AhmetMuratTekalp

Copyright©2024byTheInstituteofElectricalandElectronicsEngineers,Inc. Allrightsreserved.

PublishedbyJohnWiley&Sons,Inc.,Hoboken,NewJersey.

PublishedsimultaneouslyinCanada.

Nopartofthispublicationmaybereproduced,storedinaretrievalsystem,ortransmittedinany formorbyanymeans,electronic,mechanical,photocopying,recording,scanning,orotherwise, exceptaspermittedunderSection107or108ofthe1976UnitedStatesCopyrightAct,without eitherthepriorwrittenpermissionofthePublisher,orauthorizationthroughpaymentofthe appropriateper-copyfeetotheCopyrightClearanceCenter,Inc.,222RosewoodDrive,Danvers, MA01923,(978)750-8400,fax(978)750-4470,oronthewebatwww.copyright.com.Requeststo thePublisherforpermissionshouldbeaddressedtothePermissionsDepartment,JohnWiley& Sons,Inc.,111RiverStreet,Hoboken,NJ07030,(201)748-6011,fax(201)748-6008,oronlineat http://www.wiley.com/go/permission.

Trademarks:WileyandtheWileylogoaretrademarksorregisteredtrademarksofJohnWiley& Sons,Inc.and/oritsaffiliatesintheUnitedStatesandothercountriesandmaynotbeused withoutwrittenpermission.Allothertrademarksarethepropertyoftheirrespectiveowners. JohnWiley&Sons,Inc.isnotassociatedwithanyproductorvendormentionedinthisbook.

LimitofLiability/DisclaimerofWarranty:Whilethepublisherandauthorhaveusedtheirbest effortsinpreparingthisbook,theymakenorepresentationsorwarrantieswithrespecttothe accuracyorcompletenessofthecontentsofthisbookandspecificallydisclaimanyimplied warrantiesofmerchantabilityorfitnessforaparticularpurpose.Nowarrantymaybecreatedor extendedbysalesrepresentativesorwrittensalesmaterials.Theadviceandstrategiescontained hereinmaynotbesuitableforyoursituation.Youshouldconsultwithaprofessionalwhere appropriate.Neitherthepublishernorauthorshallbeliableforanylossofprofitoranyother commercialdamages,includingbutnotlimitedtospecial,incidental,consequential,orother damages.Further,readersshouldbeawarethatwebsiteslistedinthisworkmayhavechanged ordisappearedbetweenwhenthisworkwaswrittenandwhenitisread.Neitherthepublisher norauthorsshallbeliableforanylossofprofitoranyothercommercialdamages,includingbut notlimitedtospecial,incidental,consequential,orotherdamages.

Forgeneralinformationonourotherproductsandservicesorfortechnicalsupport,please contactourCustomerCareDepartmentwithintheUnitedStatesat(800)762-2974,outsidethe UnitedStatesat(317)572-3993orfax(317)572-4002.

Wileyalsopublishesitsbooksinavarietyofelectronicformats.Somecontentthatappearsin printmaynotbeavailableinelectronicformats.FormoreinformationaboutWileyproducts, visitourwebsiteatwww.wiley.com.

LibraryofCongressCataloging-in-PublicationDataAppliedfor:

HardbackISBN:9781394196401

CoverDesign:Wiley CoverImage:©zfL/GettyImages

Setin9.5/12.5ptSTIXTwoTextbyStraive,Chennai,India

2PhilosophicalPerspectivesoftheParadigmShiftinControl SystemsDesignandtheRe-EmergenceofData-Driven Control 35

2.1Introduction 35

2.2BackgroundMaterials 36

2.2.1ScientificTheory 36

2.2.2ScientificRevolutionsandParadigmShifts 37

2.2.3RevolutionsinControlSystemsDesignfromKuhn’sPerspective 39

2.2.4PhilosophicalIssuesinControlEngineeringandControlSystems Design 41

2.2.5AGeneralSystemClassification 43

2.3ParadigmShiftsinControlSystemsDesign 44

2.3.1Pre-historyandPrimitiveControl 44

2.3.2Pre-classicalControlParadigm 44

2.3.3GeneralSystemTheoryandthePhilosophicalFoundationsof Model-BasedControl 45

2.3.4Model-BasedDesignParadigm 46

2.3.4.1PhilosophicalDiscussionsonModelPrevalenceinFeedback Control 46

2.3.5ClassicalControlDesign 49

2.3.6ModernControlDesign 50

2.4UncertaintyCombatParadigm 54

2.4.1UncertaintyandPerformanceProblem 54

2.4.2UncertaintyCombat:theRobustControlApproach 56

2.4.3UncertaintyCombat:theAdaptiveControlApproach 57

2.4.4UncertaintyCombat:theSoftComputing-basedControlApproach 59

2.5TheParadigmShiftTowardsData-drivenControlMethodologies 61

2.5.1UnfalsifiedPhilosophyinControlSystemsDesign 64

2.6Conclusions 68 References 69

3UnfalsifiedAdaptiveSwitchingSupervisoryControl 73

3.1Introduction 73

3.2APhilosophicalPerspective 75

3.3PrinciplesoftheUnfalsifiedAdaptiveSwitchingControl 77

3.3.1BasicConceptsandDefinitionsintheUASCMethodology 78

3.3.2TheMainResults 79

3.4TheNon-MinimumPhaseController 87

3.5TheDALPhenomena 88

3.6PerformanceImprovementTechniques 91

3.6.1FilteredCostFunction 91

3.6.2ThresholdHysteresisAlgorithm 92

3.6.3Scale-IndependentHysteresisAlgorithm 93

3.7IncreasingCostLevelAlgorithmsinUASC 95

3.7.1IncreasingCostLevelAlgorithm 97

3.7.2LinearIncreasingCostLevelAlgorithm 98

3.8Time-varyingSystemsintheUASC 101

3.9Conclusion 104

Problems 106

References 108

4Multi-ModelUnfalsifiedAdaptiveSwitchingSupervisory Control 111

4.1Introduction 111

4.2TheMulti-ModelAdaptiveControl 113

4.3PrinciplesoftheMulti-ModelUnfalsifiedAdaptiveSwitching Control 116

4.4PerformanceEnhancementTechniquesintheMMUASC 126

4.4.1DifferentMMUASCCostFunctions 126

4.4.2AdaptiveWindowintheMMUASC 127

4.5Input-constrainedMulti-ModelUnfalsifiedSwitchingControl Design 129

4.5.1Multi-ModelUnfalsifiedConstrainedAnti-WindupControl 130

4.5.2TheFeasibilityProblem 135

4.5.3QuadraticInverseOptimalControl 138

4.5.4Multi-ModelUnfalsifiedConstrainedGeneralisedPredictive Control 141

4.5.5VirtualReferenceSignalintheMMUCGPCScheme 143

4.5.6SwitchingAlgorithmintheMMUCGPC 144

4.6Conclusion 147 Problems 148 References 151

5Data-DrivenControlSystemDesignBasedontheVirtual ReferenceFeedbackTuningApproach 155

5.1Introduction 155

5.2TheBasicVRFTMethodology 156

5.2.1FilterDesign 159

5.3TheMeasurementNoiseEffect 163

5.3.1TheInstrumentalVariableSelection 164

5.4TheNon-MinimumPhasePlantsChallengeintheVRFTDesign Approach 166

5.5ExtensionsoftheVRTFMethodologytoMultivariablePlants 171

5.6OptimalReferenceModelSelectionintheVRFTMethodology 177

5.6.1TheParticleSwarmOptimisationScheme 180

5.7Closed-loopStabilityoftheVRFT-BasedData-DrivenControl Systems 183

5.7.1AnIdentification-BasedApproach 183

5.7.2AnUnfalsification-BasedApproach 184

5.8Conclusions 187

Problems 188 References 190

6TheSimultaneousPerturbationStochastic Approximation-BasedData-DrivenControlDesign 193

6.1Introduction 193

6.2TheEssentialsoftheSPSAAlgorithm 195

6.2.1TheMainTheoreticalResultoftheSPSAAlgorithm 198

6.3Data-DrivenControlDesignBasedontheSPSAAlgorithm 201

6.3.1ThePIDControl 202

6.3.2TheMPCApproach 202

6.4ACaseStudy:Data-DrivenControlofUnder-actuatedSystems 205

6.4.1TheLiquidSloshExample 205

6.4.2TheBallandBeamExample 210

6.5Conclusions 212 Problems 213 References 215

7Data-drivenControlSystemDesignBasedonthe FundamentalLemma 217

7.1Introduction 217

7.2TheFundamentalLemma 218

7.3SystemRepresentationandIdentificationofLTISystems 222

7.3.1EquivalentData-drivenRepresentationsofLTISystems 222

7.3.2Data-drivenState-spaceIdentification 224

7.4Data-drivenState-feedbackStabilisation 225

7.5RobustData-drivenState-feedbackStabilisation 228

7.6Data-drivenPredictiveControl 233

7.6.1TheData-enabledPredictiveControl(DeePC) 235

7.6.1.1Input–OutputDataCollection 235

7.6.1.2StateEstimationandTrajectoryPrediction 235

7.6.1.3TheDeePCAlgorithm 237

7.6.2LTISystemswithMeasurementNoise 239

x Contents

9.3.3FFDLDataModelforNonlinearMultivariablePlants 304

9.4DesignofModel-freeAdaptiveControlSystemsforUnknown NonlinearPlants 304

9.4.1Model-freeAdaptiveControlBasedontheCFDLDataModel 305

9.4.2Model-freeAdaptiveControlBasedonthePFDLDataModel 308

9.4.3Model-freeAdaptiveControlBasedontheFFDLDataModel 310

9.5ExtensionsoftheModel-freeAdaptiveControlMethodologiesto MultivariablePlants 314

9.5.1MFACDesignBasedontheCFDLDataModelforNonlinear MultivariablePlants 314

9.5.2MFACDesignBasedonthePFDLDataModelforNonlinear MultivariablePlants 318

9.5.3MFACDesignBasedontheFFDLDataModelforNonlinear MultivariablePlants 320

9.6ACombinedMFAC–SPSAData-drivenControlStrategy 330

9.7Conclusions 337 Problems 338 References 339

Appendix 341

ANorms 341

BLyapunovEquation 343

CIncrementalStability 343

DSwitchingandtheDwell-time 344

EInverseMoments 346

FLeastSquaresEstimation 349

GLinearMatrixInequalities 351

HLinearFractionalTransformations 353 References 355

Index 357

Preface

Model-basedcontrolsystems.Model-basedcontrolsystemanalysisanddesign approacheshavebeenthedominantparadigmincontrolsystemeducationand thecornerstoneofcontrolsystemdesignfordecades.Thesemethodologies relyonaccuratemathematicalmodelsandassumptionstoachievethedesired systembehaviour.Intheearlydecadesofthelastcentury,despitethetremendous interestinmodel-basedcontrolapproaches,manyPIDcontrollersintheindustry weredesignedbasedonthedata-driventechniqueofZiegler–NicholsPID parametertuning,whichisconsideredthefirstdata-drivencontrolapproach. Later,theadvancedadaptiveandrobustmodel-basedcontroltechniquesevolved tocombattheuncertaintychallengeintheestablishedmodel-basedtechniques. Theseadvancedcontroltechniquessuccessfullycontrolledmanyreal-worldand industrialplants.Yet,bothstrategiesrequiremathematicalmodelsandprior plantassumptionsmandatedbythetheory.

Data-drivencontrolmethodologies.Thelimitationsanduncertaintiesassociatedwithmodelsandassumptions,ontheonehand,andtheemergenceof progressivelycomplexsystems,ontheotherhand,havesparkedaparadigm shifttowardsdata-drivencontrolmethodologies.Theexponentiallyincreasing numberofresearchpapersinthisfieldandthegrowingnumberofcourses offeredinuniversitiesworldwideonthesubjectclearlyshowthistrend.Thenew data-drivencontrolsystemdesignparadigmhasre-emergedtocircumvent thenecessityofderivingofflineoronlineplantmodels.Manyplantsregularly generateandstorehugeamountsofoperatingdataatspecificinstantsoftime. Suchdataencompassesalltherelevantplantinformationrequiredforcontrol, estimation,performanceassessment,decision-makingandfaultdiagnosis. Thisdataavailabilityhasfacilitatedthedesignofdata-drivencontrolsystems.

Intendedaudience.Thisbookisanintroductiontodata-drivencontrolsystems andattemptstoprovideanoverviewofthemainstreamdesignapproachesin thefield.Theselectedapproachesmaybecalledwithcautiontheconventional approaches,notincludingtheapproachesbasedonsoftcomputingtechniques.

Auniquechapterisdevotedtophilosophical–historicalissuesregardingthe emergenceofdata-drivencontrolsystemsasthefuturedominantcontroldesign paradigm.Thischapterwillbeparticularlyappealingtoreadersinterestedin gaininginsightsintothephilosophicalandhistoricalaspectsofcontrolsystem designmethodologies.Conceptsfromthephilosophyofscienceandhistorical discussionsarepresentedtoshowtheinevitableprevalenceofdata-driven techniquesinthefaceofemergingcomplexadaptivesystems.Thisbookcancover agraduatecourseondata-drivencontrolandcanalsobeusedbyanystudent orresearcherwhowishestostartworkinginthefieldofdata-drivencontrol systems.Thisbookwillpresenttheprimarymaterial,andthereadercanperceive ageneraloverviewofthedevelopingdata-drivencontroltheory.Thebook presentationavoidsdetailedmathematicalrelationsandderivationsthatare availableinthecitedtechnicalpapersoneachsubject.However,algorithmsfor easyimplementationofthemethodswithnumericalandsimulationexamples areprovided.Thesoftwarecodesareavailableuponrequestfromtheauthor. Data-drivencontrolisalsoahotresearchtopic;manyfinal-yearundergraduate andpostgraduatestudentsareinterestedinstartingaresearchprojectinits differentareas.Theavailablereadingsourcesarethetechnicalpapersandthe limitednumberofresearchmonographsandbooksonthesubject.However,the technicalpapersareveryspecialisedandinvolvedeepmathematicalderivations. Thelimitednumberofpublishedmonographsandbooksalsospecialiseinspecific subjectareasanddonotprovideageneralintroductionandoverviewofdifferent methodologiesforafirst-timereaderindata-drivencontrol.Theselectedtopics inthisbookcanbeindividuallytaughtinmanydifferentcoursesonadvanced controltheory.Also,foraninterestedresearcherinanyofthecoveredfields,it wouldbebeneficialtolearnaboutthebasicsofotheralternativemethodologies toplanaresearchprogramme.

Prerequisites.Thebookisdesignedforgraduate-levelcoursesandresearchers specialisingincontrolsystemsacrossvariousengineeringdisciplines.Thebook assumesthatthereaderpossessesasolidunderstandingoffeedbackcontrol systemsaswellasfamiliaritywiththeprinciplesofdiscrete-timecontrolsystems andoptimisationproblems.Moreover,abasicunderstandingofsystemidentification,adaptivecontrolandrobustcontrolcanenhancethereader’scomprehension andappreciationofdata-drivencontrolmethodologies.

Overviewofthebook.Thebookisorganisedasfollows.Chapter1introduces boththemodel-basedanddata-drivencontrolsystemdesignapproaches. Itdiscussestheearlydevelopmentsandthecurrentstatusquoofmodel-based controlsystems,aswellasthechallengestheyface.Thechapteralsoexplores adaptiveandrobustcontrolmethodologiesasameanstoovercomesomeofthese

Preface

challenges.Subsequently,thedata-drivencontrolsystemdesignapproachis presented,andthetechnicalaspectsofdifferentdata-drivencontrolschemesare discussed.

Chapter2takesaphilosophicalperspectivetoanalysetheparadigmshiftsin controlsystemdesign.Itpresentsscientifictheory,revolutionsandparadigm shifts,drawingparallelstotheevolutionofcontrolsystemdesignmethodologies. Thehistoricaldevelopmentofcontrolsystemsdesignparadigmsandtheir philosophicalfoundationsisintroduced,andageneralclassificationofcontrol systemsisgiven.Thechapterconcludeswithanexplorationoftheparadigm shiftstowardsdata-drivencontrolmethodologies,withafocusontheinfluence oftheunfalsifiedphilosophy.

Chapters3and4presentdata-drivenadaptiveswitchingsupervisorycontrol andmulti-modeladaptiveswitchingsupervisorycontrol,respectively.ThephilosophicalbackboneofthepresentedmethodologiesisPopper’sfalsification theory,whichisintroducedinadata-drivencontrolcontextbySafonov.Itis showninChapter3thattheunfalsifiedadaptiveswitchingsupervisorycontrol caneffectivelycontrolunknownplantswithguaranteedclosed-loopstability undertheminimumassumptionoftheexistenceofastabilisingcontroller. Althoughseveralclosed-looptransientimprovementtechniquesarepresentedin Chapter3,themulti-modelunfalsifiedadaptiveswitchingcontrolisintroduced inChapter4toensureasuperiorclosed-looptransientperformance.Itisshown thatperformanceimprovementisachievedbyutilisingamodelsettoselect theappropriatecontrollerbasedonthefalsifyingtheory.Theadaptivememory concept,input-constraineddesignproblemsandquadraticinverseoptimal controlnotionarealsodiscussed.

Chapter5presentsthevirtualreferencefeedbacktuningapproach.Itisshown thatbyformulatingthecontrollertuningproblemasacontrollerparameter identificationproblem,adata-basedcontrollerdesignmethodologyisderived. Inthisapproach,avirtualreferencesignalisintroduced,anditisassumedthat thecontrollerstructureisknownapriori.Afterintroducingthebasicconcepts andmethodology,theproblemsofappropriatefilterdesign,measurementnoise, non-minimumphasezerochallenges,closed-loopstabilityandextensionsto multivariableplantsareaddressedinthischapter.

ThesimultaneousperturbationstochasticapproximationoptimisationtechniqueisintroducedandutilisedinChapter6forthedesignofdata-driven controlsystems.Itisshownthatthiscircumventsthenecessityofananalytical closed-formsolutiontothecontroloptimisationproblemsthatrequirealmost exactmathematicalmodelsofthetrueplant.Theessentialsofthetechnique arepresentedforplantswithunknownexactmathematicalmodels.Then,after

selectingacontrollerwithafixedknownstructurebutunknownparameters,by minimisingacostfunction,thecontrollerparametersarederived.Thepresented data-drivencontrolmethodologyisthenappliedtounknown,under-actuated systemsasapracticalcasestudy.

Chapter7presentsaclassofdata-drivencontrollersbasedonWillem’sFundamentalLemma.Itisinitiallyshownthatpersistentlyexcitingdatacanbeused torepresenttheinput–outputbehaviourofalinearsystemwithouttheneedto identifythelinearsystem’smatrices.Thederivedso-calledequivalentdata-based representationsofalineartimeinvariant(LTI)systemaresubsequentlyutilised todesigndata-drivenstate-feedbackstabilisersandpredictivecontrollerscalled Data-enabledPredictiveControl,orDeePCforshort.Resultswithmeasurement noiseandnonlinearsystemsarealsogiveninthischapter.

Chapter8presentsdata-drivencontrollersbasedonKoopman’stheoryandthe FundamentalLemmapresentedinChapter7.ThefundamentalsofKoopman’s theoryarebrieflyreviewedfordata-drivencontrol.Itisshownthatnonlinear dynamicalsystemsarepresentedbyhigherdimensionallinearapproximations. Themainnotionsof lifting or embedding andtheeffectivetoolsof(extended) dynamicmodedecompositionsareintroducedandadata-drivenKoopman-based predictivecontrolschemeispresentedbyincorporatingWillem’sFundamental LemmaofChapter7.Arobuststabilityanalysisisprovided,andtheresultsare finallyappliedtotheACUREXsolarcollectorfield.

Themodel-freeadaptivecontroldesignisadata-drivencontroldesignapproach basedondynamiclinearisationmethodologiesandispresentedinChapter9. Thethreemaindynamiclinearisationsdiscussedinthischapterareshownto capturethesystem’sbehaviourbyinvestigatingtheoutputvariationsresulting frominputsignals.Thesedatamodelsareutilisedforcontrollerdesignandtheir virtualnaturemakestheminappropriateforothersystemanalysispurposes.Also, inChapter9,thevirtualdatamodelresultsandtheircorrespondingmodel-free adaptivecontrollersareextendedtomultivariableplants.

Somepreliminaryconceptsthatareusefulforthechaptersarepresentedin theAppendix.Thechaptersareaccompaniedbyproblemsetsthatprovidereaderswiththeopportunitytoreinforcetheirunderstandingandapplytheconcepts discussed.Asolutionmanualisalsoprovidedforinstructorsteachingaclasson data-drivencontrolusingthisbookbycontactingtheauthor.

October2023

● AliRezaei,mymaster’sstudentintheAdvancedControlLaboratory,forhis valuablecontributionstoChapter9andtheefforthededicatedtosimulations throughoutthebook.

Finally,Iwouldliketoexpressmysincereappreciationtotheanonymous reviewerswhoprovidedinvaluablefeedbackduringthereviewprocessof thisbook,andspecialthankstoWiley-IEEEPressfortheirexceptional professionalism,dedication,industryknowledgeandseamlesscoordination thatexceededmyexpectations.Lastbutnotleast,Iamalsogratefultomyfamily fortheircollaborationandsupport,allowingmetodedicatemostofmyholidays, weekendsandeveningstocompletingthisbook.

ListofAcronyms

ASSCAdaptiveswitchingsupervisorycontrol

BIBOBounded-inputbounded-output

CFDLCompact-formdynamiclinearisation

CSPConcentratedsolarpower

DALDehghani–Anderson–Lanzon

DDKPCData-drivenKoopmanpredictivecontrol

DeePC Data-enabled Predictive Control

DFTDiscreteFouriertransform

DMDDynamicmodedecomposition

EDMDExtendeddynamicmodedecomposition

ETFEEmpiricaltransferfunctionestimate

FFDLFull-formdynamiclinearisation

GLAGeneralisedLaplaceanalysis

GPCGeneralisedpredictivecontrol

ICLAIncreasingcostlevelalgorithm

IFACInternationalFederationofAutomaticControl

LFTLinearfractionaltransformation

LICLALinearlyincreasingcostlevelalgorithm

LLCLinearisationlengthconstant

LMILinearmatrixinequality

LQGLinearquadraticGaussian

LQRLinearquadraticregulator

LSTMLongshort-termmemory

LTILineartime-invariant

MFACModel-freeadaptivecontrol

MMUASCMulti-modelunfalsifiedadaptiveswitchingcontrol

MMUASC-RMMUASCwithresettime

MMUCGPCMulti-modelunfalsifiedconstrainedGPC

MPCModelpredictivecontrol

1.1.1TheEarlyDevelopments

Theadventofmodelsincontrolsystemstheoryanddesignisrootedintheseminal paperofMaxwell OnGoverners [1].NorbertWiener,inintroducingtheword cyberneticsinRef.[2]describestheMaxwellpaperas‘… thefirstsignificantpaper onfeedbackmechanismsisanarticleongovernors,whichwaspublishedbyClerk Maxwellin1868’andinRef.[3],Maxwellisrecognisedasthe‘fatherofcontrol theory’.TheMaxwellmagicwastointroducedifferentialequationsinmodelling thebehaviouroftheflyballgovernorfeedbackcontrolsysteminventedbyJames Wattin1788.Thisground-breakingcontributionbyMaxwellintroducedthe conceptofmathematicalmodellinginthestabilityanalysisofaclosed-loopcontrolsystem,anideathatsoonfoundmanyapplicationsandadvocatesandsolved manyuntilthenunsolvedstabilityanalysisproblems.Thedifferentialequations encounteredintheflyballgovernormodelwerenonlinear.Bylinearisingthese nonlinearequations,Maxwellmanagedtointroducethenotionsofwhatistoday calledrealpoles,imaginarypolesandthesignificanceofpolepositionintheright halfplane.Thismodel-basedapproachtotheanalysisofacontrolsystemthrough thedifferentialequationsofmotionwasperformedforthefirsttimeinthehistory ofcontroltheory.Hence,itisplausibletointroduceMaxwellasthepioneerofthe model-basedcontroltheory.

Intheearlytwentiethcentury,controlsystemdesignmethodologiessuchas the classicalcontrol techniquesinitiatedbyBode,Nyquist,EvansandNichols wereallmodel-basedapproachestocontroldesignsincethetransferfunction knowledgeofthecontrolledsystemisrequired.Thetransferfunctioncanbe derivedfromasetofalgebraicanddifferentialequationsthatanalyticallyrelate inputsandoutputs,oritcouldbeobtainedfromsimpletestsperformedonthe plantwiththeassumptionsoflinearityandtimeinvariance.Laterinthe1960s,

AnIntroductiontoData-DrivenControlSystems,FirstEdition.AliKhaki-Sedigh. ©2024TheInstituteofElectricalandElectronicsEngineers,Inc.Published2024byJohnWiley&Sons,Inc.

Kalmanintroducedthemodel-basedstate-spaceapproachthatwasmoredetailed andmathematical.

Theonlynotabledata-driventechniqueofthefirsthalfofthelastcentury istheZiegler–Nicholsproportional-integral-derivative(PID)parametertuning proposedinRef.[4],whichbecameawidelyusedcontroltechniqueinthe industry[5].ItisstatedinRef.[4]that“Apurelymathematicalapproachtothe studyofautomaticcontroliscertainlythemostdesirablecoursefromastandpoint ofaccuracyandbrevity.Unfortunately,however,themathematicsofcontrolinvolves suchabewilderingassortmentofexponentialandtrigonometricfunctionsthatthe averageengineercannotaffordthetimenecessarytoplowthroughthemtoasolution ofhiscurrentproblem.”Thisstatementfromtheeminentcontrolengineersof thattimeshowsthelong-lastinginfluenceofmathematicalmodel-baseddesign techniquesonthecontrolsystemsdesigncommunity.Indescribingtheirwork, theyimmediatelystatethat‘thepurposeofthispaperistoexaminetheactionof thethreeprincipalcontroleffectsfoundinpresent-dayinstruments,assignpractical valuestoeacheffect,seewhatadjustmentofeachdoestothefinalcontrol,andgive amethodforarrivingquicklyattheoptimumsettingsofeachcontroleffect.The paperwillthusfirstendeavortoanswerthequestion:“Howcanthepropercontroller adjustmentsbequicklydeterminedonanycontrolapplication? ”’Thisstatement canenlightenaspectsofthephilosophyofthedata-drivencontrolsystemsthat evolvedinthelatetwentiethcenturyonwards.

1.1.2Model-basedControlSystemStatusQuo

Model-basedcontrolsystemdesignisthedominantparadigmincontrolsystem educationanddesign.Thisapproachisbasedonderivedanalyticalmodelsfrom basicphysicallawsandequationsormodelsfromanidentificationprocess.Models areonlyapproximationsofrealityandcannotcaptureallthefeaturesandcharacteristicsofaplantundercontrol.High-frequencyun-modelleddynamicsarean example,asinroboticandspacecraftapplicationswheretheresidualvibration modesarenotincludedinthemodel[6].Thestructureofamodel-basedcontrol systemisshowninFigure1.1.Inthecaseofadaptivecontrolstrategies,theapproximateplantmodelisupdatedusingtheinput–outputdata.

AsisshowninFigure1.1,theplantmodel,derivedfromfirstprinciplesoridentifiedfromplant-measureddata,isusedtodesignafixed-ordercontrollersatisfying thespecifiedclosed-looprequirements.However,thedesignedcontrollerdoesnot necessarilysatisfythepre-definedrequirementswhenconnectedtotherealplant, andtheclosed-loopperformanceislimitedbythe modellingerrors.Modelling errorscanhavemanyrootcauses,suchasun-modelleddynamics,unknownor varyingplantparametersresultingfromchangingoperatingpoints,equipment ageingorfaultsandinappropriatemodelstructures.

Figure1.1 Thestructureofamodel-basedcontrolsystemdesign.

Modellingerrorsduetoun-modelleddynamicsarejustifiedinthestandardpracticeofmodel-basedcontroldesignwhenthesystemiscomplexandisofa high order,andalow-ordermodelisemployedtofacilitatethecontroldesign.Onthe otherhand,therecanbeatendencytoincreasethemodelordertofindasuitable model.ItisshowninRef.[7]thatthisisnotgenerallytrueifthemodelhastobe usedforcontroldesign.Infact,the order ofarealsystemisabadlydefinedconcept, andeventhemostaccuratemodelsareonlyanapproximationoftherealplant. Intherealworld,a full-ordermodel doesnotexist,andanydescriptionis,bydefinition,anapproximation[7].Model-basedcontroldesigncanonlybeemployedwith confidenceinreal-worldapplicationsifthemodelstructureisperfectlyknown.

Theissueofmodel-basedcontrolsystemdesignandtheparadigmshiftstoand frommodel-basedapproachesisfurtherelaboratedinChapter2.

1.1.3ChallengesofModelsinControlSystemsDesign

Theintroductionofthestate-spaceconceptbyKalmanin1960,togetherwiththe newlyestablishednotionofoptimality,resultedinaremarkabledevelopmentof model-basedcontroldesignmethods.BeforeKalman’sstate-spacetheory,most ofthecontroldesignwasbasedontransferfunctionmodels,asisintheBode andNyquistplotsortheroot-locusmethodandtheNicholschartsforlead–lag compensatordesign.

Inthecaseswherereliablemodelswereunavailable,orinthecaseofvarying parametersandchangingoperatingconditions,theapplicationofthemodel-based controlwasseverelylimited.Inthemid-1960s,thesystemidentificationstrategy evolved.TheproposedMaximumLikelihoodframeworkfortheidentificationof

input–outputmodelsresultedinthepredictionerror-typeidentifiers.Theadvent ofidentificationtheorysolvedtheproblemofcontrollingcomplextime-varying plantsusingmodel-basedcontroldesignmethodologies.

Initially,controlscientistsworkingontheidentificationmethodsaimedat developingsophisticatedmodelsandmethodologieswiththeelusivegoalof convergingtothe truesystem,undertheassumptionthatthetruesystemwas inthedefinedmodelset.Later,theyrealisedthatthetheorycouldbestachieve anapproximationofthetruesystemandcharacterisethisapproximationin termsofbiasandvarianceerrorontheidentifiedmodels.Finally,system identificationwasguidedtowardsacontrol-orientedidentification.Inallthe modellingstrategies,modellingbyfirstprinciplesorbyidentificationfromdata, modellingerrors areunescapable,and explicitquantification ofmodellingerrorsis practicallyimpossible.Hence,themodellingstrategiescannotprovideadequate practicaluncertaintydescriptionsforcontroldesignpurposes.Therefore,thefirst modellingprinciplegiveninRef.[8],thatarbitrarilysmallmodellingerrorscan leadtoarbitrarilybadclosed-loopperformance,isseriouslyalarmingforcontrol systemsdesigners.

Applicationofthe certaintyequivalenceprinciple (seeChapter2)wasbasedon theearlyoptimisticassumptionthatitispossibletoalmostperfectlymodelthe actualplantandthemathematicalmodelobtainedfromthefirstprinciplesor byidentificationfrominput–outputdataisvalidenoughtorepresentthetrue system.However,applicationsinreal-worldproblemsdidnotmeettheexpectationsofthecontrolscientistsanddesigners.Therefore,anobviousneedprevailed toguaranteeclosed-loopstabilityandperformanceinthemodel-basedcontrol designapproaches.Thisledtothedevelopmentofthemodel-basedapproaches offixed-parameterrobustcontrolandadaptivecontrolsystemdesign[9].

Themathematicalmodelsderivedfromthephysicallawshavebeeneffectively usedinpracticalapplications,providedthatthefollowingassumptionshold:

● Accuratelymodeltheactualplant.

● Prioriboundsonthenoiseandmodellingerrorsareavailable.

Also,identificationmodelshavebeenemployedinmanypracticalapplications. Theidentifiedmodelcancapturethemainfeaturesoftheplant,providedthat

● Compatibilityoftheselectedmodelstructureandparameterisationwiththe actualplant’scharacteristicsisassumed.

● Theexperimentdesignisappropriate;thatis,forcontrolproblems,theselection oftheinputsignalisinaccordancewiththeactualplant’scharacteristicsorthe persistenceofexcitation(PE)condition.

Itisimportanttonotethateveninthecaseofanaccuratelymodelledplant, iftheassumptionsabouttheplantcharacteristicsarenotmet,themathematical

thecontroller.Thisisthe data-driven approach,whichappearedattheendofthe 1990s.Manyplantsregularlygenerateandstorehugeamountsofoperatingdataat specificinstantsoftime.Suchdataencompassalltherelevantplantinformation requiredforcontrol,estimation,performanceassessment,decisionmakingsand faultdiagnosis.Thisfacilitatesthedesignofdata-drivencontrolsystems.Theterm data-drivenwasinitiallyusedincomputerscienceandhasenteredthecontrolsystemscienceliteratureinthepasttwodecades.Althoughdata-drivencontrolwas actuallyintroducedinthefirstdecadesofthetwentiethcentury(seeChapter2), theapproachwasnotcalleddata-drivenatthattime.Thedata-drivencontroland data-basedcontrolconceptsaredifferentiatedinRef.[11].Also,Ref.[12]haselaboratedonthedifferencebetweendata-basedanddata-drivencontrol.Itisstated inRef.[12]that‘data-drivencontrolonlyreferstoaclosedloopcontrolthatstartingpointanddestinationarebothdata.Data-basedcontrolisthenamoregeneral termthatcontrollersaredesignedwithoutdirectlymakinguseofparametricmodels, butbasedonknowledgeoftheplantinput-outputdata.Sortedaccordingtotherelationshipbetweenthecontrolstrategyandthemeasurements,databasedcontrolcan besummarizedasfourtypes:post-identificationcontrol,directdata-drivencontrol, learningcontrol,andobserverintegratedcontrol.’

Themainfeaturesofthedata-drivencontrolapproachescanbecategorisedas follows:

● Controlsystemdesignandanalysisemployonlythemeasuredplant input–outputdata.Suchdataarethecontrollerdesign’sstartingpoint andendcriteriaforcontrolsystemperformance.

● Noprioriinformationandassumptionsontheplant’sdynamicsorstructureare required.

● Thecontrollerstructurecanbepredetermined.

● Theclosed-loopstability,convergenceandsafeoperationissuesshouldbe addressedinadata-drivencontext.

● Adesigner-specifiedcostfunctionisminimisedusingthemeasureddatato derivethecontrollerparameters.

Thestructureofadata-drivencontrolsystemdesignisshowninFigure1.2. Severaldefinitionsfordata-drivencontrolareproposedintheliterature.The followingdefinitionfromRef.[11]ispresented.

Definition1.1

Data-drivencontrolincludesallcontroltheoriesandmethodsin whichthecontrollerisdesignedbydirectlyusingonlineorofflineinput–output dataofthecontrolledsystemorknowledgefromthedataprocessingbutnot anyexplicitinformationfromamathematicalmodelofthecontrolledprocess andwhosestability,convergenceandrobustnesscanbeguaranteedbyrigorous mathematicalanalysisundercertainreasonableassumptions.

Turn static files into dynamic content formats.

Create a flipbook
Issuu converts static files into: digital portfolios, online yearbooks, online catalogs, digital photo albums and more. Sign up and create your flipbook.
An introduction to data-driven control systems ali khaki-sedigh - Get instant access to the full ebo by Education Libraries - Issuu