基于梯度的雙率采樣數(shù)據(jù)系統(tǒng)辨識(shí)

1、江南大學(xué)碩士學(xué)位論文基于梯度的雙率采樣數(shù)據(jù)系統(tǒng)辨識(shí)姓名：陳曉明申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：控制理論與控制工程指導(dǎo)教師：丁鋒20080601AbstractAbstractMultiratesystemsareabundantinindustry；forexample，manysoft—sensordesignproblemsarerelatedtomodeling，parameteridentncation，orstateestim8tio

2、ninVolVingmultiratesystemsTheconVentlonalsampledatasystems，whoseoutputsaresampledatthesameratesasthecontrolupdatingrates，arecalledsingl爭(zhēng)ratesystemsInmanyindustryapplications，theoutputsaresampledatsloⅥ傖rratesthanthecontro

3、lupdatingrates，mostlyduetohardwarelimitationInthiscase，therewillbetwoorevenmoredi行erentoperationratesinasystemSuchsystemsarenamedasdualrate(multirate)systemsTherefore，seachingforefI＆tiveparameterdeEnicationofdualrate(mul

4、tirate)sysLemsisnotonlysignifjcantintheory，butalsopotentiallyValuableinapplicationsThisthesisisbaSedontheProject“StudyofModellingandIdenti矗cationofaClaSsofNonlinearSystems(TheNa乞ionalNatureScienceFbundationofChina)”，andr

5、esearchesparameteridentincationofdual—ratesystemsBasedonsomeexistingidentificationmethodsofdualrate(multirate)systems，thethesismainlyfocusesontheappropriatemodelthatcanbeidentifiedwithdua1一rateinput—outputdataByusingthep

6、olynomialtransformationtechnique，newalgorithmsofparameterident侑cationofdualratesystemsareobtained，andresultsarepresentedaSfollows1Byusingpolynomialtransf6rmationtechniquetotransformARmodelwjthmissingobserVa尸tiondataintoA

7、I氣MAmodelthatcanbeidentifiedwithdualrateinputoutputdatadirectly，arecursiveexten￡IedstochasticgradientidentincationmethodfortheARMAsystemwithmissingobservationdataisproposedThestochasticgradientalgorithmhaslowcomputationa

8、leHort，butltsconvergenceisrelativelyslowInordertoimprovethetrackingperformanceofthestochas—ticgradientalgorithm，aforgettingf如torisintroducedtogetthepecursiVeextendedstochaSticgradientalgorjthmw砒laforgetting‰torAssumethat

9、thesystemjsconsistentlyexciLed，byapplyingthestochasticproeesstheoremandthemartingaleconvergencetheorem，、)l，ecanproVe化attheparameterestimateserrorsareconslstentlyconvergetozeroFlnally，wet|esteff＆tl、傖nessofthealgorithminvo

10、lvedbyasimulationexample2Studyingidentincatlonmethodsfordua】一ratesampled—dataARXmodelsByusingpolynomialtransformationtechnique，dualratesampleddataARXmodelcanbetransformedintothedualrateARMAXmodelthatcanbeidenti矗edwithdua

11、1rateinputoutputdatadirectly，andaforgettingfactorbasedFecursiVeextendedstochasticgra_dientidenti行cationmethodisproposedAlso，assumethatthesystemsarepersistentexcitat沁n，weprovethatthedualra托parameterestimatemethodsareconsi

12、stentlyconvergentFinally，thesimulationexampleindjcatethatparameterestimationofthealgorithmproposedissatisfying3WhennoisemodelsareAR，MAandARMA，respectively，westudyidentmcationmethodsfordualratesampleddatasystemswjththecol