基于DE算法的DRNN网络非线性系统辨识研究

《基于DE算法的DRNN网络非线性系统辨识研究》由会员上传分享，免费在线阅读，更多相关内容在学术论文-天天文库。

1、西南交通大学硕士研究生学位论文第1I页AbstractSystemidentificationisallimportantpartofcontroltheoryresearch，andnonlinearsystemidentificationisalwaysthefocusanddifficultyinthisfield．Astheartificialneuralnetworkshavebeenanewfortheidentificationofproposed，itopensupwaycomplexnonlinearsystem．Inmanykindsofneuralnetworks，Ba

2、ck-Propagation(BP)neuralnetworkisthemostwidelyused．Asamultilayerf．eed—forwardneuralnetwork，BPneuralnetworkhasmanyproblemsanddisadvantages．Diagonalrecurrentneuralnetwork(DRNN)isadynamicnetworkwithfeed-backparts．Bystoringitsintemalstatustoreflectthedynamiccharacteristics，DRNNismoresuitabletofortheide

3、ntificationofnonlineardynamicsystem．Thelearningalgorithmisthecoreissuesinthesystemidentification．Traditionally,theBPalgorithmwhichisalwayscalledgradientdescentalgorithmCannotmeetOUrneedsofidentificationaccuracyandconvergencerate．Inordertosolvethisproblem，theimprovementandinvestigationofvariousalgor

4、ithmsarecarriedout，andtheirachievementsareprovedtobeeffective．Soastofurtherenhancetheidentificationandaccuracyconvergencerate，thisthesismakesanin—depthexplorationandinvestigationofDRNNnonlinearsystemidentificationbasedondifferentialevolution(DE)．Inthefirstpartofthethesis，thepresentsituationsofthesy

5、stemidentificationwithneuralnetworksanddifferentialevolutionaresummarizedandanalyzed．Theproblemsneedtobesolvedarealsopointedout．besidestheideasandsignificanceoftheinvestigationaregiven．ThebasicmodelandsystemidentificationprincipleofDRNNareintroduced．BPalgorithm，improvedBPalgorithmandgeneticalgorith

6、m(GA)arerespectivelyusedastheintheidentificationoftwotypicalnonlinearsystems．Throughthematlablearningalgorithmsimulation，theperformancesofthosethreealgorithmsarecompared．TheexperimentresultsshowthatthenonlinearsystemidentificationofGAismuchbetterthantheothertwoalgorithms．Afterthat，inordertofurthere

7、nhancetheidentificationaccuracy,anewlearningalgorithm—DEalgorithmisintroducedandusedasthelearningalgorithminDRNN．ThreedifferentstrategiesofDEarediscussedandanalyzed．Accordingtothemeritsofthosethreestrategie