基于神經(jīng)網(wǎng)絡(luò)的機(jī)械手控制方法研究.pdf

1、北京工業(yè)大學(xué)碩士學(xué)位論文基于神經(jīng)網(wǎng)絡(luò)的機(jī)械手控制方法研究姓名：李明申請(qǐng)學(xué)位級(jí)別：碩士專業(yè)：模式識(shí)別與智能系統(tǒng)指導(dǎo)教師：李建更于乃功20070501北京工業(yè)大學(xué)工學(xué)碩士學(xué)位論文ABSTRACTNeuralnetworksaI℃extensivelyusedinmodeling、identificationandcontrolofsystembecauseoftheirpowerfulandexcellentabilityofdatapro

2、cessingandapproximatinganynonlinearftmetionInthepapertwokindsoftypicalneuralnetworks—feedforwardneuralnetworkandrecurrentneuralnetworkareanalyzedandimprovedtespectivelyapplytobotmanipulatorsinversemodelcontrolandpathtrac

3、kingExperimentshowsagoodresultSomemainworksofthispaperarcasfollows：(1)Theslruetureelmmeteristies，laainingalgorithmandapplicationinrobotmanipttlators’controloffeedforwardandree哪tneuralnetworks黜respectivelyanalylxtandsumma

4、rizedThengavetheproblemofthesetwoapplicationsinrobotmanipulators’contr01BasedOntheaboveanalysisakindoffastspeelandheuristicalgorithmandakindofimprovedneuralnetworkslzuetureofJordannetwork玳respectivelyintroduced(2)Inthepa

5、pef’aslrategythatcombinedneuralnetworkandinversemodelcontrolisproposedtosolvetheproblemresultedfromtheinverseclynamiemodelsnonlinearcharacteristicandp鋤鋤ctcrsltulcenaintyin2DOFrobot蛐contr01Atthe蛐etime，becauseBPalgorithmha

6、sflOIltledisadvantagesofslowconvergenceandgettingeasilyintolocalminimaatthelasttimeoftrainingafast印ccdandhcul_isticalgorithmisintroducedt0lrainamultiplefeedforwardneuralnetworkandthenestabfishtheinversemodelofrobotlllman

7、dcontrolitThesimulationresultsshowthevalidityofcontrolslzategyandthefast印ecdconvergenceofalgorithm(3)BasedOntheree哪tneuralnetworks’goodabilityofmappingnonlin船taymmicsystemapaIlhtrackingcontrolsystembasedOn卸improvedJordan

8、networkisdesignedtocontrolrobotmanipulatorApproprialenetworkshud【IⅡeandtrainningalgritiara讎selectedaccordingtothefeatureofmanipulatorkinematicsFinallythesimulationsshowafastspeedandrealtimeeontroningp曲皿an∞KEYWORDS=RobotM