搜索
32120112()JournalofQingdaoUniversityofScienceandTechnology(NaturalScienceEdition)Vol.32No.1Feb.2011:16726987(2011)010085051,1*,2,1(1.,150001;2.,150001):针对非线性强耦合的空间漂浮基柔性机器人,提出了一种启发式学习算法的神经网络的前馈控制策略首先通过拉格朗日法和假设模态方法建立了漂浮基柔性空间机器人的动力学模型,然后采用两个神经网络及一个PID控制器来构建前馈在线学习控制系统,其中一个神经网络充当前馈控制器,另一个神经网络通过学习逆动态模型来为前馈控制器提供在线学习参数,而PID控制器主要作为辅助补偿控制器该控制策略不是在PID控制器的指导下进行学习,且无需预先的离线学习,因而学习精度更高,且减少了对学习样本选择不当的影响,采用固定中心参数,而扩展宽度采用启发式关系确定,网络权值采用改进的最优准则算法进行调整的算法来实现快速学习能力,具有较好的实时性仿真结果证明了所提方案的有效性:RBF神经网络;柔性空间机器人;启发式学习;PID控制:TP242:A:20100903:(CAST09C01).:(1962),,.*.HeuristicLearningAlgorithmofNeuralNetworkControlforSpaceFlexibleManipulatorsQINaiming1,ZHANGWenhui1,MAJing2,GAOJiuzhou1(1.HarbinInstituteofTechnology,SchoolofAerospace,Harbin150001,China;2.NortheastAgricultureUniversity,CollegeofEngineering,Harbin150001,China)Abstract:Accordingtononlinearstrongcouplingoffreefloatingspacerobotwithflexiblemanipulators,aneuralnetworkfeedforwardcontrolschemebasedonheuristiclearningalgorithmisproposed.Firstly,thedynamicsoffreefloatingspaceflexiblerobotisestablishedbytheLagrangeandassumedmodesmethods.AndthentwoneuralnetworkcontrollersandonePIDcontrollerareusedtoconstructfeedforwardonlinelearingcontrolsystem.Oneneuralnetworkcontrollerisusedasafeedforwardcontroller,andtheotherisappliedtostudyinversemodeloftherobotandprovideonlinelearningparametersforthefeedforwardcontroller.ThePIDcontrollerisanassistantcontroller.Thepropsedcontrolstrategycanimprovethecontrolaccuracyandtheasymptoticconvergenceoftrackingerror,withoutthehelpofPIDandofflinelearning.Theneuralnetworkparameterscanadjustadaptivelybyadoptingtheheuristiclearningalgorithm.Thelearningspeedoftheneuralnetworkisenhancedgreatlyandhasbetterrealtimeproper()32ty.Finally,Thesimulationresultshaveshownthatthepresentedstrategyiseffective.Keywords:RBFneuralnetwork;flexiblespacerobot;heuristiclearning;PIDcontrol,,[13],;,,,,[48],,,[910],,,,,[56],,,[78],,,,,,,,21PID,1,,PIDPID,,,,,,,1,1OXY(XY)B0,B1B21Fig.1ModelofspacerobotwithflexiblemanipulatorBi(i=0,1,2)Oixiyix1B1,x2B2,O0B0OB0,Oi,(i=1,2)Bi-1BiB0m0J0;B1m1J1;B2,EIB2,,,,,EulerBernoulli,u:u(x2,t)=ni=1i(x2)qi(t),i=1,2,,n,(1),i(x2)i,qi(t)i(x2),n,,861:(OXY)O,rc=rc=0,1,2,q1q2,Ti,TT=T0+T1+T2,(2)12,,,U,U=12EIl202u(x2,t)x22dx2,(3)L=T-U,(4)ddtdLdqi-dLdqi=Qi,i=1,2,,n,(5)L,qi,Qiqi,(3)(4),,:Dq+H=,(6)=[12]T,q=[q1q2]TD44H41=[1200]T2,d=[TdqT]T,=[TqT]T,e=d-=[eT0],e=d-(6),=D(d+Ke+Kde)+H,(7)KpKd,,D^,H^,=D^(d+Ke+Kde)+H^(8)(8)(7),e+Kde+Kpe=D^-1[D+H],(9)D=D-D^,H=H-H^,,RBF,,2PID,,PID22Fig.2FuzzyneuralnetworkInversemodelcontrolsystem:1)RBF,,PID2)RBF,RBF3)PID,,,PID,RBF,PID,PIDPD=Kde+Kpe(10)(6):=Dq+H=F(,,,q,q,q)(11)87()32RBFNM=N(,,,q,q,q,W,c,)(12)RBFNN=F(d,d,d,q,q,q,W,c,),(13)WRBF,c,=PD+NN,(14)RBFEE=-NM(15),,,0^y,ydij=ij{x(n),ck,k},:y^(n)y^(n)y^(n)=11121N21222N1N2NNNw1w2wN(16)RNNRBF^y=w(17),J,J(w)=12E2,(18)J(w)J(w)(w)=0,(19)-Tyd+Tw=0,(20)w=(T)-1Tyd=+yd(21)+(21),,,RBFNN,,:=dmaxk,(22)dmax,k{x,ck}=exp(-Kd2maxx-ck2)(23)RBF,,,31l0=05m,l1=l2=15m,a=1m;m0=40kg,m1=2kg,=05kgm-1;EI=300Nm2I0=25kgm2,I1=05kgm21d=2d=sin2tPID,PDKpKdPD:Kp=diag(20,20)Kd=diag(30,30);:(0)=0,1(0)=2(0)=10;1(0)=2(0)=03634PID5631(PID)Fig.3Trackingcurvesofjoint1(PIDalgorithm)881:34,PID,PIDRBF,,RBF,,PD,,,,,RBF,:,,4,,2PID,,,PIDPID,,,,,,,[1]AmorJnifene,WilliamAndrews.Experimentalstudyonactivevibrationcontrolofasinglelinkflexiblemanipulatorusingtoolsoffuzzylogicandneuralnetworks[J].IEEETransactionsonInstrumentationandMeasurement,2005,54(3):12001208.[2]DubowskyS,PapadopoulosEG.Thekinematics,dynamicsandcontroloffreeflyingspaceroboticsystems[J].IEEETransonRoboticsandAutomation,1993,9(5):531543.[3]FarzanehAbdollahi,TalebiHA,RajnikantVPatel.Astableneuralnetworkbasedobserverwithapplicationtoflexiblejointmanipulators[J].IEEETransonNeuralNetworks,2006,17(1):118129.(下转第94页)89()32,,:p=(lusex600)/ni=1area(i),:9983%,9782%,9597%4,,,:,,,,,,[1]GerhardWscher,HeikeHauner,HolgerSchumann.Animprovedtypologyofcuttingandpackingproblems[J].EuropeanJournalofOperationalResearch,2007,183:11091130.[2]AndreasBortfeldt.Ageneticalgorithmforthetwodimensionalstrippackingproblemwithrectangularpieces[J].EuropeanJournalofOperationalResearch,2006,172:814837.[3],,.[J].:,2004,25(5):452456.[4]IoriM,MartelloS,MonaciM.Metaheuristicalgorithmsforthestrippackingproblem[J].OptimizationandIndustry,2003,78:159179.[5]MarcoDorigo,ThomasSttzle.[M].,.:,2007:63115.[6].[M].:,2005:100117.[7]LevineJ,DucatelleF.Antcolonyoptimizationandlocalsearchforbinpackingandcuttingstockproblems[J].JournaloftheOperationalResearchSociety,2004,55:705716.[8],,.[J].,2006,18(7):948951.(责任编辑姜丰辉)(上接第89页)[4],贠超,陈力.柔性臂漂浮基空间机器人建模与轨迹跟[J].,2007,29(1):9296.[5],,,.[J].,2008,29(6):18651869.[6]GreenA,SasiadekJZAdaptivecontrolofaflexiblerobotusingfuzzylogic[C]//AIAA,AmericainstituteofAeronauticsandAstronautics,2004:110.[7],贠超,陈力.漂浮基空间机器人及其柔性影响下逆模[J].,2007,28(6):15101514.[8],.[J].,2008,29(6):172177.[9]YoshidaK,HashizumeK,AbikoS.ZeroreactionmaneuverflightvalidationwithETSVIIspacerobotandextensiontokinematicallyredundantarm[C]//ProceedingsoftheIEEEInternationalconferenceonRoboticsandAutomation,Piscataway,USA:IEEE,2001:441