Bp神经网络的Matlab实现

201010Oct.,2010315JournalofXiangnanUniversityVol.31No.5:2010-08-06:安徽建筑工业学院青年科研基金专项项目:石云(1980-),女,安徽庐江人,讲师,研究方向:神经网络.BpMatlab石云(,230601):介绍Bp神经网络以及利用Matlab神经网络工具箱构造Bp网络的方法,并阐述实现网络的关键问题,包括网络设计流程学习规则选择权重和阈值的训练与网络仿真.给出具体应用实例,构造一个典型三层结构的神经网络实现了具有函数逼近功能的Bp网络设计.:Bp神经网络;Matlab仿真;函数逼近:TP389.1:A:1672-8173(2010)05-0086-030(BackPropagationNet-work,Bp),,,.,Bp,,[1].,Bp,,,,,.MATLAB(NeuralNetworkToolbox,NNbox),.,MATLAB,,,,MATLAB,.图1典型Bp网络模型1BpBp,.1BP,,,.:(),,.(),,.Bp.,,,.,,,,,.,,,.862MatlabMatlab,,,,,[2].Matlab,Bp,Bp.2.创建Bp神经网络网络初始化网络训练仿真图2NNbox的应用Bp:MatlabNNboxnewff()newff,:net=newff(PR,Si,{TFi},BTF,BLF,PF)PRRx2;Sii;TFii;BTF!,;BLF!,;BF!,.newffBp,,0.,init(),:net=init(net).,,,.[net,tr]=train(NET,P,T),PT,NETnewff,net,tr(epochsperf).trainnewff.NNbox,,[3].Matlabsim().sim()y=sim(net,p).y;net;p,p,;p,.3BpMatlabMatlab3.,Bp.Bp,.L,,,L.L[4]l=m+n(1)l=0.43mn+0.12n2+2.54m+0.77n+0.35+051(2)mn.21:-0.96-0.577-0.07290.3770.6410.660.4610.1336-0.201-0.434-0.5-0.393-0.16470.09880.30720.3960.34490.1816-0.0312-0.2183-0.3201.,1,2.BPMATLABm:P=-1:0.1:1;T=[-0.96-0.577-0.07290.3770.6410.660.4610.1336-0.201-0.434-0.5-0.393-0.16470.09880.30720.3960.34490.1816-0.0312-0.2183-0.3201];plot(P,T,∀r+∀);net=newcf(minmax(P),[1,2,1],{∀logsig∀,∀tansig∀,∀purelin∀},∀traingdx∀);%net.trainParam.epochs=6000;%net.trainParam.goal=9.5238e-004;%sse=0.02%net.trainParam.lr=0.15;87图3仿真设计流程图[net,tr]=train(net,P,T);%Y=sim(net,P);%plot(P,Y,∀r-∀)holdplot(P,T,∀r+∀);holdoff4.#figure1∃:#+∃,#-∃.1.8504,0.012.#TrainingwithTRAINLM∃,,6000,,1080,0.02,epochs=1084,.,,.图4仿真结果4MATLABBp,,,,.:[1]巨军让,卓戎.Bp神经网络在MATLAB上的方便实现[J].新疆石油学院学报,1999,11(2):42.[2]楼顺天,施阳.基于MATLAB的系统分析与设计[M].西安:西安电子科技大学出版社,2002:23.[3]M.T.Hagan,H.B.Demuth.NeuralNetworkDesig-n.PWSpublishingcompany[J].Boston,1996:67.[4]高大启.有教师的线性基本函数前向三层神经网络结构研究[J].计算机学报,1998,21(1):80.(下转第111页)88StudyofArtQualityforLatinDancingProfessionalAthleteinPhysicalEducationDanceLIShan(DepartmentofPublicCourse,XingnanUniversity,Chenzhou423000,China)Abstract:Thisarticleasksforhelpofdancearttheoriesandotherrelatedcourtesies,combiningthesportsdancefeaturesanditsexperience,elaboratingtherestrictingsportsdanceartqualitiesforthedancesLatinathletics.1.psychology(1)estheticism(2)understanding(3)ideation2.artisticculture(1)humanityquality(2)contestants∀recognitionofartisticform(3)contestants∀understandingofartlanguage3.professionalquality(1)musicreceptioncapability(2)physicalquality(3)bodylanguageexpressiveability.ThequestionofhowtocultivatesportsdanceLatindanceprofessionalathleteartisticqualitiesisputforwardinthearticle.Keywords:sportsdance;Latindance;professionalathlete;artisticquality;teachingandtraining(上接第88页)RealizationofBpNeuralNetworkBasedonMatlabSHIYun(ArchitectureSchoolofMechanicalandElectricalEngineering,AnhuiUniversity,Hefei230601,China)Abstract:ThisarticleintroducesBpneuralnetwork,andamethodusingtheMatlabneuralnettoolkittodesignBpnetworkwasintroduced.Moreoverthekeyissuesofnetwork∀srealization,whichincludesdesigningprocessofnetwork,selectinglearningregulation,evolutionarytrainingweight,thresholdvalueandnetworkemulationwereillustratedindetail,andsomeapplicationexamplesaregivenout.Thedesignofatypicalthree-layerstructureBpNeuralNetworkoffunctionapproximationisimplemented.Keywords:Bpneuralnetwork;Matlabsimulation;functionapproximation111