人工神经网络的BP算法及其应用

©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.(1.,215006;2.,214064):BP,BP,,MATLABBP,:BP;;BP:TP18:A:1009-2552(2004)04-0001-04BPalgorithmanditsapplicationinartificialneuralnetworkHUJin2bin1,TANGXu2qing2(1.SchoolofComputerScienceandTechnology,SuzhouUniversity,Suzhou215006,China;2.SchoolofScience,SouthernYangtzeUniversity,Wuxi214064,China)Abstract:OnthebasisofnormalBPalgorithmofANN,thispaperinvestigatesthemathematicbaseofBPalgorithm,anddiscussesitsdefectsaswellasimprovingalgorithm.ProgrammingwhichrealizestheBPalgo2rithmisputforwardfinallyintheMATLABenvironment,andtheresultcomparingtoimprovingmethodissati2sfactory.Keywords:BPalgorithm;neuralnetwork;BPnetwork01974,WerbosBP,1982RumeehartBP,(Back-PropagationAlgorithm)(ArtificialNeuralNetwork)BP,:f:RnRm,,xiRnyiRn,g(),g(xi)=yi(i=1,2,,N),f,(),fg,,,Kommog2orow[1]BP[1]BP1949HebbHebbBP,BP,,MATLABBP,1BPn(),,,:netjk=iwjiOik(1)Ojk=f(netjk)(2)k,wjiij,Oik,Ojkf,Sigmoid:2003-11-02:(314000-52212047):(1979-),,1©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.:Ek=12j(yjk-Ojk)2Ekk;yjkj();Ojkj,(),,,E=Ek,kwjiwji,kwji-9Ek9wji,kwji-9Ek9wji(3)(1)9netjk9wji=99wji(pwjpOpk=Oik9Ek9wji=9Ek9netjk9netjk9wjijkjk=-9Ek9netji:9Ek9wji=-jkOik,E,kwji=jkOik(4),,,,BPjk,jk,:jk=-9Ek9netjk=-9Ek9Ojk9Ojk9netjk(5)(2)9Ojk9netjk=f(netjk)(5)9Ek9Ojk:j,Ek9Ek9Ojk=-(yjk-Ojk):,,(5)jk=(yjk-Ojk)f(netjk)(6)j,m9Ek9netmk9netmk9Ojk=m9Ek9netmk99OjkiwmiOik=-mmkwmj(5),jk=f(netjk)mmkwmj(7),(4)jk,(6)(7)jk,f(netjk)Sigmoid,Ojk=f(netjk)=11+e(-netjk+)(8):f(netjk)=11+e(-netjk+)1-11+e(-netjk+)(9)(8)(9)f(netjk)Ojk:f(netjk)=Ojk(1-Ojk)(10)(10)(6)(7)BP,,Sigmoid,,2,,:(1),BP2©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.(2),,n,S,f(n),(3)BP,,BP,,,,,,,,,:,,,,,,,,,:wji(k+1)=(1-mc)jpj+mcwji(k)k,mc,0.95,,,,j,,wji(k+1)=wji(k),wji=0,,,,:E=k(yk-Ok)2P2Ok1,E,E,,f(yk-Ok)(yk-Ok)2,yk=OkBaum1988:E=k121+yklog1+yk1+Ok+121-yklog1-yk1-Okyk=Ok,E=0,,Ok1,:f(n)=tanh(n)=1-e-2n1+e-2n=en-e-nen+e-nf(n)=(tanh(n))=1-f2(n)n:Ok=1-Ok2,9E9n=yk-Ok=jkjk=f(n)(yk-Ok),f(n),n,,f(n)0,f(n)w,w,1989S.Fahlman,k=(f(n)+0.1)(yk-Ok),f(n),|n|,j,,,,,,3©1994-2010ChinaAcademicJournalElectronicPublishingHouse.Allrightsreserved.:,,,;,,,E,:(n+1)=nE(n+1)E(n)nE(),,,,,,,,3BPBP,BP,,,,,,MATLAB,,,BPMATLABBP,.(3)BPP=[-1,-2,3,1;-1,1,5,-3];%T=[-1,-1,1,1];%%NEWFFnet=newff(minmax(P),[3,1],{tansig,pure2lin},traingdm);,285(1):L-Mtrainlm1BP(traingdm),224,,L-M,L-M:[1],.[M].:,2001.[2].[M].:,2001.[3],.[M].:,1998.[4],.MATLAB6.X[M].:,2002.[5]MartinTHagan,HowardBDemuth,MarkBeale.NeuralNetworkDe2sign[M].NewYorkPWSPublishingCompany,2002.:12,,,2010,1.28;1.64