BP神经网络预测的MATLAB实现

200931JournalofHarbinSeniorFinanceCollege97BP神经网络预测的MATLAB实现焦淑华1,夏冰1,徐海静2,刘莹32008-10-08:2007(:11525028)(1,150030;2,150080;3,150027):在人工神经网络的实际应用中,大多数的人工神经网络模型是采用前馈反向传播网络(Back-Propagation-Network,简称BP网络)或它的变化形式它是前向网络的核心,体现了人工神经网络最精华的部分近年来MATLAB因其编程效率高,易学易懂,被广泛应用以旅游需求预测为例,说明MATLAB可以实现BP神经网络的预测:神经网络;MATLAB;预测,,,,,,(),,1974,P.Werbos,2080,DavidRunelhart,GeoffreyHintonRonaldWilliams,DavidParkr,YannnLeCunBP1986,PDP(paralleldistributedprocession)ParallelDistributedProcessing,,BP,BPBPW-H,,KolmogorovBP,BP,,,,(Back-Propagation-Network,BP),BP,,:,,,,BP,,1:1BPBP,(一)网络层数BP,,,BP,BP(二)网络各层中神经元的个数,,555,5;1,,,,;,,,,,,,:i=n+m+ai,n,m,a1a10(1),12BPMATLAB(一)数据样本的预处理,:10-5,,1993-1998,1993-1997;1998;1994-1999,1994-1998,1999(二)确定激活函数,tansigpurelin(三)设定网络的最大学习迭代次数为6000次(四)设定网络的学习精度为0.005(五)创建和训练BP神经网络的MATLAB程序%lyycclearal;lP=[2.165522.329353.137093.202673.365143.447023.358913.865904.153804.344533.334773359572.630993.113464.016633.850123.945483.808593.944554.318894.359804.921294.046073.602703.187482.954874.244344.514934.575624.434754.278314.587954.960795.477864.795113.808593.441543.487794.534064.771244.809184.849795.195105.798085.453356.182195.379134.965714.463053.847845.623276.112435.655435.489655.697766.237725.924576.969985.936365.48528];T=[4.997144.706366.282586.26596.338215.890436.088566.944196.961087.565456.48063559884];net=newff(minmax(P),[121],{'tansig'p'urelin}',t'raingdx,''learngdm)';net.trainParam.epochs=10000;net.trainParam.goal=0.005;net.trainParam.show=500;net=train(net,P,T),411,0.00499674,(六)测试BP神经网络,(1),BPMATLAB119991121999BP152932352850024942164860103690393661960471629271516057241887227106693599710690771834171353087692097644609769967770810108874928056201177664977525012662627613980,,MATLAB,,:1.[M].:,2004.2,.MATLAB[M].:,2005.3.[M].:,2003.4.MATLAB[M].:,2003.:王丽华56