请教BP神经网络中出现的一个问题

zzh911

编写的具体程序如下：
>> P=[0.936694,0.981744,1,0.667698,0.830719;  
0.664891,0.981744,0.989899,0.667698,0.498492;
0.731759,0.975711,0.979798,0.600792,0.556788;
0.533898,0.975711,0.888889,0.600793,0.600218;
0.246123,0.496107,0.424242,0.424534,0.120379;
0.135998,0.496107,0.525253,0.424534,0.243802;  
0.334354,0.049286,0.606061,0.086876,0.210937;
0.141461,0.645391,0.444444,0.632913,0.171543;  
0.525646,0.645391,0.919192,0.632912,0.526012;
0.167553,0.022205,0.737374,0.009141,0.019267;
0.326964,0.022205,0.333333,0.009141,0.147202;
0.241284,0.325522,0.919192,0.619236,0.256637;   
0.441056,0.325522,0.989899,0.619236,0.170411]';

T=[5000;5000;5000;5000;4400;4400;4000;4750;5160;3790;5250;2300;2500]';

PR=[0,1;0,1;0,1;0,1;0,1];

net=newff(PR,[10,1],{'tansig','purelin'},'trainlm');
net.trainParam.epochs=200;
net.trainParam.goal=0.001;
net=init(net);
net=train(net,P,T);

结果出现
TRAINLM-calcjx, Epoch 0/200, MSE 1.97782e+007/0.001, Gradient 21488.9/1e-010
TRAINLM-calcjx, Epoch 10/200, MSE 493351/0.001, Gradient 2.7615e-010/1e-010
TRAINLM, Maximum MU reached, performance goal was not met.
请问该如何解决，谢谢了，十万火急啊。

frogfish

检查一下你的样本是否有问题，如果样本没有问题可以通过设置 net.trainParam.mu 来调节参数。

net.trainParam.mu 0.001          Initial Mu
net.trainParam.mu_dec 0.1      Mu decrease factor
net.trainParam.mu_inc 10        Mu increase factor
net.trainParam.mu_max 1e10 Maximum Mu

hlhl

请问楼上的mu是什么参数？是所说的变学习率吗？

F117_ren_0

目标误差没达到，
net.trainParam.epochs=200;
net.trainParam.goal=0.001;
要么把训练次数搞高点，至少3000次，要么就吧目标误差写大点，0.01