神经网络精度达到了，但是训练样本时预测值很不准。

大家好，我想问一下，这是什么原因造成的，该如何解决，我在用神经网络学习样本时精度已达到要求，但是训练时发现与真实值差别很大，这是为什么呢？我的代码如下：

P=[-1,-0.928571,-0.857143,-0.785714,-0.714286,-0.642857,-0.5,-0.428571,-0.357143,-0.214286,-0.142857,-0.0714286,0,0.0714286,0.142857,0.285714,0.357143,0.428571,0.5,0.571429,0.642857,0.785714,0.857143,0.928571,1;-1,-0.970588,-0.941176,-0.911765,-0.882353,-0.852941,-0.794118,-0.764706,-0.735294,-0.676471,-0.647059,-0.617647,-0.588235,-0.558824,-0.529412,-0.470588,-0.441176,-0.411765,-0.382353,-0.352941,-0.323529,-0.264706,-0.352941,-0.205882,-0.176471;-1,-1,-1,-1,-1,-1,-1,-1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1,1;-1,-0.984848,-0.924242,-0.884848,-0.848485,-0.80303,-0.727273,-0.69697,-0.560606,-0.287879,-0.151515,-0.106061,-0.0606061,-0.0151515,0.030303,0.0909091,0.136364,0.181818,0.227273,0.272727,0.318182,0.545455,0.69697,0.848485,1];

T=[0.299353,0.1,0.199029,0.224919,0.249515,0.131715,0.134951,0.202913,0.369256,0.336246,0.633333,0.202265,0.465696,0.399029,0.20356,0.156958,0.652751,0.133657,0.479935,0.101294,0.243042,0.704531,0.613916,0.756311,0.9;0.1,0.285636,0.27241,0.307512,0.340645,0.474806,0.56703,0.557926,0.248696,0.337998,0.204276,0.49763,0.354778,0.417996,0.594394,0.694306,0.389227,0.775082,0.9,0.863782,0.766917,0.506283,0.581703,0.522699,0.469325];

NodeNum=30;

TypeNum=2;

TF1='logsig';TF2='logsig';

net=newff(minmax(P),[NodeNum TypeNum],{TF1,TF2},'trainlm');

net.trainParam.epochs = 10000; %最大训练轮回

net.trainParam.goal = 1e-4;

net = train(net,P,T);

P_test=[-0.571429,-0.285714,0.214286,0.714286;-0.823529,-0.705882,-0.5,-0.294118;-1,1,1,1;-0.757576,-0.424242,0.0606061,0.363636];

x=sim(net,P_test)

x =

0.1111 0.3975 0.1008 0.5377

0.5515 0.2695 0.7173 0.5972

而我的样本值为

x=0.446602 0.610841 0.026699 0.673948

0.302533 0.163908 0.694006 0.519865

Answer 1

[kogipec]

NodeNum=30;

TypeNum=2;

TF1='logsig';TF2='logsig';

调一下这四个参数，尝试一下。