- 积分
- 0
- 注册时间
- 2008-3-31
- 仿真币
-
- 最后登录
- 1970-1-1
|
用matlab中提供的人工神经网络做一个简单的数据拟合,输入为19个变量,输出为1个变量,相关数据如下,在运行中提示不收敛。而且拟合结果和目标相差比较远。因为此前没有接触过神经网络,查相关资料通过增加隐层层数及神经元结点数也没能解决,还请高手指点一下啊,不胜感激!下面是MATLAB提示
TRAINCGF-srchcha-calcgrad, Epoch 0/100, MSE 1.39861e+011/0, Gradient 2.42514e+015/1e-006
TRAINCGF-srchcha-calcgrad, Epoch 4/100, MSE 1.39861e+011/0, Gradient 0/1e-006
TRAINCGF, Minimum gradient reached, performance goal was not met
下面是整个计算过程的源代码:
xx = [2124.03 2581.94 ;16208300000 19781900000 ;1702890 2299850 ;34130000 41710000 ;1705830 2296380; 47.6019 64.3748 ;4.38525e-006 5.34268e-006 ;2325.44 3138.5 ;11906300000 16068000000 ;1440880 2158870 ;36137000 53882700; 720567 1073080 ;36.1173 53.896 ;2063.13 3429.24 ;2402320000 3582360000; 1126690 1871890 ;32259900 53620800 ;263051 434757 ;26.258 43.6448];
x1=[ 2350 2581.94
18000000000 18000000000
2000000 2000000
37920000 37920000
2000000 2000000
56 56
4.87e-006 4.87e-006
2730 2730
14000000000 14000000000
1800000 1800000
45000000 45000000
900000 900000
45 45
2750 2750
3000000000 3000000000
1500000 1500000
43000000 43000000
350000 350000
35 35 ];
x2=[ 2124.03 2350
18000000000 19781900000
2000000 2000000
37920000 37920000
2000000 2000000
56 56
4.87e-006 4.87e-006
2730 2730
14000000000 14000000000
1800000 1800000
45000000 45000000
900000 900000
45 45
2750 2750
3000000000 3000000000
1500000 1500000
43000000 43000000
350000 350000
35 35 ];
x3=[ 2350 2350
16208300000 18000000000
2000000 2299850
37920000 37920000
2000000 2000000
56 56
4.87e-006 4.87e-006
2730 2730
14000000000 14000000000
1800000 1800000
45000000 45000000
900000 900000
45 45
2750 2750
3000000000 3000000000
1500000 1500000
43000000 43000000
350000 350000
35 35 ];
x=[x1 x2 x3];
y=[ -374960 -376700 -371690 -382480 -362860 -374910];
h=[38,38,38,38,19,1];
net=newff(xx,h,{'tansig','tansig','tansig','tansig','tansig','logsig'},'traincgf');
[net,tr,y1,e]=train(net,x,y); |
|