nxzcx.m

很好的bp神经网络预测模型

close all
% clear
echo on
% clc
pause
load sj
%load wbtao1;
P=guifan(1:12,:)';%定义输入样本矢量
T=p6(1:12,:)';
echo on
pause
%clc
val.P=guifan(13:16,:)';%定义验证样本
val.T=p6(13:16,:)';
% test.P=guifan(1:12,:)'
% test.T=p6(1:12,:)
pause
%clc
net=newff(minmax(P),[2,1],{'tansig','purelin'},'trainbfg');%创建一个新的前向神经网络
pause
%clc
%设计训练参数
net.trainParam.epochs=2000;%训练次数
net.trainParam.max_fail=350;%最大验证失败次数
%net.trainParam.lr=0.008;%学习速率
%net.trainParam.min_grad=1e-7;
%net.trainParam.goal=1e-8;%网络性能目标
net=init(net);
pause
%clc
%训练BP网络
[net,tr]=train(net,P,T,[],[],val);
pause
%clc
%对网络进行仿真
n1=size(p6);
%仿真训练样本
P2=guifan(1:1*12,:)';
T2=p6(1:12)'
% 仿真测试样本
P1=guifan(13:n1,:)';
T1=p6(13:n1)';

A=(sim(net,P1));
B=(sim(net,P2));
E=T1-A;         
MSE=mse(E)
pause
save ningxia net
%clc
%绘制仿真拟合结果曲线
close;
G=1:1:n1-12; 
%  红色代表仿真值,绿色代表实际值
plot(G,A,'r*',G,T1,'g*');
pause
[m,b,r]=postreg(A,T1)

d1=0.9
d2=(1-d1)/2
for i=1:n1-12;  
   t4(i)=(A(i)-d2)*(p6max-p6min)/d1+p6min;
end 

for z=1:1:12
  y12(z)=(B(z)-d2)*(p6max-p6min)+p6min;
end
Y12=y12'
T4=t4'
save Y12 T4
pause
echo off