MATLAB-BP 神经网络中的MIV算法学习

clear;
close all;
clc

%输入输出数据
data = load('data.txt');
p=data(1:end,1:end-1)';%输入数据
t=data(1:end,end)';  %输出数据
[pn,minp,maxp,tn,mint,maxt]=premnmx(p,t)

% %变量筛选MIV算法初步实现
p = p';
[m,n] = size(p);  %输入数据p的size
yy_temp = p;

%p_increase为增加10%的矩阵
for i = 1:n
    p = yy_temp;
    pX = p(:,i);
    pa = pX*1.1;
    p(:,i) = pa;
    aa = ['p_increase' int2str(i) '=p'];
    eval(aa);
end

%p_decrease为减少10%的矩阵
for i = 1:n
    p = yy_temp;
    pX = p(:,i);
    pa = pX*0.9;
    p(:,i) = pa;
    aa = ['p_decrease' int2str(i) '=p'];
    eval(aa);
end

% %利用原始函数训练一个正确的神经网络
nntwarn off;

p = p';
%bp网络建立
net = newff(minmax(p),[20,1],{'tansig','purelin'},'trainlm');
%初始化BP网络
net = init(net);
%网络训练参数设置
net.trainParam.show = 5;
%net.trainParam.lr = 0.05;
%net.trainParam.mc = 0.9;
net.trainParam.epochs = 300;
net.trainParam.goal = 1e-5;
[net,tr] = train(net,p,t);

% %变量筛选MIV算法的后续实现（差值计算）

%转置后sim

for i = 1:n
    eval(['p_increase',num2str(i),'=transpose(p_increase',num2str(i),')'])
end

for i = 1:n
    eval(['p_decrease',num2str(i),'=transpose(p_decrease',num2str(i),')'])
end

%result_in为增加10%后的输出 result_de为减少10%后的输出
for i = 1:n
    eval(['result_in',num2str(i),'=sim(net,','p_increase',num2str(i),')'])
end

for i = 1:n
    eval(['result_de',num2str(i),'=sim(net,','p_decrease',num2str(i),')'])
end

for i = 1:n
    eval(['p_increase',num2str(i),'=transpose(result_in',num2str(i),')'])
end

for i = 1:n
    eval(['p_decrease',num2str(i),'=transpose(result_de',num2str(i),')'])
end

% %MIV_n的值为各个项网络输出的MIV值，MIV被认为是在神经网络中评价变量相关的最好指标之一，其符号代表相关的方向...
%绝对值大小代表影响的重要性
for i = 1:n
    IV = ['result_in',num2str(i),'-result_de',num2str(i)];
    eval(['MIV_',num2str(i),'=mean(',IV,')'])
end
联系方式：QQ1277430842