%利用動(dòng)量梯度下降算法訓(xùn)練BP網(wǎng)絡(luò)
close all
clear
echo on
%newff 生成一個(gè)新的向前神經(jīng)網(wǎng)絡(luò)
%train 對(duì)b神經(jīng)網(wǎng)絡(luò)進(jìn)行訓(xùn)練
%sim對(duì)bp神經(jīng)網(wǎng)絡(luò)進(jìn)行仿真
%定義訓(xùn)練樣本
%p為輸入矢量
p=[-1 ,12, 3, 1;
    -1, 1, 5, -3];
%t為目標(biāo)矢量
t=[-1, -1, 1, -1]
clc
%創(chuàng)建一個(gè)新的向前神經(jīng)網(wǎng)絡(luò)
net=newff(minmax(p),[3,1],{'tansig','purelin'},'traingdm');
%當(dāng)前輸入層權(quán)值和域值
inputweights=net.iw{1,1}%產(chǎn)生隱含層個(gè)神經(jīng)元的域值
Inputbias=net.b{1}
%當(dāng)前網(wǎng)絡(luò)層權(quán)值和閾值
layerweights=net.lw{2,1}
layerbias=net.b{2}
clc
%設(shè)置訓(xùn)練參數(shù)
net.trainparam.show=50;%迭代50次顯示一次
net.trainparam.lr=0.05;%（一般大小仔0-1之間）
net.trainparam.mc=0.9;%動(dòng)量因子0.09（0-1）
net.trainparam.epochs=1000;%迭代次數(shù)
net.trainparam.goal=1e-3;%誤差
clc
%調(diào)用traingdm算法訓(xùn)練bp網(wǎng)絡(luò)
[net,tr]=train(net,p,t);
clc%對(duì)bp網(wǎng)絡(luò)進(jìn)行仿真
a=sim(net,p)
e=t-a
mse=mse(e)%計(jì)算軍方誤差
clc
echo off