#include<iostream>
#include<stdlib.h>
#include<time.h>
#include<math.h>
#define n 3 //输入维数
#define p 12 //隐层单元数
#define q 2 //输出维数
#define m 6 //样本数k循环
using namespace std;
struct annbp{
double x[n], hi[p], ho[p], yi[q], yo[q], d[q];
// 输入向量 隐含层输入向量 隐含层输出向量 输出层输入向量 输出层输出向量 期望输出向量
double wih[n][p], who[p][q], bh[p], b[q];
//输入层与隐间层的连接权值 隐含层与输出层的连接权值 隐含层各神经元的阈值 输出层各神经元的阈值
int loopmax;//最大循环次数
double a; //学习率
double e;//误差上限
};
double f(double x)
{
return 1.0/(1.0+exp(-x));
}
void InitBp(annbp *bp) { //初始化bp网络
bp->loopmax=100000;
bp->a=0.2;
bp->e=0.000001;
int i, j;
srand((unsigned)time(NULL));
rand();
for (i = 0; i < n; i++)
for (j = 0; j < p; j++)
bp->wih[i][j] = rand()/(double)(RAND_MAX);
for (i = 0; i < p; i++)
for (j = 0; j < q; j++)
bp->who[i][j] = rand()/(double)(RAND_MAX);
}
int TrainBp(annbp *bp, float x[m][n], int y[m][q])
{
//训练bp网络，样本为x，理想输出为y
double f1 = bp->e; //精度参数
double a = bp->a; //学习率
double v[n][p], w[p][q]; //权矩阵
double ChgH[p], ChgO[q]; //修改量
double O1[p], O2[q]; //隐层和输出层输出量
int LoopCout = bp->loopmax; //最大循环次数
int i, j, k, nn;
double temp;
for (i = 0; i < n; i++) // 复制结构体中的权矩阵
for (j = 0; j < p; j++)
v[i][j] = bp->wih[i][j];
for (i = 0; i < p; i++)
for (j = 0; j < q; j++)
w[i][j] = bp->who[i][j];
double e = f1 + 1;
for (nn = 0; e > f1 && nn < LoopCout; nn++) //对每个样本训练网络
{
e = 0;
for(i=0;i<m;i++){
for(k=0;k<p;k++)
{ //计算隐层输出向量
temp = 0;
for (j=0;j<n;j++)
temp+=x[i][j]*v[j][k];
O1[k]=f(temp);//隐层输出向量
}
for (k = 0; k < q; k++) { //计算输出层输出向量
temp = 0;
for (j = 0; j < p; j++)
temp = temp + O1[j] * w[j][k];
O2[k] = f(temp);//输出层输出向量
}
for (j = 0; j < q ; j++) //计算输出误差
e += (y[i][j] - O2[j]) * (y[i][j] - O2[j]);
for (j = 0; j < q; j++) //计算输出层的权修改量
ChgO[j] = O2[j] * (1 - O2[j]) * (y[i][j] - O2[j]);
for (j = 0; j < p; j++) { //计算隐层权修改量
temp = 0;
for (k = 0; k < q; k++)
temp = temp + w[j][k] * ChgO[k];
ChgH[j] = temp * O1[j] * (1 - O1[j]);
}
for (j = 0; j < p; j++) //修改输出层权矩阵
for (k = 0; k < q; k++)
w[j][k] = w[j][k] + a * O1[j] * ChgO[k];
for (j = 0; j < n; j++)
for (k = 0; k < p; k++)
v[j][k] = v[j][k] + a * x[i][j] * ChgH[k];
}
if (nn % 10 == 0)
cout<<"误差 : "<<e<<endl;
}
cout<<"总共循环次数："<<nn<<endl;
cout<<"调整后的隐层权矩阵"<<endl;
for (i = 0; i < n; i++) {
for (j = 0; j < p; j++)
cout<<v[i][j]<<"\t";
cout<<endl;
}
cout<<"调整后的输出层权矩阵："<<endl;
for (i = 0; i < p; i++) {
for (j = 0; j < q; j++)
cout<< w[i][j]<<"\t";
cout<<endl;
}
for (i = 0; i < n; i++) //把结果复制回结构体
for (j = 0; j < p; j++)
bp->wih[i][j] = v[i][j];
for (i = 0; i < p; i++)
for (j = 0; j < q; j++)
bp->who[i][j] = w[i][j];
cout<<"bp网络训练结束！"<<endl;
return 1;
}
int UseBp(annbp *bp) { //使用bp网络
float Input[n];
double O1[p];
double O2[q]; //O1为隐层输出,O2为输出层输出
while (1) { //持续执行，除非中断程序
cout<<"请输入3个数："<<endl;
int i, j;
for (i = 0; i < n; i++)
cin>>Input[i];
double temp;
for (i = 0; i < p; i++) {
temp = 0;
for (j = 0; j < n; j++)
temp += Input[j] * bp->wih[j][i];
O1[i] = f(temp);
}
for (i = 0; i < q; i++) {
temp = 0;
for (j = 0; j < p; j++)
temp += O1[j] * bp->who[j][i];
O2[i] = f(temp);
}
cout<<"结果："<<endl;
for (i = 0; i < q; i++)
cout<< O2[i]<<"\t";
cout<<endl;
}
return 1;
}
void main()
{
float x[m][n] = {
{0.8,0.5,0},
{0.9,0.7,0.3},
{1,0.8,0.5},
{0,0.2,0.3},
{0.2,0.1,1.3},
{0.2,0.7,0.8}}; //训练样本
int y[m][q] = {
{0,1},
{0,1},
{0,1},
{1,0},
{1,0},
{1,0}}; //理想输出
annbp *bp=new annbp;
InitBp(bp); //初始化bp网络结构
TrainBp(bp,x,y); //训练bp神经网络
UseBp(bp); //测试bp神经网络
}
