改进的BP神经网络算法(C语言源码)

来源：互联网   发布日期：2011-12-04 19:24:01   浏览：50575次  

#include "stdio.h"
#include "stdlib.h"
#include "time.h"
#include "math.h"
/*********************************************
inpoints 为输入神经元个数，可改变
outpoints为输出神经元个数
defaultpoints为隐层神经元个数
datagrough为样本数据个数
**********************************************

******以下数据定义可以修改*****/
#define A 0
#define a 1
#define b 1
#define c 1
#define ALFA 0.85
#define BETA 0.2 //学习率0~1
#define Total 20000
#define inpoints 9
#define outpoints 5
#define defaultpoints 28
#define datagrough 44
#define forecastdata 4

/**********定义所需变量********/
double InpointData[datagrough][inpoints],OutpointData[datagrough][outpoints]; /* 输入输出数据 */
double InpointData_MAX[inpoints],InpointData_MIN[inpoints]; /* 每个因素最大数据 */
double OutpointData_MAX[outpoints],OutpointData_MIN[outpoints]; /* 每个因素最小数据 */
double w[defaultpoints][inpoints],limen[defaultpoints],v[outpoints][defaultpoints]; /* 连接权值、阈值 */
double dlta_w[defaultpoints][inpoints],dlta_limen[defaultpoints],dlta_v[outpoints][defaultpoints]; /* 连接权、阈值修正值 */
double defaultOutpoint[defaultpoints],Outpoint_dp[outpoints],Outpoint_ep[datagrough];
/**************************读数据文件******************************/
void ReadData()
{
FILE *fp1,*fp2;
int i,j;
if((fp1=fopen("D:\\data\\训练输入.txt","r"))==NULL)
{
printf("1can not open the file\n");
exit(0);
}
for(i=0;i<datagrough;i++)
for(j=0;j<inpoints;j++)
fscanf(fp1,"%lf",&InpointData[i][j]);
fclose(fp1);

if((fp2=fopen("D:\\data\\训练输出.txt","r"))==NULL)
{
printf("2can not open the file\n");
exit(0);
}
for(i=0;i<datagrough;i++)
for(j=0;j<outpoints;j++)
fscanf(fp2,"%lf",&OutpointData[i][j]);
fclose(fp2);
}
/*****************************************************/

/*****************************************归一化******************************************************/
void unitary()
{
int i,j;
int k=0;
for(j=0;j<inpoints;j++) //找出每列的最大、最小值存放在数组InpointData_MAX[j]、InpointData_MIN[j]中
{
InpointData_MAX[j]=InpointData[0][j];
InpointData_MIN[j]=InpointData[0][j];
for(i=0;i<datagrough;i++)
if(InpointData_MAX[j]<InpointData[i][j])
InpointData_MAX[j]=InpointData[i][j];
else if(InpointData_MIN[j]>InpointData[i][j])
InpointData_MIN[j]=InpointData[i][j];
}

for(j=0;j<outpoints;j++) //找出每列的最大、最小值存放在数组OutpointData_MAX[j]、OutpointData_MIN[j]中
{
OutpointData_MAX[j]=OutpointData[0][j];
OutpointData_MIN[j]=OutpointData[0][j];
for(i=0;i<datagrough;i++)
if(OutpointData_MAX[j]<OutpointData[i][j])
OutpointData_MAX[j]=OutpointData[i][j];
else if(OutpointData_MIN[j]>OutpointData[i][j])
OutpointData_MIN[j]=OutpointData[i][j];
}
/***************将数据归一处理,处理之后的数据全部在[0，1]之间*************************/
for(j=0;j<inpoints;j++)
for(i=0;i<datagrough;i++)
if(InpointData_MAX[j]==0)
InpointData[i][j]=0;
else
InpointData[i][j]=(InpointData[i][j]-InpointData_MIN[j]+A)/(InpointData_MAX[j]-InpointData_MIN[j]+A);

for(j=0;j<outpoints;j++)
for(i=0;i<datagrough;i++)
if(OutpointData_MAX[j]==0)
OutpointData[i][j]=0;
else
OutpointData[i][j]=(OutpointData[i][j]-OutpointData_MIN[j]+A)/(OutpointData_MAX[j]-OutpointData_MIN[j]+A);

}
/*****************************************************/

/*********************初始化，随机赋初值**************************/
void Initialization()
{
int i,j;
srand((unsigned)time(NULL)); //头文件名 #include <time.h>

for(i=0;i<defaultpoints;i++) //给输入层到隐层的连接权赋随机值LianJie_w[i][j],这些值在[0,1]
for(j=0;j<inpoints;j++)
{
w[i][j]=(rand()*2.0/RAND_MAX-1)/2;
dlta_w[i][j]=0;
}

for(i=0;i<defaultpoints;i++)
{
limen[i]=(rand()*2.0/RAND_MAX-1)/2;
dlta_limen[i]=0;
}

for(i=0;i<outpoints;i++) //给隐层到输出层的连接权赋初值
for(j=0;j<defaultpoints;j++)
{
v[i][j]=(rand()*2.0/RAND_MAX-1)/2;
dlta_v[i][j]=0;
}
}
/**********************求单样本的计算输出误差*******************************/
void out_sub1(int t)
{
int i,j;
double defaultInpoint[defaultpoints];
double Outpoint_y[outpoints];
Outpoint_ep[t]=0;
for(i=0;i<defaultpoints;i++)
{
double sum=0;
for(j=0;j<inpoints;j++)
sum+=w[i][j]*InpointData[t][j];
defaultInpoint[i]=sum+limen[i];
defaultOutpoint[i]=1/(a+b*exp(-1*c*defaultInpoint[i]));//求O[i]
}

for(j=0;j<outpoints;j++)//求Y[i]
{
Outpoint_y[j]=0;
for(i=0;i<defaultpoints;i++)
Outpoint_y[j]+=v[j][i]*defaultOutpoint[i];
Outpoint_dp[j]=OutpointData[t][j]-Outpoint_y[j];
Outpoint_ep[t]+=Outpoint_dp[j]*Outpoint_dp[j]/2;
}
}
/*****************************反算权值******************************************/
void out_sub2(int t)
{
int i,j,k;
double s;
for(i=0;i<defaultpoints;i++)
{
s=0;
for(j=0;j<outpoints;j++)
{
dlta_v[j][i]=ALFA*dlta_v[j][i]+BETA*Outpoint_dp[j]*defaultOutpoint[i]; //
s+=v[j][i]*Outpoint_dp[j];
v[j][i]+=dlta_v[j][i];
}
dlta_limen[i]=ALFA*dlta_limen[i]+BETA*defaultOutpoint[i]*(1-defaultOutpoint[i])*s;//
limen[i]+=dlta_limen[i];
for(k=0;k<inpoints;k++)
{
dlta_w[i][k]=ALFA*dlta_w[i][k]+BETA*defaultOutpoint[i]*(1-defaultOutpoint[i])*s*InpointData[t][k];//
w[i][k]=w[i][k]+dlta_w[i][k];
}
}
}
/*******************************************************/
void forecast()
{
int i,j,t,k=0;
double e,e1[forecastdata]={0}; //训练误差
double sss;
double InputData_x[forecastdata][inpoints],tp[forecastdata][outpoints];
double defInpoint,defOutpoint[defaultpoints],y[forecastdata][outpoints];//y[forecastdata][outpoints]为网络检验输出
FILE *fp1,*fp3;
if((fp1=fopen("D:\\data\\预测输入.txt","r"))==NULL) //检验数据输入
{
printf("3can not open the file\n");
exit(0);
}
for(i=0;i<forecastdata;i++)
for(j=0;j<inpoints;j++)
fscanf(fp1,"%lf",&InputData_x[i][j]);
fclose(fp1);

if((fp3=fopen("D:\\data\\预测输出.txt","r"))==NULL) //实际检验结果输出
{
printf("31can not open the file\n");
exit(0);
}
for(i=0;i<forecastdata;i++)
for(j=0;j<outpoints;j++)
fscanf(fp3,"%lf",&tp[i][j]);
fclose(fp3);

for(j=0;j<inpoints;j++) // 检验数据归一化
for(i=0;i<forecastdata;i++)
if(InpointData_MAX[j]==0)
InputData_x[i][j]=0;
else
InputData_x[i][j]=(InputData_x[i][j]-InpointData_MIN[j]+A)/(InpointData_MAX[j]-InpointData_MIN[j]+A);

for(j=0;j<outpoints;j++)
for(i=0;i<forecastdata;i++)
if(OutpointData_MAX[j]==0)
tp[i][j]=0;
else
tp[i][j]=(tp[i][j]-OutpointData_MIN[j]+A)/(OutpointData_MAX[j]-OutpointData_MIN[j]+A);

do
{
Initialization(); //初始化连接权值w[i][j],limen[i],v[k][i]
k=0;
do
{
e=0;
for(t=0;t<datagrough;t++)
{
out_sub1(t); //正向计算网络输出
out_sub2(t); //反向计算，修正权值
e+=Outpoint_ep[t]; //计算输出误差
}
k++;
}while((k<Total)&&(e>0.1));
sss=0; //中间参数
for(t=0;t<forecastdata;t++)
{
e1[t]=0;
for(i=0;i<defaultpoints;i++)
{
double sum=0;
for(j=0;j<inpoints;j++)
sum+=w[i][j]*InputData_x[t][j];
defInpoint=sum+limen[i];
defOutpoint[i]=1/(a+b*exp(-1*c*defInpoint));
}
for(j=0;j<outpoints;j++)
{
y[t][j]=0;
for(i=0;i<defaultpoints;i++)
y[t][j]+=v[j][i]*defOutpoint[i];
e1[t]+=(y[t][j]-tp[t][j])*(y[t][j]-tp[t][j])/2;
y[t][j]=y[t][j]*(OutpointData_MAX[j]-OutpointData_MIN[j]+A)+OutpointData_MIN[j]-A;
}
sss+=e1[t];
}
sss=sss/forecastdata;
printf(" %lf %lf\n",e,sss);
}while(sss>0.12);
}
/********************************************************/


void main()
{
int i,j,k;
FILE *fp2;
ReadData(); //读训练数据：输入和输出
unitary(); //归一化，将输入输出数据归一，结果在[0，1]中
forecast(); //检验误差
if((fp2=fopen("D:\\data\\计算权值.txt","w"))==NULL) //文件输出训练好的权值
{
printf("6can not open the file\n");
exit(0);
}
for(i=0;i<defaultpoints;i++)
{
for(k=0;k<inpoints;k++)
fprintf(fp2," %lf ",w[i][k]);
fprintf(fp2,"\n");
}
fprintf(fp2,"\n");
for(i=0;i<defaultpoints;i++)
fprintf(fp2," %lf ",limen[i]);
fprintf(fp2,"\n\n");
for(i=0;i<defaultpoints;i++)
{
for(j=0;j<outpoints;j++)
fprintf(fp2," %lf ",v[j][i]);
fprintf(fp2,"\n");
}
fclose(fp2);

}

赞助本站