fnn.rar_dfxncbfnfn.c.fnn_模糊神经_模糊神经网络_神经

# define MAX1 200 # define MAX2 10000 # define MAX3 3000 /*各个网络的最大训练次数*/ # define A 30.0 /*A,B为NN1网络输出层激励函数参数*/ # define B 10.0 # define COEF3 0.000035 /*NN2网络的学习效率*/ # define BCOEF3 0.00005 /*NN2网络的阈值调整效率*/ # define COEF2 0.0035 /*NN1网络的学习效率*/ # define COEF1 0.8 /*NNMF网络的学习效率*/ # define ERROR3 0.002 /*各网络训练的允许误差*/ # define ERROR2 0.002 # define ERROR1 0.000028 # define ACCURACY 0.0005 /*网络要求精度*/ # define ACCURACY1 0.00004 # include "STDIO.H" # include "MATH.H" # include "STDLIB.H" /* NNMF网络学习样本及参数变量说明*/ int sample1[20][3]={{1,3,1},{1,5,2},{1,1,3},{1,3,4},{1,5,5},{5,1,4},{5,3,3},{5,5,2},{5,1,1}, {5,3,2},{1,5,3},{1,1,4},{1,3,5},{1,1,3},{5,3,2},{5,5,1},{5,1,2},{5,3,3},{5,5,4}}; double w1[5][10][10],wc1[5][10][10],b1[5][10],bc1[5][10]; double o1[5][10],netin1[5][10],d1[5][10],differ1; int count1,a1; double is1s; /*NN1网络样本及参数变量说明*/ double sample2[10][4]={{1,3,1,11.110},{1,5,2,6.521},{1,1,3,10.190},{1,3,4,6.043},{1,5,5,5.242}, {1,5,3,5.724},{1,1,4,9.766},{1,3,5,5.870},{1,5,4,5.406},{1,1,3,10.190}}; double w2[5][10][10],wc2[5][10][10],b2[5][10],bc2[5][10]; double o2[5][10],netin2[5][10],d2[5][10],differ2; int count2,a2; double is2s; /*NN2网络样本及参数变量说明*/ double sample3[10][4]={{5,1,4,19.020},{5,3,3,14.150},{5,5,2,14.360},{5,3,3,14.150},{5,3,2,15.390}, {5,3,2,15.390}, {5,5,1,19.6680},{5,1,2,21.060},{5,5,3,14.150},{5,5,4,12.680}}; double w3[5][10][10],wc3[5][10][10],b3[5][10],bc3[5][10]; double o3[5][10],netin3[5][10],d3[5][10],differ3; int count3,a3; double is3s; /*计算NN2网络隐含层和输出层的输出*/ void netout3(int m,int n) { int i,j,l; for(j=2,i=2;j<=m;j++) { netin3[i][j]=0.0; for (l=1;l<=m;l++) netin3[i][j]=netin3[i][j]+o3[i-1][l]*w3[i][l][j]; netin3[i][j]=netin3[i][j]-b3[i][j]; o3[i][j]=0.5*(exp(netin3[i][j])-exp(-netin3[i][j])); } for (j=1,i=3;j<=m;j++) { netin3[i][j]=0.0; for (l=1;l<=m;l++) netin3[i][j]=netin3[i][j]+o3[i-1][l]*w3[i][l][j]; netin3[i][j]=netin3[i][j]-b3[i][j]; o3[i][j]=0.5*(exp(netin3[i][j])-exp(-netin3[i][j])); } for (j=1,i=4;j<=m;j++) { netin3[i][j]=0.0; for (l=1;l<=m;l++) netin3[i][j]=netin3[i][j]+o3[i-1][l]*w3[i][l][j]; netin3[i][j]=netin3[i][j]-b3[i][j]; o3[i][j]=0.5*(exp(netin3[i][j])-exp(-netin3[i][j])); } } /*计算NN2网络的反向传播误差*/ void calculd3(int m,int n) { int i,j,l; double t; a3=count3-1; d3[4][1]=(o3[4][1]-sample3[a3][3])*0.5*(exp(netin3[4][1])+exp(-netin3[4][1])); for (j=1,i=3;j<=m;j++) {t=0.00; for(l=1;l<=n;l++) t=t+w3[i+1][j][l]*d3[i+1][l]; d3[i][j]=0.5*(exp(netin3[i][j])+exp(-netin3[i][j]))*t; } for (j=1,i=2;j<=m;j++) {t=0.00; for(l=1;l<=m;l++) t=t+w3[i+1][j][l]*d3[i+1][l]; d3[i][j]=0.5*(exp(netin3[i][j])+exp(-netin3[i][j]))*t; } } /*计算NN2网络权值W的调整量*/ void calculwc3(int m,int n) {int i,j,l; for (i=1,l=4;i<=m;i++) for(j=1;j<=n;j++) wc3[l][i][j]=-COEF3*d3[l][j]*o3[l-1][i]+0.5*wc3[l][i][j]; for (i=1,l=3;i<=m;i++) for(j=1;j<=m;j++) wc3[l][i][j]=-COEF3*d3[l][j]*o3[l-1][i]+0.5*wc3[l][i][j]; for (i=1,l=2;i<=m;i++) for(j=1;j<=m;j++) wc3[l][i][j]=-COEF3*d3[l][j]*o3[l-1][i]+0.5*wc3[l][i][j]; } /*计算NN2网络阈值的调整量*/ void calculbc3(int m,int n) { int j; for (j=1;j<=m;j++) { bc3[2][j]=BCOEF3*d3[2][j]; bc3[3][j]=BCOEF3*d3[3][j]; } for(j=1;j<=n;j++) bc3[4][j]=BCOEF3*d3[4][j]; } /*计算NN2网络权值W*/ void changw3(int m,int n) {int i,j; for (i=1;i<=3;i++) for(j=1;j<=m;j++) w3[2][i][j]=w3[2][i][j]+wc3[2][i][j]; for (i=1;i<=m;i++) for(j=1;j<=m;j++) w3[3][i][j]=w3[3][i][j]+wc3[3][i][j]; for (i=1;i<=m;i++) for(j=1;j<=n;j++) w3[4][i][j]=w3[4][i][j]+wc3[4][i][j]; } /*计算NN2网络阈值*/ void changb3(int m,int n) {int j; for (j=1;j<=m;j++) { b3[2][j]=b3[2][j]+bc3[2][j]; b3[3][j]=b3[3][j]+bc3[3][j]; } for(j=1;j<=n;j++) b3[4][j]=b3[4][j]+bc3[4][j]; } /*清除NN2网络权值变化量Wc*/ void clearwc3(void) {int i,j,l; for (i=0;i<5;i++) for (j=0;j<10;j++) for (l=0;l<10;l++) wc3[i][j][l]=0.00; } /*清除NN2网络阈值变化量bc*/ void clearbc3(void) {int i,j; for (i=0;i<5;i++) for (j=0;j<10;j++) bc3[i][j]=0.00; } /*初始化NN2网络权值W*/ void initialw3(void) { int i,j,l,x; double weight; for (i=0;i<5;i++) for (j=0;j<10;j++) for (l=0;l<10;l++) { x=100+random(400); weight=(double)x/50000000.00; w3[i][j][l]=weight; } } /*初始化NN2网络阈值*/ void initialb3(void) {int i,j,x; double fazhi; for (i=0;i<5;i++) for (j=0;j<10;j++) {x=100+random(400); fazhi=(double)x/50000000.00; b3[i][j]=fazhi; } } /*计算NN2网络单个样本的误差*/ void calculdiffer3(void) {a3=count3-1; differ3=0.5*(o3[4][1]-sample3[a3][3])*(o3[4][1]-sample3[a3][3]); } /*计算NN2网络全体样本均方差*/ void calculis3(void) {int i; is3s=0.0; for (i=0;i<=9;i++) {o3[1][1]=sample3[i][0];o3[1][2]=sample3[i][1];o3[1][3]=sample3[i][2]; netout3(8,1); is3s=is3s+(o3[4][1]-sample3[i][3])*(o3[4][1]-sample3[i][3]); } is3s=is3s/10; } /*训练NN2网络*/ void trainnn3(void) {long int time; int i,x3[4]; initialw3(); initialb3(); for (time=1;time<=MAX3;time++) {count3=0; while(count3<=9) {o3[1][1]=sample3[count3][0];o3[1][2]=sample3[count3][1]; o3[1][3]=sample3[count3][2]; count3=count3+1; clearwc3(); clearbc3(); netout3(8,1); calculdiffer3(); while (differ3>ERROR3) { calculd3(8,1); calculwc3(8,1); calculbc3(8,1); changw3(8,1); changb3(8,1); netout3(8,1); calculdiffer3(); } } printf("This is %d times training NN2...\n",time); calculis3(); printf("is3s==%f\n",is3s); if(is3s<ACCURACY) break; } } /*计算NN1网络的各层输出*/ void netout2(int m,int n) { int i,j,l; for (j=1,i=2;j<=m;j++) {netin2[i][j]=0.0; for(l=1;l<=3;l++) netin2[i][j]=netin2[i][j]+o2[i-1][l]*w2[i][l][j]; netin2[i][j]=netin2[i][j]-b2[i][j]; o2[i][j]=A/(1+exp(-netin2[i][j]/B)); } for (j=1,i=3;j<=m;j++) {netin2[i][j]=0.0; for(l=1;l<=m;l++) netin2[i][j]=netin2[i][j]+o2[i-1][l]*w2[i][l][j]; netin2[i][j]=netin2[i][j]-b2[i][j]; o2[i][j]=A/(1+exp(-netin2[i][j]/B)); } for (j=1,i=4;j<=n;j++) {netin2[i][j]=0.0; for(l=1;l<=m;l++) netin2[i][j]=netin2[i][j]+o2[i-1][l]*w2[i][l][j]; netin2[i][j]=netin2[i][j]-b2[i][j]; o2[i][j]=A/(1+exp(-netin2[i][j]/B)); } } /*计算NN1网络的反向传播误差*/ void calculd2(int m,int n) { int i,j,l; double t; a2=count2-1; d2[4][1]=(o2[4][1]-sample2[a2][3])*(A/B)*exp(-netin2[4][1]/B/pow(1+exp(-netin2[4][1]/B),2)); for (j=1,i=3;j<=m;j++) {t=0.00; for(l=1;l<=n;l++) t=t+w2[i+1][j][l]*d2[i+1][l]; d2[i][j]=t*(A/B)*exp(-netin2[i][j]/B/pow(1+exp(-netin2[i][j]/B),2)); } for (j=1,i=2;j<=m;j++) {t=0.00; for(l=1;l<=m;l++) t=t+w2[i+1][j][l]*d2[i+1][l]; d2[i][j]=t*(A/B)*exp(-netin2[i][j]/B/pow(1+exp(-netin2[i][j]/B),2)); } } /*计算NN1网络的权值变化量*/ void calculwc2(int m,int n) { int i,j,l; for (i=1,l=4;i<=m;i++) for(j=1;j<=n;j++) wc2[l][i][j]=-COEF2*d2[l][j]*o2[l-1][i]+0.5*wc2[l][i][j]; for (i=1,l=3;i<=m;i++) for(j=1;j<=m;j++) wc2[l][i][j]=-COEF2*d2[l][j]*o2[l-1][i]+0.5*wc2[l][i][j]; for (i=1,l=2;i<=3;i++) for(j=1;j<=m;j++) wc2[l][i][j]=-COEF2*d2[l][j]*o2[l-1][i]+0.5*wc2[l][i][j]; } /*计算NN1网络各节点的阈值变化量*/ void calculbc2(int m,int n) {int j; for (j=1;j<=m;j++) {bc2[2][j]=COEF2*d2[2][j]; bc2[3][j]=COEF2*d2[3][j]; } for (j=1;j<=n;j++) bc2[4][j]=COEF2*d2[4][j]; } /*计算NN1网络的权值*/ void changw2(int m,int n) {int k,i,j; for (i=1,k=2;i<=3;i++) w2[k][i][j]=w2[k][i][j]+wc2