%BP neural network
clear
inputNums=5;
outputNums=6;
hideNums=3;
maxcount=2000;
samplenum=3;
precision=0.001;
alpha=0.5;%设定值
a=0.5;%设定值
count=1;
error=zeros(1,count);
errorp=zeros(1,samplenum);
v=rand(inputNums,hideNums);
deltv=zeros(inputNums,hideNums);
dv=zeros(inputNums,hideNums);
st1=zeros(1,hideNums);%隐含层的偏置函数
w=rand(hideNums,outputNums);
deltw=zeros(hideNums,outputNums);
dw=zeros(hideNums,outputNums);
st2=zeros(1,outputNums); %输出层的偏置函数
load Data;
samplelist=Data(:,1:4)';%[1,0,0;0,1,0;0,0,1];
expectlist=Data(:,5:7)';%[1,0,0;0,1,0;0,0,1];
while (count<=maxcount)
c=1;
while (c<=samplenum)
for k=1:outputNums
d(k)=expectlist(c,k);%获得期望输出的向量
end
for i=1:inputNums
x(i)=samplelist(c,i);%获得输入的向量（数据）
end
%Forward();
for j=1:hideNums
net=0.0;
for i=1:inputNums
net=net+x(i)*v(i,j);
end
net=net-st1(j);
y(j)=1/(1+exp(-net)); %隐含层的函数
end
for k=1:outputNums
net=0.0;
for j=1:hideNums
net=net+y(j)*w(j,k);
end
net=net-st2(k);
o(k)=1/(1+exp(-net));
end
%BpError(c);
errortmp=0.0;
for k=1:outputNums
errortmp=errortmp+(d(k)-o(k))^2;
end
errorp(c)=0.5*errortmp;
%end
%Backward();
for k=1:outputNums
yitao(k)=(d(k)-o(k))*o(k)*(1-o(k));
end
for j=1:hideNums
tem=0.0;
for k=1:outputNums
tem=tem+yitao(k)*w(j,k);
end
yitay(j)=tem*y(j)*(1-y(j));
end
%调整各层权值
for j=1:hideNums
for k=1:outputNums
deltw(j,k)=alpha*yitao(k)*y(j);
w(j,k)=w(j,k)+deltw(j,k)+a*dw(j,k);
dw(j,k)=deltw(j,k);
end
end
for k=1:outputNums
st2(k)=st2(k)+alpha*yitao(k);
end
for i=1:inputNums
for j=1:hideNums
deltv(i,j)=alpha*yitay(j)*x(i);
v(i,j)=v(i,j)+deltv(i,j)+a*dv(i,j);
dv(i,j)=deltv(i,j);
end
end
for j=1:hideNums
st1(j)=st1(j)+alpha*yitay(j);
end
%end
c=c+1;
end
double tmp;
tmp=0.0;
for i=1:samplenum
tmp=tmp+errorp(i)*errorp(i);
end
tmp=tmp/c;
error(count)=sqrt(tmp);
if (error(count)<precision)
break;
end
count=count+1;%训练次数加1
end
p=1:count
plot(p,error(count),'-');
评论0