神经网络实验报告BP网络实现对Hermit函数逼近.docx

实验报告
——基于BP网络的多层感知器
电气工程及其自动化四班
20080240402 曹建
20080240406 邓宏
一、神经网络结构图
O
输出层
W

隐层

Y0 V
输入层
X0 X
二、程序
:
function main()
clc
close all
p=1;Pmax=100 ;
q=1;Qmax=30000;
n=;l=5;
Emin=;
Erme=[];
rand('state',sum(100*clock));
w=rand(1,l);w0=;
v=rand(1,l ); v0=rand(1,l);
y0=-1;x0=-1;

for p=1:Pmax
interrupt(p)=normrnd(0,);
x(p)=8*(rand()-);
d(p)=*(1-x(p)+2*x(p)^2)*exp((-x(p)^2)/2);
end

while(q<=Qmax)
for p=1:Pmax

y(p,:)=logsig(v*x(p)+v0*x0);
o1(p)=w*y(p,:)'+y0*w0;
o(p)=o1(p)+interrupt(p);
ek=d(p)-o(p);
for i=1:l
ej(i)=w(i)*(d(p)-o(p))*y(p,i)*(1-y(p,i));
end
wc=n*ek*y(p,:);w0c=n*ek*y0;vc=n*ej*x(p);v0c=n*ej*x0;
w=w+wc;w0=w0+w0c;v=v+vc;v0=v0+v0c;
E(p)=d(p)-o(p);

end
s2=0;
for p=1:Pmax
s2=E(p)^2+s2;
end
e=(s2/Pmax)^;
Erme=[Erme e];
if e<=Emin
break;
end
q=q+1;
end

Erme;
q
x=linspace(-4,4);
for p=1:100
d(p)=*(1-x(p)+2*x(p)^2)*exp((-x(p)^2)/2);
end
for p=1:100
y(p,:)=logsig(v*x(p)+v0*x0);
o1(p)=w*y(p,:)'+y0*w0;
end
plot(Erme);
xlabel('误差的收敛曲线')
figure;
plot(x,d,'-r');
hold on;
plot(x,o1,'-.b');
xlabel('Hermit多项式曲线与所构建BP网络输出曲线')
最好的一次运行结果:学习率=,隐节点数=5
q =
2500
:
function main()
clc
close all
samnum=100;
testsamnum=101;
hiddenunitnum=5;
indim=1;outdim=1;
rand('state',sum(100*clock))
noisevar=;
noise=noisevar*randn(1,samnum);
samin=8*rand(1,samnum)-4;
samoutnonoise=*(1-samin+2*samin.^2).*exp(-samin.^2/2);
samout=samoutnonoise+noise;
testsamin=-4::4;
testsamout=*(1-testsamin+2*testsamin.^2).*exp(-testsamin.^2/2);
figure
hold on
grid
plot(testsamin,testsamout,'k--')
xlabel('inputx');ylabel('outputy');
maxepochs=30000;
lr=; alpha=;
e0=;
w1=*rand(hiddenunitnum,indim);
b1=*rand(hiddenunitnum,1);
w2=*rand(outdim,hiddenunitnum);
b2=*rand(outdim,1);
w1ex=[w1 b1];
w2ex=[w2 b2];
dw1ex=zeros(hiddenunitnum,2);
dw2ex=zeros(1,hiddenunitnum+1);
saminex