whk_b.asv

魏海坤编著的《神经网络结构设计的理论与方法》 国防工业出版社出版

function main()
SamNum=100;
TestSamNum=101;
HiddenUnitNum=10;
InDim=1;
OutDim=1;

% 根据目标函数获得样本输入输出
rand('state',sum(100*clock))
NoiseVar=0.1;
Noise=NoiseVar*randn(1,SamNum);
SamIn=8*rand(1,SamNum)-4;
SamOutNoNoise=1.1*(1-SamIn+2*SamIn.^2).*exp(SamIn.^2/2);
SamOut=SamOutNoNoise+Noise;

TestSamIn=-4:0.08:4;
TestSamOut=1.1*(1-TestSamIn+2*TestSamIn.^2).*exp(TestSamIn.^2/2);

figure
hold on
grid
plot(SamIn,SamOut,'k+')
plot(TestSamIn,TestSamOut,'k--')
xlabel('Input x');
ylabel('Output y');

MaxEpochs=20000;
lr=0.003;
EO=0.5;

W1=0.2*rand(HiddenUnitNum,InDim)-0.1;
B1=0.2*rand(HiddenUnitNum,1)-0.1;
W2=0.2*rand(OutDim,HiddenUnitNum)-0.1;
B2=0.2*rand(OutDim,1)-0.1;

W1Ex=[W1 B1];
W2Ex=[W2 B2];

SamInEx=[SamIn' ones(SamNum,1)];
ErrHistory=[];

for i=1:MaxEpochs
    % 正向传播计算网络输出
    HiddenOut=logsig(W1Ex*SamInEx);
    HiddenOutEx=[HiddenOut' ones(SamNum,1)]';
    NetworkOut=W2Ex*HiddenOutEx;
    
    % 停止学习判新
    Error=SamOut-NetworkOut;
    SSE=sumsqr(Error)
    
    % 纪录每次权值调总后的训练误差
    ErrHistory=[ErrHistory SEE];
    
    if SEE<E0,break,end
    
    % 计算反向传播误差
    Delta2=Error;
    Delta1=W2'*Delta2.*HiddenOut.*(1-HiddenOut);
    
    % 计算权值调节量
    dW2Ex=Delta2*HiddenOutEx';
    dW1Ex=Delta1*SamInEx';
    
    % 权值调节
    W1Ex=W1Ex+lr*dW1Ex;
    W2Ex=W2Ex+lr*dW2Ex;
    
    % 分离隐层到输出层的初始权值，以便以后使用
    W2=W2Ex(:,1:HiddenUnitNum);
end

% 显示计算结果
i
W1=W1Ex(:,1:InDim)
B1=W1Ex(:,InDim+1)
W2
B2=W2Ex(:,1+HiddenUnitNum);

% 测试
TestHiddenOut=logsig(W1*TestSamIn+repmat(B1,1,TestSamNum));
TestNNOut=W2*TestHiddenOut+repmat(B2,1,TestSamNum);
plot(TestSamIn,TestNNOut,'k-')

% 绘制学习误差曲线
figure
hold on
grid
[xx,Num]=size(ErrHistory);
plot(1:Num,ErrHistory,'k-');