whk_c.m

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;
InDim=1;
ClusterNum=10;
Overlap=1.0;

% 根据目标函数获得输入输出
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');

Centers=SamIn(:,1:ClusterNum);

NumberInClusters=zeros(ClusterNum,1);
IndexInClusters=zeros(ClusterNum,SamNum);
while 1,
    NumberInClusters=zeros(ClusterNum,1);
    IndexInClusters=zeros(ClusterNum,SamNum);
    
% 按最小距离原则对所有样本进行分类
  for i=1:SamNum
      AllDistance=Dist(Centers',SamIn(:,i));
      [MinDist,Pos]=min(AllDistance);
      NumberInClusters(Pos)=NumberInClusters(Pos)+1;
      InDexInClusters(Pos,NumberInClusters(Pos))=i;
  end

% 保存旧的聚类中心
OldCenters=Centers;

% 重新计算各类的聚类中心
  for i=1:ClusterNum
      Index=IndexInClusters(i,1:NumberInClusters(i));
      Centers(:,i)=mean(SamIn(:,Index)')';
  end

% 判断新旧聚类中心是否一致，如果是，则结束聚类
  EqualNum=sum(sum(Centers==OldCenters));
  if EqualNum==InDim*ClusterNum,
    break,
  end
end
% 计算各隐节电的扩散常数（宽度）
AllDistances=dist(Centers',Centers);
Maxinum=max(max(AllDistances));
for i=1:ClusterNum
    AllDistances(i,i)=Maxinum+1;
end
Spreads=Overlap*min(AllDistances)';

% 计算各隐节电的输出权值
Distance=dist(Centers,SamIn);
SpreadsMat=repmat(Spreads,1,SamNum);
HiddenUnitOut=radbas(Distance./SpreadsMat);
HiddenUnitOutEx=[HiddenUnitOut' ones(SamNum,1)]';
W2Ex=SamOut*pinv(HiddenUnitOutEx);
W2=W2Ex(:,1:ClusterNum);
B2=W2Ex(:,ClusterNum+1);

% 测试
TestDistance=dist(Centers',TestSamIn);
TestSpreadsMat=repmat(Spreads,1,TestSamNum);
TestHiddenUnitOut=radbas(TestDistance./TestSpreadsMat);
TestNNOut=W2*TestHiddenUnitOut+B2;
plot(TestSamIn,TestNNOut,'k-')

display(W2)
display(B2)