MATLAB源代码，PCA优化RBF神经网络源代码

%杨铮 %河钢数字信达科技 clear; load hbis.txt;%5000行15列数据 Pall=hbis(:,1:14); Tall=hbis(:,15); X=Pall; Y=Tall; [mX,nX]=size(X); [mY,nY]=size(Y); [Xn,meanX,stdX,Yn,meanY,stdY] = premnmx(X',Y'); nTest=150; %Xn1=Xn(:,1:(mX-nTest)); %Xn2=Xn(:,(mX-nTest+1):mX); Yn1=Yn(:,1:(mX-nTest)); Yn2=Yn(:,(mX-nTest+1):mX); Y1=Y(1:(mX-nTest),:); Y2=Y((mX-nTest+1):mX,:); %Preprocess of data [T,P,W] = pca(Xn',7); t1=T'; y1=Y'; [tn,mint,maxt,yn,miny,maxy] = premnmx(t1,y1); tn1=tn(:,1:(mX-nTest)); tn2=tn(:,(mX-nTest+1):mX); yn1=yn(:,1:(mX-nTest)); yn2=yn(:,(mX-nTest+1):mX); [S2,Q] = size(T'); %minmax returns the Rx2 matrix of minimum and maximum values for each %row of M. %Network Training eg = 0.8; % Sum-squared error goal sc = 0.75; % Spread constant net = newrb(tn1,yn1,eg,sc); %simulate the network: Ybn1 = sim(net,tn1); Yb1 = postmnmx(Ybn1,miny,maxy); Ybn2 = sim(net,tn2); Yb2 = postmnmx(Ybn2,miny,maxy); disp('fitting Error is:'); Err_Bp1=sqrt(sumsqr(Y1'-Yb1)/(size(Yn1,2))); %Err_Bp1=sumsqr(Y1'-Yb1)/(size(Yn1,2)); disp(Err_Bp1); e1=Y1'-Yb1; e1=abs(e1); er1=e1./Y1'; er1=sum(er1)/(size(er1,2)); disp(er1); er3=sum(e1)/(size(e1,2)); disp(er3); disp('testing Error is:'); Err_Bp2=sqrt(sumsqr(Y2'-Yb2)/(size(Yn2,2))); %Err_Bp2=sumsqr(Y2'-Yb2)/(size(Yn2,2)); disp(Err_Bp2); e2=Y2'-Yb2; e2=abs(e2); er2=e2./Y2'; er2=sum(er2)/(size(er2,2)); disp(er2); er4=sum(e2)/(size(e2,2)); disp(er4); figure(2); ct=1:1:(mX-nTest); plot(ct,Y1,'b:',ct,Yb1,'r-'); legend('Analysis value','PCA-RBF output'); xlabel('Sample No'); ylabel('Value'); %Network Testing figure(3); ct=(mX-nTest+1):1:Q; plot(ct,Y2,'b:',ct,Yb2,'r-'); legend('Analysis value','PCA-RBF output'); xlabel('Sample No'); ylabel('Value'); figure(4);%画出检验样本的拟合效果图 Yb3=[Yb1 Yb2]; ct=1:1:Q; plot(ct,Y,'r:',ct,Yb3,'b-'); legend('value desired','PCA-RBF out'); xlabel('sample No.'); ylabel('Value'); %save net.IW, net.b,net.LW RBFw.dat -ascii; %celldisp(net.IW);%Display IW; %celldisp(net.LW);%Display LW; %celldisp(net.b);%Display b;