11.m

RBF实现神经网络预测程序

data=[-0.22914	-0.059577	-0.88853	-0.25353	-0.28422	0.096958;
-0.48264	0.63645	0.025257	-0.36327	0.47082	-0.087841;
-0.4228	0.28586	0.34107	0.048845	-0.76611	0.18407;
-0.38717	-0.46772	0.14487	-0.23132	-0.089108	-0.74088;
-0.40248	-0.5382	0.19561	-0.22917	0.24525	0.63041;
-0.47317	-0.035252	-0.18518	0.83387	0.20625	-0.051838];
x=data(:,1:5);t=data(:,6);
c1=x(1:3,:);  %初值
%第一次分类
for i=1:1:6
     for j=1:1:3
    d1(i,j)=(x(i,:)-c1(j,:))*((x(i,:)-c1(j,:))');
end
end
% x1   d1=[0	   1.9658	 1.9924;     
 % x2    1.9658	0	     1.9261;
 % x3    1.9924	1.9261	  0
 % x4    1.298	1.5735	 1.1445
 % x5    1.7154	1.4841	 1.8008
 % x6    1.9779	1.9987	 1.9443]     
%类1（x1)
%类2（x2,x5)
%类3（x3,x4,x6)
%中心c1=[-0.22914	 -0.059577	 -0.88853	 -0.25353	 -0.28422
%       -0.48264	0.63645	   0.025257	   -0.36327	   0.47082
%       -0.4228	    0.28586	   0.34107	   0.048845	   -0.76611]

%第二次聚类
%新的中心c
c2=[x(1,:);0.5*(x(2,:)+x(5,:));(x(3,:)+x(4,:)+x(6,:))./3];
%c2=[ -0.22914	  -0.059577	 -0.88853	 -0.25353	 -0.28422
%     -0.44256	  0.049124	 0.11043   	 -0.29622	  0.35803
%    -0.42772	  -0.072373	  0.10025	  0.21713	-0.21632]
%求d2
for i=1:1:6
     for j=1:1:3
    d2(i,j)=(x(i,:)-c2(j,:))*((x(i,:)-c2(j,:))');
end
end
%x1      d2=[0	    1.4696	  1.2434
%x2          1.9658	0.37103	  1.3201
%x3          1.9924	1.4924	  0.51694
%x4          1.298	0.47553	  0.37723
%x5          1.7154	0.37103	  0.63896
%x6          1.9779	1.3956	  0.64385]
%聚类结果
%类1(x1)
%类2(x2,x5）
%类3(x3,x4,x6)
%
%可知，c不再变化，故，分类结束
%
%最后的结果：c＝[ -0.22914	  -0.059577	 -0.88853	 -0.25353	 -0.28422
%                -0.44256	  0.049124	 0.11043   	 -0.29622	  0.35803
%                -0.42772	  -0.072373	  0.10025	  0.21713	-0.21632]
%  求RBF基函数的宽度delta
dd2=d2';
dsum=sum(dd2);
delta=[dsum(1) 0.5*(dsum(2)+dsum(5)) (dsum(2)+dsum(4)+dsum(6))./3];
%delta=[2.713	3.1912	3.275]
%据上述可知，隐含层数（采用高斯核函数）为3，输出层为线性输出
for j=1:1:3
for i=1:1:6
a(j,i)=((x(i,:)-c2(j,:)))*((x(i,:)-c2(j,:))');
    a(j,i)=(-a(j,i)./delta(j));
end
end
P=1:1:6;
T=t;
%figure;subplot(2,2,1);plot(P,t);title('待逼近的函数样本点');
%axis([1,6,-1,1])
  p=a;
 r=radbas(p);
 err_goal=0.99;
 sc=1;
 net=newrb(p,T',err_goal,sc);
%NEWRB, neurons = 0, SSE = 0.999213
Y=sim(net,p);
%axis([1,6,-0.4,0.4]);
figure;
plot(P,T,'r');
hold on;
plot(P,Y,':*');
title('RBF网络拟合曲线图11');
legend('化验值','估计值');
ylabel('淀粉利用率（％）');
xlabel('样本个数');
axis([1,6,-1,1]);