rbf_nn.m

類神經網路的RBF這對於任何研究都非常之友幫助歡迎下載內有說明

clc
clear all
close all
epoch=1000;
learn=0.5;
hidden_units=10; 
sigma=1;
rand('seed',123456);
x=(-10:1:10);
% m=[-8 -4 0 4 8];  %% 5
% m=[-8 -5 -2 0 2 5 8];  %% 7
% m=[-10 -8 -5 -2 0 2 5 8 10];  %% 9
m=[-10 -8 -7 -6 -4 -2 0 2 4 6 7 8 10];  %% 13
% m=[-10 -9 -8 -7 -6 -4 -3 -2 -1 0 1 2 3 4 6 7 8 9 10];  %% 19
% aa=-10:20/hidden_units:10;
% m=[aa];  %% 25
y=0.5.*(x.^3)-0.2.*(x.^2)-3.*x+20;
d=((y-(-470))/(470-(-470)))*0.6+0.2; 
b=1;
woh=0.5*rands(hidden_units,1); 

for l=1:epoch
     for j=1:hidden_units
        for i=1:21
            phi(i,j)=exp(-(norm(x(:,i)-m(:,j))^2)/(2*sigma^2));
            rr(i,j)=x(:,i)-m(:,j);
        end
     end
     
phi1=woh'*phi';
bb=b*ones(1,21);
F=phi1+bb;

h1=ones(1,hidden_units);
delta_woh=mean(learn*(d-F)'*h1.*phi);   
woh=woh+delta_woh';
delta_b=mean(learn*(d-F));
b=b+delta_b;

    for k=1:hidden_units
        ccc=learn*(d-F);
        delta_m=mean(ccc.*woh(k,:).*phi(:,k)'.*(1/(sigma^2)).*(rr(:,k))');
        m=m+delta_m;
        delta_sigma=mean(ccc.*woh(k,:).*phi(:,k)'.*(1/(sigma^3)).*(rr(:,k).^2)');
        sigma1=sigma*ones(1,hidden_units);
        sigma1=sigma1+delta_sigma;
        sigma=mean(sigma1);  
    end

eave(l)=mean((1/2)*(d-F).^2);
end

F_denormalize=(F-0.2)./0.6*940-470;

figure(1)
subplot(311)
plot(eave)
subplot(312)
plot(x,y)
hold on
plot(x,F_denormalize,'g*')
subplot(313)
plot(x,phi)

disp('Eave(end) =')
disp(eave(end))