Plz tell why the output is not following the desired ?

clc
close all
clear all
%type1 network fuzzy

yd1=-.1;
yd0=-.07;
u3=0;
u2=0;
u1=0;
u0=0;
for t=1:1000
    if t<250
        u(t)=sin(pi*t/25);
    elseif t<500 && t>249
        u(t)=1;
    elseif t<750 && t>499
        u(t)=-1;
    else
        u(t)=0.3*sin(pi*t/25)+0.1*sin(pi*t/32)+0.6*sin(pi*t/10);
    end
end

yd(1)=0.72*yd0+0.025*yd1*u1+0.01*(u2)^2+0.2*u3;
yd(2)=0.72*yd(1)+0.025*yd0*u0+0.01*(u1)^2+0.2*u2;
yd(3)=0.72*yd(2)+0.025*yd(1)*u(1)+0.01*(u0)^2+0.2*u1;
yd(4)=0.72*yd(3)+0.025*yd(2)*u(2)+0.01*(u(1))^2+0.2*u0;
for t=4:999
    yd(t+1)=0.72*yd(t)+0.025*yd(t-1)*u(t-1)+0.01*(u(t-2))^2+0.2*u(t-3);
end
figure
plot(yd);

m=5;      %number of input 
U=.4;      %learning rate
n=length(yd);
c=0.4*ones(m,n); %mean of Gaussian mf
s=0.4*ones(m,n); %variance of Gaussian mf
%defining weight
W = 0.1*ones(m);

for i=1:m
    for j=1:n
        A(:,i)=exp(-0.5*((u(i)-c(:,i))./s(:,i)).^2); % n no. of mf
        R(i)=min(A(:,i));          %fuzzy rule
        y(i)=sum(u(i)*W(:,i));
    end
    for j=1:n
         B(i)=((sum(R(i)*y(i))/sum(R(i)))); %output of fuzzy nural structure
         e(j) = yd(i)-B(i);     %erroe
    end
end

%update rule for weight,mean ,variance
for i=1:m
    for j=1:n
        W(:,i) = W(:,i)+m*e(j)*u(i);
       c(:,i) = c(:,i) + (0.5*e(j)*(exp(-(u(i)-c(:,i)./s(:,i)).^2)));
       s(:,i) = s(:,i) + (0.5*e(j)*(exp(-(u(i)-c(:,i)./s(:,i)).^2)));
end

MSE(n+1-i) = sum(e.^2);
end

figure;
plot(u,'r') %output plot
figure;
plot(e,'k') %error plot
figure
plot(MSE) % meansquare error plot
e(i)^2

DESCRIPTION This is fuzzy neural network having n input. and by fuzzyfication and rule and defuzzyfication output is obtained. which is compared with desired and error is calculated. and also the weight, mean and variance is updated.

But problem is the out put is not following the desired. PLZ help.

[the cyclist]

I am able to run your code, and see four figures as a result.

Can you please describe in more detail what you mean by "output is not following the desired"?

[pinak parida]

thakyou sir, in the program the 'yd' is the desire output and 'B' is the output of fuzzy block. and the the output 'B' should be look alike 'yd'.

it is like this..... i have given input to fuuzy block whose output is 'B'.compared with the desired output 'yd'. error is obtained and fed to the fuuzy block to update the weights using simple LMS algorithm.

[pinak parida]

inside fuuzy block

[pinak parida]

x=[1 6]; mu=.004; yd=[15 21]; m=2; n=3; c=[1 2 3;4 5 6] s=[1 2 3;4 5 6] w=[1 2; 3 4; 5 6] %3x2 %A=[1 2 3; 4 5 6] %2x3 epoc=100; while(epoc) for j=1:m for i=1:n A(j,:)=exp(-0.5*((x(j)-c(j,:))./s(j,:)).^2); R(i)=min(A(:,i)); %1x3 y=w*x'; end p=y'*R'; Q=sum(p); V=sum®; B(j)=Q/V e(j)=yd(j)-B(j);

end w=w+mu*x*e'; c=c+mu*x*e'; s=s+mu*x*e'; epoc=epoc-1; plot(B) end

hi friend, what is the problem here that i am not getting the desired that is nearly equal to [15 21]. instead i get both the value same at last step what ever may be the epoc. plz help i could not find out it.

Answer 1

[Richard McCulloch]

I would check your indices. Try this:

clc
close all
clear all
%type1 network fuzzy
yd1=-.1;
yd0=-.07;
u3=0;
u2=0;
u1=0;
u0=0;
for t=1:1000
    if t<250
        u(t)=sin(pi*t/25);
    elseif t<500 && t>249
        u(t)=1;
    elseif t<750 && t>499
        u(t)=-1;
    else
        u(t)=0.3*sin(pi*t/25)+0.1*sin(pi*t/32)+0.6*sin(pi*t/10);
    end
end
yd(1)=0.72*yd0+0.025*yd1*u1+0.01*(u2)^2+0.2*u3;
yd(2)=0.72*yd(1)+0.025*yd0*u0+0.01*(u1)^2+0.2*u2;
yd(3)=0.72*yd(2)+0.025*yd(1)*u(1)+0.01*(u0)^2+0.2*u1;
yd(4)=0.72*yd(3)+0.025*yd(2)*u(2)+0.01*(u(1))^2+0.2*u0;
for t=4:999
    yd(t+1)=0.72*yd(t)+0.025*yd(t-1)*u(t-1)+0.01*(u(t-2))^2+0.2*u(t-3);
end
figure
plot(yd);
title('yd')
m=5;      %number of input 
U=.4;      %learning rate
n=length(yd);
c=0.4*ones(m,n); %mean of Gaussian mf
s=0.4*ones(m,n); %variance of Gaussian mf
%defining weight
W = 0.1*ones(m,1);
for i=1:m
    for j=1:n
        A(:,j)=exp(-0.5*((u(j)-c(:,i))./s(:,i)).^2); % n no. of mf
        R(j)=min(A(:,i));          %fuzzy rule
        y(j)=sum(u(j)*W(i));
    end
    for j=1:n
         B(j)=((sum(R.*y)./sum(R))); %output of fuzzy nural structure
         e(j) = yd(j)-B(j);     %erroe
    end
end
%update rule for weight,mean ,variance
for i=1:m
    for j=1:n
        W(i) = W(i)+m*e(j)*u(i);
       c(:,i) = c(:,i) + (0.5*e(j)*(exp(-(u(i)-c(:,i)./s(:,i)).^2)));
       s(:,i) = s(:,i) + (0.5*e(j)*(exp(-(u(i)-c(:,i)./s(:,i)).^2)));
end
MSE(n+1-i) = sum(e.^2);
end
figure;
plot(u,'r') %output plot
title('u')
figure;
plot(e,'k') %error plot
title('error')
figure
plot(MSE) % meansquare error plot
title('MSE')
figure
plot(B)
title('B')
e(i)^2

It kind of works, but it gives B a value of zero, so your error ends up being yd. If you can reverse that you're set, but try checking the indices, especially in the "j" loops.

[pinak parida]

thanks for your response

but i need 'B' similar to 'yd' but it is zero. i take your advice checking with 'j'.

[pinak parida]

CAN'T. can anyone detect the fault of solve.

[Jan]

When you omit the useless clear all you can set some breakpoints in the code and use the debugger to see, what's going on inside.