Particle swarm optimization implementation error

@all please I need your suggestion and help on this code using pso

for num_acur= 1:1:num_aopt
    options=optimset('MaxIter',400);
    a_init=1*rand(4,1);
    [a_opt, Fval, exitFlag]=fminunc(@W,a_init,options); 
    a_result(num_acur,:)=[a_opt',Fval,a_init',exitFlag];
end

Rewriting this Using Pso

for num_acur = 1:1:num_aopt
    fun = @W;
    lb = [-30,-30,-30,-30];
    ub =[100,10,10,10];
   % options = optimoptions('particleswarm','SwarmSize',100);
   %options = optimoptions('particleswarm','SwarmSize',50,'HybridFcn',@fmincon);
   nvars = 4;
   [a_opt, Fval, exitFlag] = particleswarm(fun,nvars,lb,ub);%,options);
   a_result(num_acur,:)=[a_opt',Fval,lb',exitFlag];
end

Error

Error using horzcat

Dimensions of arrays being concatenated are not consistent.

Error in code ()

a_result(num_acur,:)=[a_opt',Fval,lb',exitFlag];

Answer 1

[Walter Roberson]

   [a_opt, Fval, exitFlag] = particleswarm(fun,nvars,lb,ub);%,options);
   a_result(num_acur,:)=[a_opt',Fval,lb',exitFlag];
   

The first output from particleswarm is a row vector with length equal to the number of variables -- so 1 x 4 in your case. The second output is the scalar function result.

You use conjugate transpose on the 1 x 4, which will give you a 4 x 1. You use it with [] with the scalar Fval, so you are trying to horizontally concatenate a vector with 4 rows and a vector with 1 row. That is not a permitted operation.

You need to either not transpose a_opt and lb, or else you need to use ; instead of , in the [] so that you construct a column.

[Odesanmi Gbenga Abiodun]

Thanks for your response,this really work but it doesnt solve my problem yet. the output seems to converge easilly I dont know why... and while using fminuc it responds to nesseary input.

any suggestion please??

[Walter Roberson]

There are relatively few kinds of functions for which PSO is known to converge. It is not necessarily all that good for other kinds of functions.

It would help to have the code for W to test with.

[Odesanmi Gbenga Abiodun]

x1=[0.89 0.7 0.6 0.4 0.5 0.2];
x2=[0.02 0.05 -0.02 0.01 -0.3 -0.03];
y=[0.7;1.6;0.9;1;0.2;0.4];
x=[x1;x2];
function  A=W(k)
t=size(x,2);
Kern_Cov=ones(t,t);
for i=1:1:t
     for j=1:1:t
           Kern_Cov(i,j)=k(1).^2*exp((-1/2)*((x(:,i)-x(:,j))'*inv(diag([k(2).^2,k(3).^2]))*(x(:,i)-x(:,j))))+k(4).^2*uita(x(:,i),x(:,j));
     end
end
    
A=-(-1/2.*(y'*inv(Kern_Cov)*y)-1/2*log(det(Kern_Cov))-t/2*log(2*pi));
end
 
function  guita=uita(x1,x2)
        if(x1==x2)
            guita=1;
        else
            guita=0;
        end
 
end

[Walter Roberson]

That code looks like it will give an error as x x1 and x2 are not defined inside the functions. You are using some of the variables as if they are shared variables, but I do not not see any enclosing function.

[Odesanmi Gbenga Abiodun]

exactly but each time i compute the next point of x1 and x2 using this code below, it converge

for num_acur = 1:1:num_aopt
    fun = @W;
    lb = [-30,-30,-30,-30];
    ub =[100,10,10,10];
   % options = optimoptions('particleswarm','SwarmSize',100);
   %options = optimoptions('particleswarm','SwarmSize',50,'HybridFcn',@fmincon);
   nvars = 4;
   [a_opt, Fval, exitFlag] = particleswarm(fun,nvars,lb,ub);%,options);
   a_result(num_acur,:)=[a_opt',Fval,lb',exitFlag];
end

[Walter Roberson]

Are you defining W as a nested function with x being a shared variable used in W ? If so then we need your actual code.

[Odesanmi Gbenga Abiodun]

No, their is another function for the nested x

Tota_number =find(a_result(:,5)==min(a_result(:,5)));
a_prama_tra=a_result(Tota_number,1:4);
a_prama = a_prama_tra
%[opt_x_current, fun_negEI, exitFlag]=fmincon(@nested,opt_x_init,[-1,1.6;1,-1.6],[-0.2;1],[],[],[G_low,E_low],[G_high,E_high],[]); 
[opt_x_current, fun_negE, exitFlag]=particleswarm(@nested,2,lb,ub);
opt_x_nested=opt_x_current;

Bys_flag=0;
x=[x,opt_x_nested];

function  Kern_Cov_xixj=Aug(xi,xj)

        global a_prama       
        Kern_Cov_xixj=(a_prama(1)).^2*exp(-(1/2)*((xi-xj)'*inv(diag([(a_prama(2)).^2,( a_prama(3)).^2]))*(xi-xj)))+a_prama(4).^2*uita(xi,xj); 
        
end


function E_ne_o=nested(X_opt)

global  x y E_m E_v 
y_n=size(y,1); 
K=ones(y_n,y_n);
for i=1:1:y_n
    for j=1:1:y_n
        K(i,j)=Aug(x(:,i),x(:,j));     
    end
end




 k=ones(y_n,1);
 for i=1:1:y_n
     k(i,1)=Aug(X_opt,x(:,i)); 
 end
           

           E_m=k'*inv(K)*y;
           E_v=Aug(X_opt,X_opt)-k'*inv(K)*k;
           if(E_v>0)
               f_best=min(y);
               u_star=(f_best-E_m)/sqrt(E_v);
               E_o=(f_best-E_m).*normcdf(u_star)+sqrt(E_v).*normpdf(u_star);

           else
                E_o=0;
           end
            E_ne_o=-E_o;   
           
end

[Walter Roberson]

There would need to be a function header at the top and an end at the bottom of that in order for that to make use of nested functions and shared variables.