Error in user-supplied fitness function evaluation – Genetic Algorithm
Hello everyone. I am currently using the ga function in order to find the best parameters for a discrete GPC controller. However, more often than not, the execution of this algorithm ends up in error.
First, let me show the fitness function that I am currently using:
function J = simplanta(individuo)
Np = individuo(1);
Nc = individuo(2);
rw = individuo(3);
ts=1/2000; % sampling time
t=0:ts:1; % sim time
f0=50;
r=1+1.5.*sin(2*pi*f0*t); % reference signal
Ldata=length(t); % # of time points
%Discrete Model
Ad =[0.4134 0.4536 0; -0.2404 0.7876 0; -0.4366 0.4224 0.7738];
Bd =[0.1331; 0.4528; 0.1273];
Cd=[0 0 1];
% Specified Gain Calc
n=length(Ad); %order%
Omega = 2*pi*f0*ts;
gamma = ( 2*cos(Omega) ) + 1;
E=[0 1 0; 0 0 1; 1 -gamma gamma];
sigma = [0;0;1];
% prediction matrices
A=[Ad zeros(n,3); sigma*Cd*Ad E];
B=[Bd; sigma*Cd*Bd];
C=[zeros(1,3) sigma’ ];
% MPC
nau = length(A);
Rb = rw*eye(Nc,Nc);
% F e Phi
F=zeros(Np,nau);
for i=1:Np
F(i,:)=C*(A^i);
end
Phi = zeros(Np, Nc);
%1st column:
for linha=1:Np
Phi(linha,1) = C*(A^(linha-1))*B;
end
%other column
for col =2: Nc
Phi(col:Np, col) = Phi(1:Np-(col-1),1);
end
PTP = Phi’*Phi;
if (Nc == 1)
vetor10 = [1];
else
vetor10 = [ 1 zeros(1,Nc-1)];
end
Hessiana=inv(PTP + Rb);
Kbs = -1*vetor10*Hessiana*(Phi’)*F;
% control sim
x = zeros(n,Ldata+1); %
y = zeros(1,Ldata); %
erro =zeros(1,Ldata); %
u =zeros(1,Ldata); %
xk_1=zeros(n,1); %x(k-1)
xk_2=zeros(n,1); %x(k-2)
xk_3=zeros(n,1); %x(k-3)
uk_1=0; %u(k-1)
uk_2=0; %u(k-2)
uk_3=0; %u(k-3)
ek_1=0; %e(k-1)
ek_2=0; %e(k-2)
% closed loop sim
for k=1:Ldata
xk = x(:,k); %
rk = r(k);
yk = Cd*xk; %
ek = rk-yk; %
qk = xk_3+(-gamma*xk_2)+(gamma*xk_1)-xk;
xbarra= [qk;ek_2;ek_1;ek];
vk=Kbs*xbarra;
uk= uk_3 + gamma*(uk_1-uk_2)-vk;
%
x(:,k+1) = (Ad*xk)+(Bd*uk);
%
erro(k) = ek;
u(k) = uk;
y(k) = yk;
%
xk_3 = xk_2;
xk_2 = xk_1;
xk_1 = xk;
uk_3 = uk_2;
uk_2 = uk_1;
uk_1 = uk;
ek_2 = ek_1;
ek_1 = ek;
end
%Fit func
up = max(abs(u))/1000; %peak control law
erro_rmse = rmse(r,y); % RMSE
% settling time
max_e = max(erro); %
valor_acom = 0.02*max_e; %
ind_acom = find(erro >= -valor_acom & erro <= valor_acom, 1); %
t_ae = t(ind_acom); %
J=0.2*up + 0.7*erro_rmse + 0.1*t_ae;
With this fitness function I am calling the ga function with the following parameters:
A_des = [0 0 -1; -1 1 0; 1 0 0; 0 -1 0];
b_des = [0 0 100 -2]’;
intcon = [1 2]; os
lb = [2 2 0];
ub = [100 99 60];
%Declaração das opções de funcionamento do AG
options = optimoptions("ga","CreationFcn","gacreationuniformint",…
"Display","iter","MutationFcn","mutationpower",…
"SelectionFcn","selectionroulette",…
"PlotFcn",["gaplotdistance", "gaplotselection","gaplotstopping",…
"gaplotbestf","gaplotbestindiv"]);
[best_ind, cost] = ga(@simplanta, 3, A_des, b_des, [], [], lb, ub, [],…
intcon, options)
param = infoPlant(best_ind)
However, sometimes (and to be quite frankly more often than not) MATLAB shows me this error during the execution of this code:
Unable to perform assignment because the size of the left side is 1-by-1 and the
size of the right side is 1-by-0.
Error in fcnvectorizer (line 19)
y(i,:) = feval(fun,(pop(i,:)));
Error in gaminlppenaltyfcn
Error in gapenalty
Error in stepGA (line 63)
nextScore = FitnessFcn(pop);
Error in galincon (line 86)
[score,population,state] = stepGA(score,population,options,state,GenomeLength,FitnessFcn);
Error in gapenalty
Error in ga (line 412)
[x,fval,exitFlag,output,population,scores] = gapenalty(FitnessFcn,nvars,…
Error in ga_implementation (line 60)
[best_ind, cost] = ga(@simplanta, 3, A_des, b_des, [], [], lb, ub, [],…
Caused by:
Failure in user-supplied fitness function evaluation. GA cannot continue.
I am quite frankly at loss as of why that would happen, since it sometimes runs smoothly, so it seems that it isn’t necessarily a code syntax error of my part. Or perhaps it IS and I am just not able to detect it.
Any help and input would be quite appreciatedHello everyone. I am currently using the ga function in order to find the best parameters for a discrete GPC controller. However, more often than not, the execution of this algorithm ends up in error.
First, let me show the fitness function that I am currently using:
function J = simplanta(individuo)
Np = individuo(1);
Nc = individuo(2);
rw = individuo(3);
ts=1/2000; % sampling time
t=0:ts:1; % sim time
f0=50;
r=1+1.5.*sin(2*pi*f0*t); % reference signal
Ldata=length(t); % # of time points
%Discrete Model
Ad =[0.4134 0.4536 0; -0.2404 0.7876 0; -0.4366 0.4224 0.7738];
Bd =[0.1331; 0.4528; 0.1273];
Cd=[0 0 1];
% Specified Gain Calc
n=length(Ad); %order%
Omega = 2*pi*f0*ts;
gamma = ( 2*cos(Omega) ) + 1;
E=[0 1 0; 0 0 1; 1 -gamma gamma];
sigma = [0;0;1];
% prediction matrices
A=[Ad zeros(n,3); sigma*Cd*Ad E];
B=[Bd; sigma*Cd*Bd];
C=[zeros(1,3) sigma’ ];
% MPC
nau = length(A);
Rb = rw*eye(Nc,Nc);
% F e Phi
F=zeros(Np,nau);
for i=1:Np
F(i,:)=C*(A^i);
end
Phi = zeros(Np, Nc);
%1st column:
for linha=1:Np
Phi(linha,1) = C*(A^(linha-1))*B;
end
%other column
for col =2: Nc
Phi(col:Np, col) = Phi(1:Np-(col-1),1);
end
PTP = Phi’*Phi;
if (Nc == 1)
vetor10 = [1];
else
vetor10 = [ 1 zeros(1,Nc-1)];
end
Hessiana=inv(PTP + Rb);
Kbs = -1*vetor10*Hessiana*(Phi’)*F;
% control sim
x = zeros(n,Ldata+1); %
y = zeros(1,Ldata); %
erro =zeros(1,Ldata); %
u =zeros(1,Ldata); %
xk_1=zeros(n,1); %x(k-1)
xk_2=zeros(n,1); %x(k-2)
xk_3=zeros(n,1); %x(k-3)
uk_1=0; %u(k-1)
uk_2=0; %u(k-2)
uk_3=0; %u(k-3)
ek_1=0; %e(k-1)
ek_2=0; %e(k-2)
% closed loop sim
for k=1:Ldata
xk = x(:,k); %
rk = r(k);
yk = Cd*xk; %
ek = rk-yk; %
qk = xk_3+(-gamma*xk_2)+(gamma*xk_1)-xk;
xbarra= [qk;ek_2;ek_1;ek];
vk=Kbs*xbarra;
uk= uk_3 + gamma*(uk_1-uk_2)-vk;
%
x(:,k+1) = (Ad*xk)+(Bd*uk);
%
erro(k) = ek;
u(k) = uk;
y(k) = yk;
%
xk_3 = xk_2;
xk_2 = xk_1;
xk_1 = xk;
uk_3 = uk_2;
uk_2 = uk_1;
uk_1 = uk;
ek_2 = ek_1;
ek_1 = ek;
end
%Fit func
up = max(abs(u))/1000; %peak control law
erro_rmse = rmse(r,y); % RMSE
% settling time
max_e = max(erro); %
valor_acom = 0.02*max_e; %
ind_acom = find(erro >= -valor_acom & erro <= valor_acom, 1); %
t_ae = t(ind_acom); %
J=0.2*up + 0.7*erro_rmse + 0.1*t_ae;
With this fitness function I am calling the ga function with the following parameters:
A_des = [0 0 -1; -1 1 0; 1 0 0; 0 -1 0];
b_des = [0 0 100 -2]’;
intcon = [1 2]; os
lb = [2 2 0];
ub = [100 99 60];
%Declaração das opções de funcionamento do AG
options = optimoptions("ga","CreationFcn","gacreationuniformint",…
"Display","iter","MutationFcn","mutationpower",…
"SelectionFcn","selectionroulette",…
"PlotFcn",["gaplotdistance", "gaplotselection","gaplotstopping",…
"gaplotbestf","gaplotbestindiv"]);
[best_ind, cost] = ga(@simplanta, 3, A_des, b_des, [], [], lb, ub, [],…
intcon, options)
param = infoPlant(best_ind)
However, sometimes (and to be quite frankly more often than not) MATLAB shows me this error during the execution of this code:
Unable to perform assignment because the size of the left side is 1-by-1 and the
size of the right side is 1-by-0.
Error in fcnvectorizer (line 19)
y(i,:) = feval(fun,(pop(i,:)));
Error in gaminlppenaltyfcn
Error in gapenalty
Error in stepGA (line 63)
nextScore = FitnessFcn(pop);
Error in galincon (line 86)
[score,population,state] = stepGA(score,population,options,state,GenomeLength,FitnessFcn);
Error in gapenalty
Error in ga (line 412)
[x,fval,exitFlag,output,population,scores] = gapenalty(FitnessFcn,nvars,…
Error in ga_implementation (line 60)
[best_ind, cost] = ga(@simplanta, 3, A_des, b_des, [], [], lb, ub, [],…
Caused by:
Failure in user-supplied fitness function evaluation. GA cannot continue.
I am quite frankly at loss as of why that would happen, since it sometimes runs smoothly, so it seems that it isn’t necessarily a code syntax error of my part. Or perhaps it IS and I am just not able to detect it.
Any help and input would be quite appreciated Hello everyone. I am currently using the ga function in order to find the best parameters for a discrete GPC controller. However, more often than not, the execution of this algorithm ends up in error.
First, let me show the fitness function that I am currently using:
function J = simplanta(individuo)
Np = individuo(1);
Nc = individuo(2);
rw = individuo(3);
ts=1/2000; % sampling time
t=0:ts:1; % sim time
f0=50;
r=1+1.5.*sin(2*pi*f0*t); % reference signal
Ldata=length(t); % # of time points
%Discrete Model
Ad =[0.4134 0.4536 0; -0.2404 0.7876 0; -0.4366 0.4224 0.7738];
Bd =[0.1331; 0.4528; 0.1273];
Cd=[0 0 1];
% Specified Gain Calc
n=length(Ad); %order%
Omega = 2*pi*f0*ts;
gamma = ( 2*cos(Omega) ) + 1;
E=[0 1 0; 0 0 1; 1 -gamma gamma];
sigma = [0;0;1];
% prediction matrices
A=[Ad zeros(n,3); sigma*Cd*Ad E];
B=[Bd; sigma*Cd*Bd];
C=[zeros(1,3) sigma’ ];
% MPC
nau = length(A);
Rb = rw*eye(Nc,Nc);
% F e Phi
F=zeros(Np,nau);
for i=1:Np
F(i,:)=C*(A^i);
end
Phi = zeros(Np, Nc);
%1st column:
for linha=1:Np
Phi(linha,1) = C*(A^(linha-1))*B;
end
%other column
for col =2: Nc
Phi(col:Np, col) = Phi(1:Np-(col-1),1);
end
PTP = Phi’*Phi;
if (Nc == 1)
vetor10 = [1];
else
vetor10 = [ 1 zeros(1,Nc-1)];
end
Hessiana=inv(PTP + Rb);
Kbs = -1*vetor10*Hessiana*(Phi’)*F;
% control sim
x = zeros(n,Ldata+1); %
y = zeros(1,Ldata); %
erro =zeros(1,Ldata); %
u =zeros(1,Ldata); %
xk_1=zeros(n,1); %x(k-1)
xk_2=zeros(n,1); %x(k-2)
xk_3=zeros(n,1); %x(k-3)
uk_1=0; %u(k-1)
uk_2=0; %u(k-2)
uk_3=0; %u(k-3)
ek_1=0; %e(k-1)
ek_2=0; %e(k-2)
% closed loop sim
for k=1:Ldata
xk = x(:,k); %
rk = r(k);
yk = Cd*xk; %
ek = rk-yk; %
qk = xk_3+(-gamma*xk_2)+(gamma*xk_1)-xk;
xbarra= [qk;ek_2;ek_1;ek];
vk=Kbs*xbarra;
uk= uk_3 + gamma*(uk_1-uk_2)-vk;
%
x(:,k+1) = (Ad*xk)+(Bd*uk);
%
erro(k) = ek;
u(k) = uk;
y(k) = yk;
%
xk_3 = xk_2;
xk_2 = xk_1;
xk_1 = xk;
uk_3 = uk_2;
uk_2 = uk_1;
uk_1 = uk;
ek_2 = ek_1;
ek_1 = ek;
end
%Fit func
up = max(abs(u))/1000; %peak control law
erro_rmse = rmse(r,y); % RMSE
% settling time
max_e = max(erro); %
valor_acom = 0.02*max_e; %
ind_acom = find(erro >= -valor_acom & erro <= valor_acom, 1); %
t_ae = t(ind_acom); %
J=0.2*up + 0.7*erro_rmse + 0.1*t_ae;
With this fitness function I am calling the ga function with the following parameters:
A_des = [0 0 -1; -1 1 0; 1 0 0; 0 -1 0];
b_des = [0 0 100 -2]’;
intcon = [1 2]; os
lb = [2 2 0];
ub = [100 99 60];
%Declaração das opções de funcionamento do AG
options = optimoptions("ga","CreationFcn","gacreationuniformint",…
"Display","iter","MutationFcn","mutationpower",…
"SelectionFcn","selectionroulette",…
"PlotFcn",["gaplotdistance", "gaplotselection","gaplotstopping",…
"gaplotbestf","gaplotbestindiv"]);
[best_ind, cost] = ga(@simplanta, 3, A_des, b_des, [], [], lb, ub, [],…
intcon, options)
param = infoPlant(best_ind)
However, sometimes (and to be quite frankly more often than not) MATLAB shows me this error during the execution of this code:
Unable to perform assignment because the size of the left side is 1-by-1 and the
size of the right side is 1-by-0.
Error in fcnvectorizer (line 19)
y(i,:) = feval(fun,(pop(i,:)));
Error in gaminlppenaltyfcn
Error in gapenalty
Error in stepGA (line 63)
nextScore = FitnessFcn(pop);
Error in galincon (line 86)
[score,population,state] = stepGA(score,population,options,state,GenomeLength,FitnessFcn);
Error in gapenalty
Error in ga (line 412)
[x,fval,exitFlag,output,population,scores] = gapenalty(FitnessFcn,nvars,…
Error in ga_implementation (line 60)
[best_ind, cost] = ga(@simplanta, 3, A_des, b_des, [], [], lb, ub, [],…
Caused by:
Failure in user-supplied fitness function evaluation. GA cannot continue.
I am quite frankly at loss as of why that would happen, since it sometimes runs smoothly, so it seems that it isn’t necessarily a code syntax error of my part. Or perhaps it IS and I am just not able to detect it.
Any help and input would be quite appreciated ga, fitness function MATLAB Answers — New Questions
