How Can I Customize Simulated Annealing Algorithm?

I have an optimization problem with discrete design variables, all of which are picked from a set {-45, 0, 45, 90}, to solve with built-in simulannealbnd function of Optimization Toolbox. In other words, optimal (x*) will be a combination of those values from the set. I read up on it from respective documentation and found out that it is by default for continuous variables. Again in the documentation there are ways to customize it so I decided to tweak with 'AnnealingFcn' option. This way I will randomly generate next iteration point as I wanted. I wrote my code as follows:

rng(7, 'twister')

ply_nums = 64;
E_11 = 127600 * 1e6;
E_22 = 13000 * 1e6;
G_12 = 6400 * 1e6;
v_12 = 0.3;
q = Q([E_11, E_22, G_12], v_12);

a = 0.508;
b = 0.254;
t = 0.127 * 1e-3;
z0 = 0;

N_xx = 1;
N_yy = 1;
N_xy = 0;

ply_qs = uniform_ply_qs(q, ply_nums);
ply_ts = uniform_ts(ply_nums, t);
ply_locations = PLY_ZS(ply_ts, z0);

lb = -90 * ones(1, ply_nums);
ub = 90 * ones(1, ply_nums);

values = [-45, 0, 45, 90];

x0 = values(randi(length(values), 1, ply_nums));

F = @(x) -lambda_critical(ply_qs, x, ply_locations, a, b, N_xx, N_yy, N_xy);

function newx = mynewx(optimValues, ~)
    
    newx = values(randi(length(optimValues.x)), 1, length(optimValues.x));

end

options = optimoptions(@simulannealbnd, ...
                             'Display', ...
                                'iter', ...
               'AnnealingFcn', @mynewx, ...
                'MaxIterations', 2000);

[x, F, exitflag, output] = simulannealbnd(F, x0, lb, ub, options);

disp('x : ')
disp(x)
disp('F(x) : ')
disp(F)

However, this code throws the following error when I run it:

Incorrect number or types of inputs or outputs for function values.

Error in sa_run>mynewx (line 40)
    newx = values(randi(length(optimValues.x)), 1, length(optimValues.x));

Error in globaloptim.simulannealbnd.sanewpoint (line 17)
     newx(:) = options.AnnealingFcn(optimvalues,problem);

Error in globaloptim.simulannealbnd.saengine (line 29)
    solverData = globaloptim.simulannealbnd.sanewpoint(solverData,problem,options);

Error in globaloptim.simulannealbnd.driver (line 28)
solverData = globaloptim.simulannealbnd.saengine(solverData,problem,options);

Error in simulannealbnd (line 197)
    globaloptim.simulannealbnd.driver(FUN, x0, [], [], [], [], lb, ub, options, defaultopt);

Error in sa_run (line 50)
[x, F, exitflag, output] = simulannealbnd(F, x0, lb, ub, options);

How can I solve the error here? What do I miss here?

Answer 1

[Abhimenyu]

Hi Burhan,

From the information shared, I could infer that you are solving an optimization problem with discrete design variables using a custom annealing function. The error message indicates that there is an issue with the function values returned by the custom annealing function, "mynewx". Specifically, the dimensions or types of the output are not as expected. The problem lies in the line where "newx" is updated, as "values(randi(length(optimValues.x)), 1, length(optimValues.x))" results in an incorrect shape for "newx".

To fix this, a single value must be generated from the "values" array for each design variable. Please refer to the below-mentioned corrected MATLAB code for the "mynewx" function:

%Corrected mynewx function
function newx = mynewx(optimValues, ~)
    newx = values(randi(length(values), 1, length(optimValues.x)));
end

In this corrected version, "randi(length(values), 1, length(optimValues.x))" generates a row vector of random integers where each integer is between 1 and the length of "values", and each integer corresponds to an index in "values". This vector is then used to index into "values", producing a new array "newx" where each element is randomly chosen from "values".

I hope this helps!

[Burhan Burak]

I have tried something else, which is a bit longer, to solve it. However, this helps too along with defining values = [-45, 0, 45, 90]; inside the function otherwise I get 'values' undefined error. Thanks a lot!