how to calculate fitness function when simulation is running?

Hi everyone, I have a simulink model and it returns two output. One of them is real and the other is predicted. Both of them have 500*1 size. For the optimization I wrote genetic algorithm code. There is no problem about the ga code. I have a problem writing my cost function. When i run the ga.m my fitness function returns constant value. But it is impossible, for 500 values it must be varied and converge to 0 in optimization process (am i wrong?). Could you please help me?

this is my fitness function

function MSE = optimizationfunc(real,predicted)

real = [2;3;6;.....]; %it has 500 values and i didnt write it here
predicted = get_param('my_sys_simulink/predicted','predicted');

MSE = ((abs((real(:,1)-predicted(:,1)).^2)));

end

[Walter Roberson]

I suspect you want to either sum() or mean() those squared differences, not abs() them.

[Selim]

i also used sum and mean, but i faced the same problem. i want to calculate fitness iteratively ( i mean for each 500 values). i have to change my code. perhaps i might use assignin or evalin functions, but i dont know how to use.

Answer 1

[Alan Weiss]

I have trouble understanding what you are doing. What are your control variables? I mean, which variables should the genetic algorithm change in order to try to minimize something, perhaps a sum of squares?

Whatever that vector of variables is, let's call it x, a row vector.

Your fitness function should accept that row vector and return a real number. Probably, it should look something like this:

 function y = optimfunc(x)
 [real,predicted] = mysimulationresults(x);
 y = norm(real - predicted);

Obviously, I do not exactly know how you are going to write the function mysimulationresults(x), but I assume that you do.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation