GA Optimization not working because fitness function must be a function handle.

Hi,

I am trying to optimize u for a customized function called fuelconsumption. This function takes as an input u. Note that u is a vector of size 1801x1.

When trying to optimize u using GA, I get the following error:

Error using ga

Fitness function must be a function handle.

Error in ga_optimization (line 26)

[x fval] = ga(prob,1,[],[],[],[],[],[],[],options);

Here is my code for reference:

%% Problem Definition

tic

u = optimvar("u","LowerBound",-21,"UpperBound",25);

f=@(u) fuelconsumption(u);

prob = optimproblem("Objective",fuelconsumption(u));

options= optimoptions(@ga, 'PopulationSize',TimeCycle)

options = optimoptions('ga','UseVectorized',true);

[x fval] = ga(prob,1,[],[],[],[],[],[],[],options);

toc

Any help would be really appreciated!

[Chris]

1. If u is a vector, you overwrite it when you assign it using optimvar. A possible fix:

    u = optimvar("u",1801,1,"LowerBound",-21,"UpperBound",25);

2. You overwrite your options struct the second time you call optimoptions. Name-Value pairs can be entered all at once:

    options = optimoptions('ga','UseVectorized',true,'PopulationSize',TimeCycle)

or you can use dot indexing:

options = optimoptions('ga');
options.UseVectorized = true;
options.PopulationSize = TimeCycle; % Assuming TimeCycle is a variable you've defined

(or some combination of the two methods)

This should help a little bit, but it's still not a full answer. Here's some more useful help:

https://www.mathworks.com/help/optim/ug/problem-based-workflow.html

https://www.mathworks.com/help/matlab/matlab_prog/creating-a-function-handle.html

[Josee Rizkallah]

For some reason, the ga function only returns one value for x and fval when it should return a vector of 1801x1, but with your help, this is what I got:

prob = @(u) fuelconsumption(u);
options = optimoptions('ga','UseVectorized',true,'PopulationSize',TimeCycle,'MaxStallGenerations',3,'FunctionTolerance',1e-04,'MaxGenerations',75)
[x fval]= ga(prob,1,[],[],[],[],-21,25,[],options);

Answer 1

[Matt J]

There is no real reason to use optimproblem for a simple 1D bounded problem. Just do,

      [x fval] = ga(@fuelconsumption,1,[],[],[],[],-21,25,[],options);

or, for that matter

           [x fval] = fminbnd(@fuelconsumption,-21,25);

[Josee Rizkallah]

When I try to use

[x fval] = fminbnd(@fuelconsumption,-21,25);

I get the following error:

Error using fminbnd

User supplied objective function must return a scalar value.

Which makes sense since my function returns a vector. How can I fix this?

[Matt J]

Your function should not be returning a vector. It should be returning the scalar quantity that you are trying to minimize.