error in fitness function (gaoptimset) ??? despite convex function gaoptimset does not optimise 2010b

I have a single variable objective function. This variable goes into my for loop.

irrespective of what lower and upper bonds I assign; initial range i give and with vectorize option 'on'.

does not optimize. I get following values

x: 100.000000.
fval: 8231923.58.

My objective function does have a convex behavior. this plot was generated by manually setting the forloop upto 10000

And went I set the options = gaoptimset(options,'Vectorized', 'off');

??? Subscripted assignment dimension mismatch.

Error in ==> fcnvectorizer at 14
            y(i,:) = feval(fun,(pop(i,:)));

numberOfVariables = 1; % always 1
PopulationSize_input = 100; % can vary between 20 to 100

%%Bounds on our flowrate
lb =100; %100;
ub = 1000; %0.00050

%%---- parametarising Genetic algorithm-----------%%

options = gaoptimset;

% Modify options setting

%options = gaoptimset(options,'PopInitRange' ,[0 ; 0.00050]);

options = gaoptimset(options,'PopInitRange' ,[lb ; ub]);

options = gaoptimset(options,'MutationFcn' ,@mutationadaptfeasible);

options = gaoptimset(options,'CrossoverFcn' ,@crossoverarithmetic);

options = gaoptimset(options,'PopulationSize', PopulationSize_input);

options = gaoptimset(options,'Display', 'off');
options = gaoptimset(options,'PlotFcns', {  @gaplotbestf @gaplotbestindiv @gaplotrange @gaplotselection });
options = gaoptimset(options,'Vectorized', 'off');

options = gaoptimset(options,'UseParallel', 'never');

%options = gaoptimset(options,'HybridFcn' ,@fminsearch);

%options = gaoptimset('OutputFcns', @gaoutputgen);

%options = gaoptimset(options,'OutputFcns' ,{ @gaplotbestf @gaplotbestindiv @gaplotrange @gaplotselection });

FitFcnn = @(f) HydraulicCOAX(f);
    [x, fval] = ga(FitFcnn,numberOfVariables,[],[],[],[],lb,ub,[],options);

      fprintf('The flow rate: %6.6f.\n',x);

      fprintf('The Qeng: %6.2f.\n',fval);

[Matt J]

What happens when you use a different optimizer, e.g., fminsearch?

[Swarup]

Hi Matt,

1) I tried using fminsearch as options = gaoptimset(options,'HybridFcn' ,@fminsearch);

??? Undefined function or variable "Q".

Error in ==> HydraulicCOAXfmin at 54 q=size(Q,2);

Error in ==> @(f)HydraulicCOAXfmin(f)

Error in ==> validate>@(x)fitness(x,FitnessFcnArgs{:}) at 136 fitness = @(x) fitness(x,FitnessFcnArgs{:});

and fminsearch

Optimization running. Error running optimization. Subscripted assignment dimension mismatch.

*****************************************

Is it because the for loop in my objective function takes integer values??

and the f (which come from PopulationSize =100) are not integers.

function [Peng] = HydraulicCOAXfmin(f)

for i=1:f

Q(i)= f/10/3600

rho=999.702;

eta=1305.875E-6;

vi(i)=Q(i)/((rii^2)*pi);

...

...

end

q=size(Q,2)

for j=1:q %Rückskalieren auf m^3/h für plot

    Q(j)=Q(j)*3600;

end

end

Answer 1

[Alan Weiss]

I don't know why you think that the solver failed. From your graph, the minimum value of your objective function occurs at the point x = 0, and rises in a convex fashion as x increases. So you give a lower bound of 100 and an upper bound of 1000, and of course you find that the best point in this range is at x = 100.

It looks to me as if the solver is doing its job.

However, I am not at all sure why you are using ga on this type of function. For a smooth function, use an Optimization Toolbox solver such as fmincon.

Alan Weiss

MATLAB mathematical toolbox documentation

[Matt J]

For a smooth function, use an Optimization Toolbox solver such as fmincon.

Since it is a one-variable problem, you could even just use fminsearch or fminbnd by itself (not within ga).

[Swarup]

Hi Matt and Alan,

Thanks, I agree, but a small argument. My single variable problem a convex shape (local minima), why dosent, GA find this mimina?

Now I am tangled a bit in my own problem, I would have to assign constraints for fminsearch and fminbnd..

[c ceq] =fltrcn(f)

c = ??

ceq = []

end

Thanks guys. I have been scratching my head since monday!!

Answer 2

[Swarup]

Thanks for your suggestions Matt and Alan. Regards, Swarup