Tips for specifying Population Size, No. of Generations, and other Options for GA with large nvars

Hello, I have a large number of GA nvars (few hundreds and may reach one thousand), may I get some tips for Options settings in order to improve the results ...

should I use the default values ?

is increasing Population Size (1000) and No. of generations while decreasing TolFun and TolCon always enhance the results? or there are limits? right now, I am using PS ~ 1000 No. of Generations ~ 300 TolFun ~ 1e-15 TolCon ~ 1e-15

do everything depend on trials and extensive runs ?

thank you in advance

Answer 1

[Alan Weiss]

Depending on your problem, GA can have a lot of trouble with such a high-dimensional problem. Is GA really the right optimizer? If you have a smooth problem, try fmincon first. If your problem is not smooth, try patternsearch.

But if you insist on using GA, then I agree that you need a much larger population, perhaps 2000. Setting such a small value of TolFun will probably cause GA to take many iterations, a good thing. The small value of TolCon indicates to me that you have nonlinear constraints. If that is true, I strongly suggest that you use fmincon or patternsearch instead, becuse the GA nonlinear constraint solver can be especially slow.

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[Mamdouh]

A fast, clear, sincere, and informative answer... as I used from you, thank you Alan, :)