i am getting different results for different times of running the genetic algorithm

hi i'm working on a project in which i have to minimize the net annual cost( objective function). the variable range is in 500-5000. i am getting different results when i run it several times and also the results are weird. can anyone help me i have attached my objective function file and can anyone give me suggestions in choosing the parameters like mutation rate etc. thank you in advance

[Jan]

There is no attachment here.

[trilochan]

https://docs.google.com/document/d/1xHRkwEXGynsP6ElWm7TumkwVyVk-p5uCZs6uFTrXFaU/edit?usp=sharing

this is my function. sorry for not uploading my file for the first time

[kalim ullah]

I am facing the same problem, I am using Multi objective particle swarm optimization for 2 objective functions.

Answer 1

[Alan Weiss]

I didn't see any attachments, so I cannot comment directly on your problem.

Are you sure you want to use the genetic algorithm to solve your problem? If your problem is smooth, Optimization Toolbox solvers such as fmincon are usually faster and more reliable. And if your problem is not smooth, then patternsearch is usually faster and easier to tune than ga.

But if you insist on using ga, I suggest you try the following:

• Give a good initial range
• Use a large population, 200 or even more if you have a lot of variables
• Set a large StallGenLimit to try to keep the solver from stopping prematurely

Basically, ga is a random search algorithm, so if you want better answers, you have to let it search a bigger space and take more tries.

I suggest again you use a different algorithm, either fmincon or patternsearch. You can start either of those solvers at random points within your bounds by using

 x0 = lb + rand(size(lb)).*(ub - lb);

Good luck,

Alan Weiss

MATLAB mathematical toolbox documentation

[trilochan]

https://docs.google.com/document/d/1xHRkwEXGynsP6ElWm7TumkwVyVk-p5uCZs6uFTrXFaU/edit?usp=sharing

this is my function. sorry for not uploading my file for the first time

[Sushrut Dandawate]

Hello, I am using pattern search to solve a non-smooth combined control and design problem with 22 decision variables. Why do I get different results each time I run a pattern search problem? Is there a way to make the results deterministic? Or pattern search always gives different results unless the problem is convex?

[Alan Weiss]

You would probably get more visibility for your question by creating a new question rather than adding to an existing question.

Pattern search is, by default, deterministic, meaning that unless you have changed some default options the algorthm does not give different results each time. To get reproducible results, do not specify options that can change results (such as MADS polling, parallel processing, or random poll order). If your results continue to vary, the reason lies elsewhere.

Alan Weiss

MATLAB mathematical toolbox documentation

[kalim ullah]

Please suggest algorithm for 3 objectives optimization problem. Is it possible to use optimization toolbox for 3 objective functions?

[Alan Weiss]

You would probably get more visibility for your question by creating a new question rather than adding an irrelevant answer to an existing question.

In Global Optimization Toolbox there are gamultiobj and paretosearch. In plain Optimization Toolbox there is fgoalattain. See Multiobjective Optimization and Generate and Plot Pareto Front.

Alan Weiss

MATLAB mathematical toolbox documentation