GA results obtains "Function Tolerance but constraints are not satisfied."

I tried to solve fitness function and nonlinear constraints. variables are almost 200 (150 variables are integer value)

GA optoolbox shows me this message.

 Objective function value: 1612176.7897540922
Optimization terminated: average change in the penalty fitness value less than options.FunctionTolerance
but constraints are not satisfied.

However, if i set 250 generations, i got this message.

Optimization running.
Objective function value: 1592795.7559062957
Optimization terminated: maximum number of generations exceeded.

I do not fully understand why ga obtains wrong answer during long iterations.

Can you help me which one i should follow?

Answer 1

[Walter Roberson]

Your population is getting stuck in a local minima, in which all of the function values for the population are so close together that ga cannot figure out which direction to go to get out of the minima. But at the same time, the local minima it is stuck in is one where the constraints are not being met. It could be the case that none of the generated values ever meet the constraints and it spends its time hunting for a place the constraints are met. At the end of 250 generations it has not yet given up on the possibility of finding a place where the constraints are met, but with the larger number of iterations it decides "No, I tried everything reasonable under the options I was given, and I am sure now that I cannot find anywhere near here that meets the constraints."

[Matlab Noob]

Thank you for your answer.

So, there is no way to obtain local minimum satisfied with constraints...? or can you suggest any way to solve this problem.?

I tried to change some parameter related with constraints. Ga shows me same message.

only way to evade 'constraints are not satisfied.' is to reduce decision variable from 150 to 10 :( (I've never checked between 10 to 150, yet.) However, it is not realistic...

[Walter Roberson]

Without knowing more about your function and your constraints, it is difficult to say. Sometimes the randomly generated populations just don't happen to satisfy the constraints. Sometimes the constraints are not feasible.

[Alan Weiss]

You might want to try to find some initial feasible points and give them to ga as part of an initial population. The more feasible initial points you find, the better ga will do.

Alan Weiss

MATLAB mathematical toolbox documentation