Failure of genetic algorithm in finding feasible point to a nonlinear binary constrained problem

Dear all,

I am currently having trouble finding feasible points using ga for my nonlinear binary constrained problem.

I have a huge problem with 6210 binary variables, 206 linear equality and 78 linear inequality constraints. I transformed all equality constraints into inequality so I end up having 490 inequality constraints now. I have left out the parameter Intcon and instead supplied my own initial population creation, crossover, mutation operators, which ensure that they produce binary variables, as I round all variables at the end of each operator process.

Since I could generate solutions for the feasibility problem that is without the nonlinear objective function of the same problem, obviously solutions exist (I do that taking advantage of the CPLEX solver in Matlab). However, even if I run the ga with population size 50000, it minimizes, minimizes and reports in the end that the constraints are not respected within the given tolerance.

Clearly, my solution space is very restricted through the many constraints. I am aware that the initial population that I create through random numbers does not lie in the feasible region at all. Is this a problem? Is it possible for the ga in Matlab to find feasible solutions, even if the initial population is not feasible?

I appreciate your short comment.