A coevolutionary genetic search for a layout problem

This chapter is devoted to an application of genetic algorithms and coevolutionary principles to a large optimization problem. Starting point is a mixed integer linear program which models our problem - in this case a facility layout problem. As the number of binary variables increases quadratically with the problem size, currently available solvers fail already for small problem instances. Using an genetic search our algorithm reduces the number of binary variables by setting a considerable part of them. The genetic operators were specially designed to yield a high precentage of feasible variable settings. In order to further speed up the computation of large problems we propose a partition into interdipendent subproblems. Each subproblem ("species") is evolved by a genetic algorithm respecting the constraints ("environment") generated by the others. Numerical experiments verify this coevolutionary approach.