A genetic learning strategy in constrained search spaces

Abstract

The performance of an adaptive learning algorithm based on evolution (the genetic algorithm) is investigated in constrained boolean search spaces where some solutions may be infeasible. This paper describes a randomized validation procedure to limit the genetic search to feasible regions of the search space. Analysis of the effect of the validation procedure on genetic optimization is presented. The performance of the modified genetic search on the set covering problem is used to illustrate the usefulness of the analysis in selecting the algorithm's parameters.<>