Embedded self-adaptation to escape from local optima

Abstract

Self-adaptation in genetic algorithms has been suggested as a strategy to enhance evolutionary algorithms for optimization tasks. We consider continuous self-adaptation schemes called strategies that are governed by evolutionary rules, and suggest to incorporate multiple strategies to improve the performance of genetic algorithms. We show that employing multiple strategies, and letting evolutionary pressure steer adaptation, can overcome the problem of premature convergence. To demonstrate the power of our method we apply it to optimization problems of uncapacitated facility location. The method outperforms both methods with a single strategy and previously reported methods on several benchmarks. In these runs, algorithms that incorporate multiple strategies avoid getting stuck in local optimum, and converge to better solutions. Copyright 2009 ACM.