Some aspects of parallel genetic algorithms with population re-initialization

In case of highly non-smooth search/optimization problems it is not easy to avoid the premature convergence of the genetic algorithm. For that reason it is important to provide for a high measure of population diversity of the GA. In such a case, an effective means is the population re-initialization. In this paper the influence of population re-initialization on the parallel genetic algorithm (PGA) performance is experimentally analyzed. In various PGA architectures three types of re-initialization are described. Next the following factors are studied: re-initialization period and the number of re-initialized nodes. The results are demonstrated on the minimization of real number test functions.