Genetic algorithms with noisy fitness (Q1922199)

This paper is devoted to the convergence analysis of genetic algorithms with noisy fitness information. Since the identification of the best individual in each generation is needed, one has to decide which individual has maximal fitness. In the proposed scheme, sign tests are used to compare sample fitness values. The sequence of best individuals is considered, and the algorithm is regarded as convergent if this sequence converges with probability one to a point with optimal average fitness. The results of numerical experiments are given and commented.