On the effect of the steady-state selection scheme in multi-objective genetic algorithms

Abstract

Genetic Algorithms (GAs) have been widely used in single-objective as well as in multi-objective optimization for solving complex optimization problems. Two different models of GAs can be considered according to their selection scheme: generational and steady-state. Although most of the state-of-the-art multi-objective GAs (MOGAs) use a generational scheme, in the last few years many proposals using a steady-state scheme have been developed. However, the influence of using those selection strategies in MOGAs has not been studied in detail. In this paper we deal with this gap. We have implemented steady-state versions of the NSGA-II and SPEA2 algorithms, and we have compared them to the generational ones according to three criteria: the quality of the resulting approximation sets to the Pareto front, the convergence speed of the algorithm, and the computing time. The results show that multi-objective GAs can profit from the steady-state model in many scenarios. © Springer-Verlag 2009.