Multiobjective GA optimization using reduced models

In this paper, we propose a novel method for solving multiobjective optimization problems using reduced models. Our method, called objective exchange genetic algorithm for design optimization (OEGADO), is intended for solving real-world application problems. For such problems, the number of objective evaluations performed is a critical factor as a single objective evaluation can be quite expensive. The aim of our research is to reduce the number of objective evaluations needed to find a well-distributed sampling of the Pareto-optimal region by applying reduced models to steady-state multiobjective GAs. OEGADO runs several GAs concurrently with each GA optimizing one objective and forming a reduced model of its objective. At regular intervals, each GA exchanges its reduced model with the others. The GAs use these reduced models to bias their search toward compromise solutions. Empirical results in several engineering and benchmark domains comparing OEGADO with two state-of-the-art multiobjective evolutionary algorithms show that OEGADO outperformed them for difficult problems.