Hybrid methods for multi-objective evolutionary algorithms

Abstract

Hybrid methods of using evolutionary algorithms with a local search method are often used in the context of single-objective real-world optimization. In this paper, we discuss a couple of hybrid methods for multiobjective real-world optimization. In the posteriori approach, the obtained non-dominated solutions of a multiobjective evolutionary algorithm (MOEA) run are modified using a local search method. In the online approach, a local search method is applied to each solution obtained by genetic operations in a MOEA run. Both these approaches are compared on three engineering shape optimization problems for a fixed number of overall function evaluations. Simulation results suggest important insights about the extent of local search and the extent of an MOEA needed to achieve an overall efficient hybrid approach.