Swarm intelligence guided by multi-objective mathematical programming techniques

Abstract

Since the early days of multi-objective particle swarm optimizers (MOPSOs), researchers have looked for appropriate mechanisms to define the set of leaders (or global best set) from the swarm. At the beginning, leaders were randomly selected from the set of nondominated solutions currently available. However, over the years, researchers realized that random selection schemes were not the best choice, and additional information was incorporated in the leader selection mechanism (namely, information related to density estimation). Here, we study the use of mathematical programming techniques for defining the leader selection mechanism of a MOPSO. The proposed approach decomposes a multi-objective optimization problem (MOP) into several single objective optimization problems by using traditional multi-objective mathematical programming techniques. Our preliminary results indicate that our proposed approach is a viable choice for solving MOPs, since it is able to outperform a state-of-the-art multi-objective evolutionary algorithm (MOEA).