An experimental study for multi-objective optimization by particle swarm with graph based archive

Particle Swarm Optimization is a stochastic multi point search algorithm mimicking social behaviors of animals, such as bird flock. Recently, many researchers pay attentions to Particle Swarm Optimization applying to multi-objective problems. In multi-objective optimization problems, it is desired that solutions cover Pareto-optimal front widely and uniformly. Generally multi-objective particle swarm optimization employs archiving method to store non-dominated solutions which are found in searching and the guide is selected from the archived solutions. In this paper, we consider a topology-based archive updating and guide selection in multi-objective particle swarm optimization in order to keep balance between exploration and exploitation. In the proposed method, each particle has an archive (sub-archive) and the sub-archive is updated by itself and its neighborhood particles. Since it takes some iterations that members in the sub-archive of the particle affect the behaviors of all particle, this method prevents early convergence and the diversity of solutions are mainlined. The performances of the proposed methods with regular graph topology are evaluated by using well known benchmark problems for the evolutionary multi-objective optimization algorithms.