A Novel Diversity Guided Particle Swarm Multi-objective Optimization Algorithm

Particle swarm optimization which is inspired by social behavior of bird flocking or fish schooling was developed to solve single objective problem by Dr. Eberhart and Dr. Kennedy in1995. Afterward the Particle swarm algorithm has been used to solve multi-objective, optimization problem. Multi-objective particle swarm algorithm has become one of popular method to solve multi-objective optimization problem, which has two or more conflicting objectives. Much research work has been focus on improving this method’s performance, such as "An Effective Use of Crowding Distance in Multi-objective Particle Swarm Optimization (MOPSO-CD)". However, to improve simultaneously the two performances as convergence to the Pareto-optimal and maintenance of diversity in the solutions of Pareto sets is a challenging and significant research issue now.This thesis presents a novel diversity guided particle swarm multi-objective optimization algorithm named MOPSO-AR which increases diversity performance of multi-objective Particle Swarm optimization by using Attraction and Repulsion (AR) mechanism. AR mechanism uses a diversity measure to control the swarm. Being attractive and repulsive will help overcome the problem of premature convergence. AR mechanism together with crowding distance computation and mutation operator maintains the diversity of non-dominated set in external archive. The approach is verified by several test function experiments. Results demonstrate that the proposed approach is highly competitive in distributed of non-dominated solutions but still keeps convergence towards the Pareto front.The experiment result shows that the proposed method improves effectively the convergence performance and diversity of the Pareto optimal solutions. The algorithm can be used in multi-objective decision, parameters optimization, combination optimization and etc. In the next step, the method would be applied for some science research and engineering multi-objective optimization problems, such as grid tasks scheduling, intelligent vehicle decision and so on.