Abstract: Multi-objective particle swarm optimizers are often trapped in local optima, converge slowly and cost more function evaluations. Therefore, a diversity-guided two-stage MOPSO (DTSPSO) was proposed. DTSPSO dynamically selects different mutation operators according to current population diversity and divides into two stages according to its ways of selecting leaders. In addition, Pareto dominance ranking and crowding distance were used to fix the size of the external archive. Experiments were carried out on several classical benchmark functions for multi-objective optimization problems and the results show that DTSPSO is effective in solving various multi-objective optimization problems.

Key words: leader particle, diversity, particle swarm optimizer, multi-objective optimization