An improved multi-objective particle swarm optimization algorithm

This paper proposes a dynamic sub-swarms multi-objective particle swarm optimization algorithm (DSMOPSO). Based on solution distribution of multi-objective optimization problems, it separates particles into multi subswarms, each of which adopts an improved clustering archiving technique, and operates PSO in a comparably independent way. Clustering eventually enhances the distribution quality of solutions. The selection of the closest particle to the gbest from archiving set and the developed pbest select mechanism increase the choice pressure. In the meantime, the dynamic set particle inertial weight, namely, particle inertial weight being relevant to the number of dominating particles, effectively keeps the balance between the global search in the preliminary stage and the local search in the later stage. Experiments show that this strategy yields good convergence and strong capacity to conserve the distribution of solutions, specially for the problems with non-continuous Pareto-optimal front.