A Parallel Implementation of Multiobjective Particle Swarm Optimization Algorithm Based on Decomposition

Multiobjective particle swarm optimization based on decomposition (MOPSO/D) is an effective algorithm for multiobjective optimization problems (MOPs). This paper proposes a parallel version of MOPSO/D algorithm using both message passing interface (MPI) and OpenMP, which is abbreviated as MO-MOPSO/D. It adopts an island model and divides the whole population into several subspecies. Based on the hybrid of distributed and shared-memory programming models, the proposed algorithm can fully use the processing power of today's multicore processors and even a cluster. The experimental results show that MO-MOPSO/D can achieve speedups of 2x on a personal computer equipped with a dual-core four-thread CPU. In terms of the quality of solutions, it can perform similarly to the serial MOPSO/D but greatly outperform NSGA-II. An additional experiment is done on a cluster, and the results show the speedup is not obvious for small-scale MOPs and it is more suitable for solving highly complex problems.