Stability-Guided Multi-Guide Particle Swarm Optimization

This paper proposes a multi-guide particle swarm optimization (MGPSO) algorithm which does not require tuning of its control parameters. Control parameter values are randomly sampled to satisfy theoretically derived stability conditions, eliminating the need for computatinally expensive parameter tuning. In addition, the feasibility of utilizing dynamically decreasing tournament sizes in the selection of the archive guide, as well as a ring neighbourhood topology, is investigated. The results show that random control parameter sampling is a viable alternative to static tuning, most notably when applied to higher numbers of objectives. However, the results show no clear benefit or detriment to utilizing dynamic tournament selection sizes and ring neighbourhood topologies.