Dynamic Multi-Swarm Fractional-best Particle Swarm Optimization for Dynamic Multi-modal Optimization

Abstract

While many particle swarm optimization (PSO) algorithms have been developed to find multiple optima to multimodal optimization problems, very few PSO algorithms exist to both find and track multiple optima in dynamically changing search landscapes. This paper presents a novel multi-swarm PSO algorithm, where the number of sub-swarms change dynamically over time to more efficiently adapt to problems where the number of optima changes over time. In addition, a repelling mechanism is employed to prevent sub-swarms from converging to the same optimum. Instead of designating one particle as the global best position, the global best position is determined by combining the best components from different particles. The new algorithm, called the dynamic multi-swarm fractional-best PSO algorithm, is compared to the best available dynamic multi-modal PSO algorithms on a large set of dynamic optimization problems with varying dynamics. The results show that the dynamic multi-swarm fractional-best PSO performs the best with reference to offline error, and second best with reference to the average number of optima found. The new algorithm's offline error is also shown to be insensitive to change severity.