A dynamic archive based niching particle swarm optimizer using a small population size

Many niching techniques have been proposed to solve multimodal optimization problems in the evolutionary computing community. However, these niching methods often depend on large population sizes to locate many more optima. This paper presents a particle swarm optimizer (PSO) niching algorithm only using a dynamic archive, without relying on a large population size to locate numerous optima. To do this, we record found optima in the dynamic archive, and allow particles in converged sub-swarms to be re-randomized to explore undiscovered parts of the search space during a run. This algorithm is compared with lbest PSOs with a ring topology (LPRT). Empirical results indicate that the proposed niching algorithm outperforms LPRT on several benchmark multimodal functions with large numbers of optima, when using a small population size.