Abstract
In this paper, we present a new multimodal optimization algorithm, role based particle swarm optimization (RPSO), for finding and maintaining multiple optima in objective function landscape. Instead of generating all trial vectors randomly, the swarm population is divided into three kinds of roles, each part of swarms generating offsprings with different strategy. A species conservation procedure is employed during the optimization process to save the newly found peaks. Numerical experiments are performed to compare the proposed method with canonical species conservation GA on a series of benchmark functions. Based on the results, we conclude that the proposed technique is comparatively effective on selected benchmark functions in terms of locating and maintaining the multiple optima.