Abstract
Particle Swarm Optimization (PSO) is a type of evolutionary computation algorithm, and PSO searches using information from each particle and information from the collective group. In general, the convergence of PSO is faster than other evolutionary computation methods because PSO quickly converges on a local optima in the early stages of the search cycle. As a trade-off, as the particles converge on a local minimum, the particles lose velocity and thus the particle group loses diversity in the solution candidates. In other evolutionary computation algorithms, Distributed Genetic Algorithm (DGA) sustains diversity by creating several search groups. In DGA, each group can search in parallel for improved search performance. To improve the diversity in PSO, we proposed a PSO based search method that divides the particles into several search groups similar to DGA. In the proposed method, the particles are grouped according to the calculated fitness, in which high fitness particles form a local search group, and low fitness particles form a wide area global search group. In the proposed PSO, it becomes possible for the groups to search in parallel, to improve the search speed as well as to keep the diversity of the solution set during the search. In this research, we implemented the proposed PSO method as a parallel computation application, and investigated the effectiveness of this method.
