Dynamically tuning the population size in particle swarm optimization

Abstract

In this paper, we investigate the benefits of dynamically varying the population size in the Particle Swarm Optimization (PSO) model. For this purpose, two well-known population resizing techniques, originally developed for Genetic Algorithms (GAs), were adapted to the PSO context, giving birth to the APPSO and PRoFIPSO variants. Contrary to some previous work that has indicated that the PSO model is not sensitive to the population dimension, the simulation results we have obtained over some benchmark numerical optimization problems suggest that the dynamic variation of the number of particles may be instrumental for bringing about performance improvements in long-term runs, mainly when considering high-dimensional problem instances. In general, the novel PSO variants have compared more favorably to their GA counterparts in targeting the optimal solutions. However, regarding PRoFIPSO specifically, the price to be paid in terms of resources used to reach the optimum point is as a rule very high.