Improved Particle Swarm Optimization in Constrained Numerical Search Spaces

Abstract

This chapter presents a study about the behavior of Particle Swarm Optimization (PSO) in constrained search spaces. A comparison of four well-known PSO variants used to solve a set of test problems is presented. Based on the information obtained, the most competitive PSO variant is detected. From this preliminary analysis, the performance of this variant is improved with two simple modifications related with the dynamic control of some parameters and a variation in the constraint-handling technique. These changes keep the simplicity of PSO i.e. no extra parameters, mechanisms controlled by the user or combination of PSO variants are added. This Improved PSO (IPSO) is extensively compared against the original PSO variants, based on the quality and consistency of the final results and also on two performance measures and convergence graphs to analyze their on-line behavior. Finally, IPSO is compared against some state-of-the-art PSO-based approaches for constrained optimization. Statistical tests are used in the experiments in order to add support to the findings and conclusions established.