Looking Inside Particle Swarm Optimization in Constrained Search Spaces

In this paper, the behavior of different Particle Swarm Optimization (PSO) variants is analyzed when solving a set of well-known numerical constrained optimization problems. After identifying the most competitive one, some improvements are proposed to this variant regarding the parameter control and the constraint-handling mechanism. Furthermore, the on-line behavior of the improved PSO and some of the most competitive original variants are studied. Two performance measures are used to analyze the capabilities of each PSO to generate feasible solutions and to improve feasible solutions previously found i.e. how able is to move inside the feasible region of the search space. Finally, the performance of this improved PSO is compared against state-of-the-art PSO-based algorithms. Some conclusions regarding the behavior of PSO in constrained search spaces and the improved results presented by the modified PSO are given and the future work is established.