An inner-point modification of PSO for constrained optimization
Purpose – The purpose of this paper is to propose a modification of the original PSO algorithm in order to avoid the evaluation of the objective function outside the feasible set, improving the parallel performances of the algorithm in the view of its application on parallel architectures. Design/methodology/approach – Classical PSO iteration is repeated for each particle until a feasible point is found: the global search is performed by a set of independent sub-iteration, at the particle level, and the evaluation of the objective function is performed only once the full swarm is feasible. After that, the main attractors are updated and a new sub-iteration is initiated. Findings – While the main qualities of PSO are preserved, a great advantage in terms of identification of the feasible region and detection of the best feasible solution is obtained. Furthermore, the parallel structure of the algorithm is preserved, and the load balance improved. The results of the application to real-life optimization problems, where constraint satisfaction sometime represents a problem itself, gives the measure of this advantage: an improvement of about 10 percent of the optimal solution is obtained by using the modified version of the algorithm, with a more precise identification of the optimal design variables. Originality/value – Differently from the standard approach, utilizing a penalty function in order to discharge unfeasible points, here only feasible points are produced, improving the exploration of the feasible region and preserving the parallel structure of the algorithm.