Título: RB-MOPSO: A PSO algorithm based on reference points
Autores: Castro Jr, Olacir R.; Britto, André; Pozo, Aurora
Resumo: Many-objective problems refer to problems containing large number of objective functions to be optimized, typically more than three. As most existing algorithms based on Pareto dominance are not efficient in handling this kind of problem, researchers have been working on alternatives to overcome these limitations. Recently, Deb et al. proposed an improved NSGA-II, the principal characteristic of which is the use of reference points to obtain a larger covered portion of the Pareto front. Inspired by the results of the improved NSGA-II, in this paper we propose a new algorithm based on Particle Swarm Optimization grounded on these concepts. The algorithm, called RB-SMPSO, presents as features improved mechanisms for (i) the update of the external archive to consider the reference points, and (ii) the selection of social leaders. In order to validate the proposal, a comparative study is presented considering a state of art MOPSO-based algorithm (Speed-constrained Multi-Objective Particle Swarm Optimization, SMPSO) using two different leader selection methods. All algorithms are evaluated using well-known scalable problems with 2, 3, 5, 10, 15 and 20 objectives, respectively. The results point out that RB-SMPSO presents the better overall performance between the compared approaches.
Palavras-chave: Particle Swarm Optimization; Multi-objective; Many-objective; Optimization
Código DOI: 10.21528/CBIC2013-056
Artigo em pdf: bricsccicbic2013_submission_56.pdf
Arquivo BibTex: bricsccicbic2013_submission_56.bib