Fast Solutions Enhancing using a Copula-based EDA and SVM for many-objective problems
Résumé
In this paper we propose a new Copula-based Estimation of Distribution Algorithm, to solveMany-objective optimization problem and to get new optimal solutions in very court time. Our algorithmuses the proprieties of Copula and exploits their statistical properties to make new solutions using thefounded optimal solutions through the estimation of their distribution. The first step of the proposedCopula-based Estimation of Distribution Algorithm (CEDA-SVM) is taking initial solutions offeredby any MOEA (Multi Objective Evolutionary Algorithm), and then creates Copulas to estimate theirdistribution, and we use Support Vector Machine (SVM) to learn the Pareto solutions model; thoseCopulas will be used to generate new solutions and SVM to avoid the expensive function evaluations.The idea of using the estimated distribution of the optimal solutions helps CEDA-SVM to avoid runningthe optimizer ( MOEA ) every time we need new alternatives solutions when the found ones are notsatisfactory. We tested CEDA-SVM on a set of many-objective benchmark problems traditionally usedby the community, namely DTLZ (1, 2, 3, and 4) with different dimensions (3, 5, 8, 10, and 15). Weused CEDA along with MOEA/D-Schy and MOEA/D-Bpi as two examples of MOEA thus resultingin two variants CEDA-MOAE/D-Scy and CEDA-MOEA/D-Bi and compare them with MOEA/D-Schyand MOEA/D-Bpi. The results of our experiments show that, with both variants of CEDA-SVM, newsolutions can be obtained in a very small time compared with the other algorithms