A Preference-guided Multiobjective Evolutionary Algorithm based on Decomposition

Abstract

Multiobjective evolutionary algorithms based on decomposition (MOEA/Ds) represent a class of widely employed problem solvers for multicriteria optimization problems. In this work we investigate the adaptation of these methods for incorporating preference information prior to the optimization, so that the search process can be biased towards a Pareto-optimal region that better satisfies the aspirations of a decision-making entity. The incorporation of the Preference-based Adaptive Region-of-interest (PAR) framework into the MOEA/D requires only the modification of the reference points used within the scalarization function, which in principle allows a straightforward use in more sophisticated versions of the base algorithm. Experimental results using the UF benchmark set suggest gains in diversity within the region of interest, without significant losses in convergence.