Studying the effect of techniques to generate reference vectors in many-objective optimization

A large number of Multi-Objective Evolutionary Algorithms employ reference directions in order to establish relative preferences for each objective function. Uniform Design (UD), Simplex Lattice Design (SLD) and their variants are techniques commonly used to generate a set of uniformly distributed reference directions with the aim of capturing the whole Pareto optimal front. In this paper, we present a comparative study of UD and SLD methods when solving Many-objective Optimization problems and we design a new strategy that combines SLD with multiple layers and UD techniques. Our preliminary results indicate that our proposed approach is able to outperform state-of-the-art methods in many-objective optimization problems.