Shake them all! Rethinking selection and replacement in MOEA/D

Abstract

In this paper, we build upon the previous efforts to enhance the search ability of Moea/d (a multi-objective decomposition-based algorithm), by investigating the idea of evolving the whole population simultaneously. We thereby propose new alternative selection and replacement strategies that can be combined in different ways within a generic and problem-independent framework. To assess the performance of our strategies, we conduct a comprehensive experimental study on bi-objective combinatorial optimization problems. More precisely, we consider ρMNK-landscapes and knapsack problems as a benchmark, and experiment a wide range of parameter configurations for Moea/d and its variants. Our analysis reveals the effectiveness of our strategies and their robustness to parameter settings. In particular, substantial improvements are obtained compared to the conventional Moea/d.