On Sharing Information Between Sub-populations in MOEA/S

摘要

This work investigates the effect of information exchange in decomposition methods that work with multi-membered populations as sub-problems. As an algorithm framework, we use the Multi-objective Evolutionary Algorithm based on Sub-populations (MOEA/S). This algorithm uses parallel sub-populations that can exchange information via migration and/or recombination. For this work, each sub-population is constructed by a few weighted utility functions, grouped by distance between their weighting vectors. The question investigated in this paper is: How is the distance between sub-populations and the mechanism of information exchange influencing the performance of MOEA/S? The study considers two ways of transferring information: (1) migration of individuals, (2) recombination using parents from two different sub-populations. A matrix describing the linkage patterns between sub-populations governs migration and recombination mechanisms. This work conducts a systematic study using the multi-objective knapsack problem (MOKP) and multi-objective traveling salesperson (MOTSP) for two and three objectives test problems. The results motivated a restriction policy for sharing information. We compare an algorithm using this policy with other state-of-the-art MOEAs, including NSGA Ⅲ, MOEA/D, and the previous version of MOEA/S.