An Overview of Pair-Potential Functions for Multi-objective Optimization

摘要

Recently, an increasing number of state-of-the-art Multi-objective Evolutionary Algorithms (MOEAs) have incorporated the so-called pair-potential functions (commonly used to discretize a manifold) to improve the diversity within their population. A remarkable example is the Riesz s-energy function that has been recently used to improve the diversity of solutions either as part of a selection mechanism as well as to generate reference sets. In this paper, we perform an extensive empirical study with respect to the usage of the Riesz s-energy function and other 6 pair-potential functions adopted as a backbone of a selection mechanism used to update an external archive which is integrated into MOEA/D. Our results show that these pair-potential-based archives are able to store solutions with high diversity discarded by the MOEA/D's main population. Our experimental results indicate that the utilization of the pair-potential-based archives helps to circumvent the known MOEA/D's performance dependence on the Pareto front shapes without meddling with the original definition of the algorithm.