Analysis of Real-World Constrained Multi-Objective Problems and Performance Comparison of Multi-Objective Algorithms

Abstract

Real-world multi-objective optimization problems usually have multiple constraints. To solve constrained multi-objective optimization problems (CMOPs), researchers have proposed various evolutionary multi-objective optimization (EMO) algorithms with constraint handling techniques. Those EMO algorithms explicitly or implicitly assume the existence of a large infeasible region in the objective space between initial solutions and the Pareto front. As a result, they use some special mechanisms to traverse such an infeasible region (e.g., push-and-pull search). However, it is not clear whether real-world CMOPs have similar characteristics. It is also unclear whether state-of-the-art EMO algorithms that proposed for artificial CMOPs work well on real-world CMOPs. In this paper, we examine the characteristics of some real-world CMOPs. We find that the examined real-world CMOPs have no large infeasible region near the Pareto front. We also compare the performance of some constrained EMO algorithms on artificial CMOPs and real-world CMOPs. Our experimental results show that performance comparison results on real-world CMOPs are clearly different from those on artificial CMOPs. It is also shown that some recently-proposed constrained EMO algorithms are outperformed by NSGA-II with the basic constraint domination principle when they are compared on real-world CMOPs.