Cooperative based hyper-heuristic for many-objective optimization
Gian Fritsche, Aurora Pozo · Proceedings of the Genetic and Evolutionary Computation Conference · 2019
Multi-Objective Evolutionary Algorithms (MOEAs) have shown to be effective, addressing Multi-Objective Problems (MOPs) suitably. Nowadays, there is a variety of MOEAs proposed in the literature. However, it is a challenge to select the best MOEA within a specific domain problem, since the MOEAs present different performance depending on the problem characteristics. Moreover, it is difficult to configure the parameter values or to select the operators properly. Considering this, we propose a new hyper-heuristic approach based on the cooperation of MOEAs (HH-CO). The main characteristic of HH-CO is that every MOEA has a population and exchanges information between them. This paper presents experimental results of HH-CO for the benchmark from CEC'18 competition on many-objective optimization. We present two comparisons: the first one, where HH-CO is compared to each MOEA that composes the pool and to a state-of-the-art hyper-heuristic, and the second one, that compares HH-CO to state-of-the-art algorithms winners of CEC'18 competition. The results were evaluated using a set of quality indicators and were statistically analyzed. The conclusion is that HH-CO is a suitable approach mainly for challenging problems, with complicated fitness landscape, including multi-modality, bias and a high number of objectives.