Comparative Study of Crossovers for Decision Space Diversity of Non-Dominated Solutions
Motoki Sato, Akira Oyama · 2021 IEEE Symposium Series on Computational Intelligence (SSCI) · 2021
Capturing diversity of non-dominated and dominated solutions in decision space is important for realworld multiobjective optimization to provide a decision maker many options. This paper studies how different crossover operators affect diversity of non-dominated and dominated solutions in decision space obtained by multiobjective evolutionary algorithms (MOEA). We compare the solutions obtained by NSGA-II with simulated binary crossover (SBX), unimodal normally distributed crossover (UNDX), reproduction process of differential evolution (DE), or blend crossover (BLX-α) for speed reducer design (SRD) problem and Mazda problem. The result shows that selection of crossover operator significantly affects diversity of non-dominated and dominated solutions in the decision space obtained by MOEA.