A Knee Point Based NSGA-II Multi-objective Evolutionary Algorithm

Abstract

Many evolutionary algorithms (EAs) can’t select the solution which can accelerate the convergence to the Pareto front and maintain the diversity from a group of non-dominant solutions in the late stage of searching. In this article, the method of finding knee point is embedded in the process of searching, which not only increases selection pressure solutions in later searches but also accelerates diversity and convergence. Besides, niche strategy and special crowding distances are used to solve multimodal features in test problems, so as to provide decision-makers with multiple alternative solutions as much as possible. Finally, the performance indicators of knee point are compared with the existing algorithms on 14 test functions. The results show that the final solution set of the proposed algorithm has advantages in coverage area of the reference knee regions and convergence speed.