A Novel Diversity Preservation Strategy Based on Ranking Integration for Solving Some Specific Multi-Objective Problems
Long Yu, Pan Wang, Zhu Haoshen · 2010
In recent decades, multi-objective evolutionary algorithms (MOEAs) are developed as powerful tools to solve multi-objective optimization problems. While the diversity of Pareto front (PF) plays an important role in the performance evaluation of MOEAs, various diversity preservation strategies (DPS) have been developed. In this paper, a novel approach that inspired from the crowding distance technique is proposed to maintain the diversity of solutions in multi-objective problems (MOPs) with quite different spans of value range. In order to improve its performance, this approach is applied in a well-know MOEA NSGA II by replacing its original DPS. According to 3 test MOPs, the modified NSGA II shows a better diversity and distribution in the PF compared with the original version. Furthermore, the influence of the spans of value range on the performance of original DPS in NSGA II is discussed and the robustness of the new DPS is illustrated.