A Many-objective Evolutionary Algorithm Based on Angle Penalized Distance
Xiaojun Bi, Chao Wang · 2018
In order to balance between convergence and distribution in Multi-Objective Evolutionary Algorithms (MOEAs), a Many-Objective Evolutionary Algorithm based on Angle Penalized Distance (MaOEA-APD) is proposed. Firstly, considering the importance of convergence and diversity in the different stages of the evolutionary process, an angle penalized distance is constructed to dynamically balance between them. Then, the environmental selection based on removing the worse individual is designed to maintain the distribution and improve the convergence. Finally, the mating selection is designed based on the principle of the environmental selection. Both are complement and coordinated to each other for improving the evolutionary efficiency of the algorithm. Compared with three state-of-the-art many-objective evolutionary algorithms (MaOEAs), the experimental results on WFG test suite show that MaOEA-APD has more advantage than other algorithms in terms of the overall performance.