A many-objective evolutionary algorithm with epsilon-indicator direction vector

Abstract The major difficulty in multi-objective optimization evolutionary algorithms (MOEAs) is how to find an appropriate solution which is able to converge towards the true Pareto Front with high diversity. In order to strengthen the selection pressure of the algorithms, indicator-based algorithms have been proposed to handle many-objective optimization problems (MaOPs), among which binary addition quality indicator I e + is superior to other indicators in terms of low computational complexity. However it often has edge effects which degrade the performance of MOEA. In this work, we devise a new MOEA approach, which is able to combine binary addition quality indicator I e + with direction vector (EDV), to address MaOPs. At the same time, an efficient resource allocation strategy is developed to improve the diversity distribution of solutions. Simulation results are presented to show that EDV outperforms state-of-the-art approaches in all problems considered in this paper, and takes a great advantage in solving the black box problem.