Indicator selection and density estimation deletion-based many-objective evolutionary algorithm

Although a variety of many-objective evolutionary algorithms have been proposed, the difficulty in balancing the convergence and diversity of the population still remains. To address this issue, we proposes an indicator selection and density estimation deletion-based many-objective evolautionary algorithm (MaOEA/IS-DED). In this algorithm, the selection strategy based on the begin{document}$I_{varepsilon+}(x, y)$end{document} indicator and the deletion mechanism based on shifted-based density estimation cooperate for deleting these individuals with poor convergence and diversity one by one, so that the proposed algorithm can make the population converge to the true Pareto front from search directions with good diversity and further achieve balance the converge and diversity. More specifically, the former is designed to find a pair of individuals with the minimum begin{document}$I_{varepsilon+}(x, y)$end{document} indicator values, which denotes these selected individuals have the most similar search directions in space. The latter utilizes own characteristic, taking the convergence and diversity of the population into account, to compare these selected individuals and delete the worse one. Experimental results demonstrate that the MaOEA/IS-DED can gain the highly competitive performance when solving many-objective optimization problems. Copyright ©2023 Control and Decision.