High-dimensional multi-objective evolutionary algorithm based on adaptive penalty parameters and improved association methods

Abstract

The balance between convergence and diversity is a crucial and challenging aspect of evolutionary multi-objective optimization. In order to balance them, this paper proposes a high-dimensional multi-objective evolutionary algorithm based on adaptive penalty parameters and an improved association method (MOEAAP). An adaptive penalty parameter is first proposed and utilized in non-dominated sorting based on reference points, allowing for the dynamic assignment of Pareto ranks to the population according to changes in iteration counts and objective dimensions. Additionally, an improved association method is employed to categorize candidate solutions into associated and non-associated solutions, while associated solutions are given preferential consideration. To verify the feasibility of the proposed algorithm, it is compared with several mainstream algorithms through simulation experiments, and is applied to multiple practical applications. Experimental results show the algorithm effectively balances convergence and diversity in high-dimensional, complex multi-objective optimization.