Many-Objective Decomposition-based Evolutionary Algorithm with Adaptive Weights

Abstract

Many-objective evolutionary algorithms are difficult to maintain good individual convergence and population diversity. As the number of objectives increases, there are more and more non-dominated solutions, and some existing diversity metrics are no longer useful. In this paper, a many-objective decomposition-based evolutionary algorithm with adaptive weights (MaOEA-AWM) is proposed. The convergence and diversity of individuals in high-dimensional target space are balanced in the proposed MaOEA-AWM through the use of a scaling method known as angle penalty distance. Additionally, a weight vector adaptation approach is proposed to adjust the weight vector distribution. Experiments show that MaOEA-AWM has strong competitiveness in many-objective optimization problems compared with seven advanced algorithms.