Many-Objective Evolutionary Optimization using Density Peaks Scoring Selection Strategy

Siamak Ghodsi, Sahar Tahmasebi, Mahdi Jalili, Parham Moradi

Abstract: In this paper we propose a many-objective optimization method tailored to efficiently address challenges posed by problems with high numbers of objectives. Our method aims to generate solutions that exhibit both convergence and diversity across the objective space. Moreover, a fast non-dominated sorting procedure is employed to ensure the convergence of the evolutionary population. To maintain the diversity of solutions, the method utilizes a novel scoring strategy based on both local and global densities of individuals in the solution space. It selects a population in which each member demonstrates a high local density, ensuring they are appropriately spaced apart to ensure diversity. The result of experiments on two widely used benchmarks alongside with six popular baselines show that the proposed method outperforms its competitors.