A Many-objective Evolutionary Algorithm with Reference Points-based Strengthened Dominance Relation

Abstract The main issues for the optimization of many-objective evolutionary are about two aspects: the balance between convergence and diversity, and increasing the selection pressure toward the true Pareto-optimal front. To overcome these difficulties, a new Reference Points-based Strengthened dominance relation (RPS-dominance) is proposed and integrated into NSGA-II, named RPS-NSGA-II. It introduces a reference point set and convergence metric Cov to distinguish Pareto-equipment solutions and further stratifies them. The performance of RPS-NSGA-II is evaluated by the WFG and MaF series benchmark problems. Extensive experimental results demonstrate that RPS-NSGA-II has the competitiveness and frequently better results when compared against the main existing algorithm(five recently proposed decomposition-based MOEAs) on 90 commonly-used benchmark problems involving up to 20 objectives.