A multi-population evolutionary algorithm with single-objective guide for many-objective optimization

Abstract

This paper develops a multi-population evolutionary algorithm with single-objective guide to tackle many-objective optimization problems. It exploits the merits of both multiple populations and single-objective optimization to balance diversity and convergence of the evolution process. Specifically, the single-objective guide process helps to construct the better ideal point and reference points. A novel objective space partitioning mechanism is developed to transform a many-objective optimization problem into multiple subproblems, each of which is tackled by a subpopulation. A novel information sharing mechanism between subpopulations is proposed to balance diversity and convergence. Finally the subpopulations are merged together and deal with many-objective optimization problems to further enhance the convergence. We have compared the performance of the proposed algorithm with nine state-of-the-art algorithms on 85 test instances of 21 benchmark problems with up to 15 objectives. Experimental results show that the proposed algorithm has the superior performance in solving multi- and many-objective optimization problems.