Enhanced θ dominance and density selection based evolutionary algorithm for many-objective optimization problems

摘要

Many multi-objective evolutionary algorithms (MOEAs) have been developed for many-objective optimization. This paper proposes a new enhanced ? dominance and density selection based evolutionary algorithm (called ?-EDEA) for many-objective optimization problems. We firstly construct an m-dimension hyper-plane using the extreme point on the each dimension. Then we replace the distance between the origin point and projection of solution on the reference line of ? dominance which recently is proposed in ? dominance based evolutionary algorithm (?-DEA), with the perpendicular distance between each solution and the hyper-plane to develop an enhanced ? dominance. Finally, in order to maintain better diversity, ?-EDEA employs density based selection mechanism to select the solution for the next population in the environment selection phase. ?-EDEA still inherits clustering operator and ranking operator of ?-DEA to balance diversity and convergence. The performance of ?-EDEA is validated and compared with five state-of-the-art algorithms on two well-known many-objective benchmark problems with three to fifteen objectives. The results show that ?-EDEA is capable of obtaining a solution set with better convergence and diversity.