多目标优化的一种通用锥超体积进化算法

It has been experimentally proven that the conical area evolutionary algorithm (CAEA) has a significantly higher run-time efficiency than the multi-objective evolutionary algorithm based on decomposition (MOEA/D) for bi-objective optimization. However, CAEA can not tackle problems with more than two objectives. In this paper, a conical hypervolume evolutionary algorithm (CHEA) is proposed to extend CAEA from bi-objective spaces to higher dimensional objective spaces. CHEA uses a universal conical partition strategy to partition high dimensional objective spaces into a series of conical subregions. Then a MOP is decomposed into a series of scalar subproblems, each of which is assigned a conical subregion and adopts a local hypervolumn indicator as its scalar objective. Experimental results on tri-objective test problems have revealed that CHEA can obtain the satisfactory overall performance on both run-time efficiency and solution quality.