Hybrid Sampling Evolution Strategy for Solving Single Objective Bound Constrained Problems

Geng Zhang, Yuhui Shi

Abstract: This paper proposes an evolution strategy (ES) algorithm called hybrid sampling-evolution strategy (HS-ES) that combines the covariance matrix adaptation-evolution strategy (CMA-ES) and univariate sampling method. In spite that the univariate sampling has been widely thought as a method only to separable problems, the analysis and experimental tests show that it is actually very effective for solving multimodal nonseparable problems. As the univariate sampling is a complementary algorithm to the CMA-ES which has obvious advantages for solving unimodal nonseparable problems, the proposed HS-ES tries to take advantages of these two algorithms to improve its searching performance. Experimental results on CEC-2018 demonstrate the effectiveness of the proposed HS-ES.