Differential Evolution Using Mutation Strategy With Adaptive Greediness Degree Control

Abstract

Differential evolution (DE) has been demonstrated to be one of the most promising evolutionary algorithms (EAs) for global numerical optimization. DE mainly differs from other EAs in that it employs difference of the parameter vectors in mutation operator to search the objective function landscape. Therefore, the performance of a DE algorithm largely depends on the design of its mutation strategy. In this paper, we propose a new kind of DE mutation strategies whose greediness degree can be adaptively adjusted. The proposed mutation strategies utilize the information of top t solutions in the current population. Such a greedy strategy is beneficial to fast convergence performance. In order to adapt the degree of greediness to fit for different optimization scenarios, the parameter t is adjusted in each generation of the algorithm by an adaptive control scheme. This way, the convergence performance and the robustness of the algorithm can be enhanced at the same time. To evaluate the effectiveness of the proposed adaptive greedy mutation strategies, the approach is applied to original DE algorithms, as well as DE algorithms with parameter adaptation. Experimental results indicate that the proposed adaptive greedy mutation strategies yield significant performance improvement for most of cases studied.