Dynamic fitness landscape-based adaptive mutation strategy selection mechanism for differential evolution

Abstract

Differential evolution (DE) is the most efficient evolutionary algorithm widely used to solve continuous or discrete numerical optimization problems. However, the performance of DE highly depends on the choice of mutation strategy. In addition, for a given optimization problem, a different mutation strategy configuration of DE may be more effective than a single mutation strategy in searching for the optimal solution. Based on these observations, a dynamic fitness landscape-based adaptive mutation strategy selection differential evolution (DFLDE) is proposed in this paper. In DFLDE, the optimal mutation strategy selection is based on each optimization problem's dynamic fitness landscape characteristics during the evolutionary process. The CEC2017 benchmark function set is used to evaluate the performance of the proposed DFLDE algorithm. The experimental results indicate that the DFLDE is superior to the other five well-known DE variants in searching for the optimal value, convergence speed, and robustness.