Self-adaptive differential evolution with neighborhood search

In this paper we investigate several self-adaptive mechanisms to improve our previous work on NSDE [1], which is a recent DE variant for numerical optimization. The selfadaptive methods originate from another DE variant, SaDE [2], but are remarkably modified and extended to fit our NSDE. And thus a Self-adaptive NSDE (SaNSDE) is proposed to improve NSDE's performance. Three self-adaptive mechanisms are utilized in SaNSDE: self-adaptation for two candidate mutation strategies, self-adaptations for controlling scale factor F and crossover rate CR, respectively. Experimental studies are carried out on a broad range of different benchmark functions, and the proposed SaNSDE has shown significant superiority over NSDE. © 2008 IEEE.