Synchronizing Differential Evolution with a modified affinity-based mutation framework

Differential Evolution is a stochastic, population-based optimization algorithm that has gained wide popularity these days for solving multi-modal, non-smooth, non-convex, and ill-behaved optimization problems. In this research article, we propose a restrictive mutation strategy that helps to probabilistically select individuals for mutation based on the information conveyed by neighboring individuals. The strategy is to develop a generalized approach that can restrict the stochastic selection by a more guided technique depending on distribution of adjacent individuals. Our approach takes into account both the proximity and the gradient estimation of the neighboring members of an individual to compute the selection probability. This framework can be easily integrated with basic DE and its state-of-the-art variants with minor changes. Experimental analysis reveals the superiority of our framework over the original variants when tested on the real parameter benchmark problems proposed in the IEEE Congress on Evolutionary Computation 2005 competition.