A Smart Scheme for Variable Selection in Partial Opposition-based Differential Evolution

Abstract

The concept of opposition-based learning (OBL) has been applied to a growing number of research works, as such we proposed a new scheme which improves effectiveness of a population-based optimization algorithm, called Differential Evolution. In this paper, we propose a novel partial opposition-based DE scheme. Given that the partial OBL reduces the likelihood of converting good variables to their opposites, the main challenge of this technique is still the selection of variables which should be replaced with their opposites. The proposed scheme addresses this issue by creating a reference solution using averaging on the best candidate solutions. This reference solution plays a crucial role in intelligent selection of variables to compute their corresponding the opposite value based on their distance to the set of best solutions. The proposed partial opposition-based scheme is embedded in Differential Evolution (DE) algorithm and compared with opposition-based DE (ODE) and DE using the set of CEC-2014 benchmark functions on dimensions 30, 50, and 100. Experimental results indicate a promising improvement over DE and ODE.