Enhancing differential evolution algorithm through a population size adaptation strategy

Abstract

As one of the three basic control parameters of the differential evolution algorithm (DE), the population size (PS) has attracted extensive attention. In general, the most appropriate population size varies for different types of problems and problems with different dimensions. As a result, the performance of an algorithm with a fixed population size is limited to some extent. In this paper, a new enhanced algorithm with a population entropy based population adaptation strategy has been proposed under the framework of SHADE (PE-SHADE). Firstly, a method to calculate the entropy of the population is introduced, through which the distribution state of the population is also characterized. Secondly, the population size is adapted according to the distribution state with a population increasing strategy and a population reduction strategy. In order to evaluate the performance of the proposed algorithm, experiments on the standard benchmark CEC2014 have been conducted, as well as the sensitivity experiments for the extra parameters. The performance comparisons with SHADE, L-SHADE, and some other well-known DE variants are analyzed, which statistically supports the effectiveness of the proposed algorithm.