A New Method of Population Initialization for Enhancing Performance of Evolutionary Algorithms

Evolutionary algorithms have been used in numerous real-world applications and proven to have virtuous convergence properties. Despite several drawbacks, population initialization in absence of any priori information available is one of them to affect the convergence rate. To address this problem, we propose a new method of population initialization based on the opposite point. Differential evolution (DE), one of the well-known evolutionary algorithms, has been embedded with this new method of population initialization. Differential evolution with this initialization method is called opposite point-based differential evolution (OPDE). In this paper, we discuss and evaluate the performance of DE and OPDE algorithms. In particular, an empirical comparison between results obtained using DE and OPDE is presented along with those reported in the literature. To assure a fair comparison, we test the algorithms using twenty-four well-known unimodal and multimodal, low-dimensional and high-dimensional, benchmark functions with three different termination criteria. The results demonstrate that OPDE successfully accelerates with a high convergence rate and outperforms the DE and other opposition learning-based differential evolution reported in the literature for most of the benchmark functions considered in the present study.