Improving SHADE with Center-based Mutation for Large-scale Optimization

Differential Evolution is a powerful and efficient approach for numerical optimization. A Success-History Based Parameter Adaptation (SHADE) is the recent variant of the adaptive DE that utilizes a historical performance of the successful control parameter. In this paper, we propose a center-based mutation for SHADE algorithm (CSHADE). In this mutation scheme, the base vector for SHADE's mutation is replaced with center-based sampled candidate solution using the normal distribution. The proposed method is evaluated on CEC-2010 and CEC-2013 LSGO benchmark functions with dimension 1000. The experimental results show that CSHADE outperforms SHADE algorithm over the majority of benchmark functions in terms of solution accuracy.

I. Introduction

Many real-world applications involve an optimization process that handles a large number of decision variables. Hence, it becomes a necessity to develop efficient algorithms in order to tackle complex large-scale global optimization (LSGO) problems. Several population-based meta-heuristic algorithms have been proposed to tackle such problems including Genetic algorithm (GA), Particle Swarm Optimization (PSO), and Differential Evolution (DE). These algorithms posses a number of advantages, such as global search capability, robustness, and potential capability of parallelism [1].