Improved Differential Evolution for Large-Scale Black-Box Optimization

Abstract

The demand for solving large-scale complex problems continues to grow. Many real-world problems are described by a large number of variables that interact with each other in a complex way. The dimensionality of the problem has a direct impact on the computational cost of the optimization. During the last two decades, differential evolution has been shown to be one of the most powerful optimizers for a wide range of optimization problems. In this paper, we investigate its appropriateness for large-scale problems. We propose a new variation of differential evolution that exhibits good results on difficult functions with a large numbers of variables. The proposed algorithm incorporates the following mechanisms: the use of three strategies, the extended range of values for self-adapted parameters $F$ and $\mathit {CR}$ , subpopulations, and the population size reduction. The algorithm was tested on the CEC 2013 benchmark suite for large-scale optimization, and on two real-world problems from the CEC 2011 benchmark suite on real-world optimization. A comparative analysis was performed with recently proposed algorithms. The analysis shows the superior performance of our algorithm on most complex problems, described by overlapping and non-separable functions.