A review of the recent use of Differential Evolution for Large-Scale Global Optimization: An analysis of selected algorithms on the CEC 2013 LSGO benchmark suite

Abstract

Abstract This paper gives a review of recent extensions of the Differential Evolution (DE) algorithm for use in Large-Scale Global Optimization (LSGO) and presents an empirical analysis of DE-based and some other state-of-the-art algorithms for LSGO on the CEC 2013 LSGO benchmark suite. Since witnessing the first successful applications of DE for a wide variety of optimization problems in the early nineties, researchers have developed several new algorithms in this field. In this paper, we are especially interested in algorithms for solving LSGO. As LSGO is one of the most active research lines, not only in DE, but in many evolutionary and meta-heuristic algorithms, we discuss general approaches for dealing with LSGO first. The main focus of the paper is DE. We review its basic algorithm and discuss several extensions used for coping with large-scale problems. This paper has two main objectives: (1) To propose, from a theoretical point of view, the grouping of DE mechanisms for dealing with LSGO into nine groups, and (2) To evaluate sixteen recently proposed algorithms for LSGO empirically. Many benchmark suites were designed with the aim of providing a suitable evaluation platform for testing and comparing large-scale optimization algorithms. In this paper, the CEC 2013 LSGO benchmark suite was chosen for comparison, because it resembles the following features of real-world problems: Non-uniform subcomponent sizes; imbalance in the contribution of subcomponents; and functions with interdependent overlapping subcomponents. The performances of state-of-the-art algorithms are compared, and the algorithms are ranked using three different metrics, which evaluate the performance from different perspectives. The conducted research shows that DE is among the best algorithms for LSGO on the CEC 2013 LSGO benchmark suite, especially when used with other mechanisms for dealing with large numbers of variables. Finally, the analysis has shown that there is still some room for further improvements in DE towards the solution of LSGO problems.