An Efficient Recursive Differential Grouping for Large-Scale Continuous Problems

Cooperative co-evolution (CC) is an efficient and practical evolutionary framework for solving large-scale optimization problems. The performance of CC is affected by the variable decomposition. An accurate variable decomposition can help to improve the performance of CC on solving an optimization problem. The variable grouping methods usually spend many computational resources obtaining an accurate variable decomposition. To reduce the computational cost on the decomposition, we propose an efficient recursive differential grouping (ERDG) method in this article. By exploiting the historical information on examining the interrelationship between the variables of an optimization problem, ERDG is able to avoid examining some interrelationship and spend much less computation than other recursive differential grouping methods. Our experimental results and analysis suggest that ERDG is a competitive method for decomposing large-scale continuous problems and improves the performance of CC for solving the large-scale optimization problems.

I. Introduction

Large-scale optimization problems involve at least thousands of decision variables [1], [2]. It is challenging for evolutionary algorithms (EAs) [3] to solve such kind of large-scale optimization problem [4]–[6]. Cooperative co-evolution (CC) [7] adopts the divide-and-conquer strategy [8]–[10] to solve optimization problems. CC divides the variables into several subcomponents and optimizes the subcomponents separately. The divide-and-conquer strategy can decrease the difficulty of solving the large-scale optimization problems [11]–[15].