Decentralizing and coevolving differential evolution for large-scale global optimization problems

Abstract

This paper presents a novel decentralizing and coevolving differential evolution (DCDE) algorithm to address the issue of scaling up differential evolution (DE) algorithms to solve large-scale global optimization (LSGO) problems. As most evolutionary algorithms (EAs) display their weaknesses on LSGO problems due to the exponentially increasing complexity, the cooperative coevolution (CC) framework is often used to overcome such weaknesses. However, the cooperative but greedy coevolution of CC sometimes gives inferior performance, especially on non-separable and multimodal problems. In the proposed DCDE algorithm, to balance the search behavior between exploitation and exploration, the original population is decomposed into several subpopulations in ring connection, and the multi-context vectors according to this connection are introduced into the coevolution. Moreover, a novel “DE/current-to-SP-best-ring/1” mutation operation is also adopted in the DCDE. On a comprehensive set of 1000- dimensional benchmarks, the performance of DCDE compared favorably against several state-of-the-art LSGO algorithms. The experimental analysis results suggest that DCDE is a highly competitive optimization algorithm on LSGO problems, especially on some non-separable and multimodal problems.