Differential Evolution Algorithm with Dual Information Guidance

Abstract

As an effective tool to solve continuous optimization problems, differential evolution (DE) algorithm has been widely used in numerous fields. To enhance the performance, in recent years, many DE variants have been developed based on the idea of multiple strategies. However, there still exists an issue for them that the strategy selection method relies on the historical search experience. The experience may be suitable for the problems with smooth fitness landscapes, but not for the problems with rugged fitness landscapes. To alleviate this issue, in this work, a new multiple strategies-based DE variant with dual information guidance is proposed, called DIGDE. In the DIGDE, to avoid the unreliable historical search experience, the fitness information and spatial information are utilized simultaneously as a guidance to estimate the evolutionary states for each individual, and then the most appropriate strategy can be chosen correspondingly. To verify the effectiveness of the DIGDE, 52 test functions are included in the experiments, and three well-established DE variants and four other evolutionary algorithms are involved in the performance comparison. The results show that the DIGDE achieves competitive performance on not only the result accuracy but also the convergence rate.