Enhancing Differential Evolution with Interactive Information

Li Ming Zheng,Lu Liu,Sheng Xin Zhang,Shao Yong Zheng

DOI: https://doi.org/10.1007/s00500-017-2740-2

IF: 3.732

2017-01-01

Soft Computing

Abstract:Differential evolution (DE) is well known for its simple structure and excellent performance among various evolutionary algorithms. Difference vectors have a dominant effect on the evolution progress. But the difference vectors in mutation operators for the conventional DE are simply generated by selecting individuals from the current population without any selective pressure. Besides, the directional information only depends on the existing individuals and hardly exploits the interaction between individuals. Therefore, a novel interactive information scheme called IIN is proposed to overcome this weakness. It attempts to provide more effective directional information during the evolution process and achieve a good balance between exploration and exploitation. In IIN, both the ranking information based on fitness and the interactive information between individuals is fully considered. The interaction between individuals is implemented by the mathematically weight-based combination according to ranking information. Hence, the interactive information inherited from existing individuals acts as a directional vector. In this way, IIN-DE utilizes the directional information to speed up convergence. The proposed scheme can be easily incorporated into different mutation strategies to provide useful directional information. To verify the effectiveness, the proposed IIN is incorporated into the original DEs based on several mutation operators as well as several state-of-art DE variants. With the incorporation of IIN, significant improvements can be achieved for most of the compared DEs, as demonstrated by the experimental results.