中国物理B ›› 2016, Vol. 25 ›› Issue (12): 128904-128904.doi: 10.1088/1674-1056/25/12/128904
• INTERDISCIPLINARY PHYSICS AND RELATED AREAS OF SCIENCE AND TECHNOLOGY • 上一篇
Kai-Quan Cai(蔡开泉), Yan-Wu Tang(唐焱武), Xue-Jun Zhang(张学军), Xiang-Min Guan(管祥民)
Kai-Quan Cai(蔡开泉)1, Yan-Wu Tang(唐焱武)1, Xue-Jun Zhang(张学军)1, Xiang-Min Guan(管祥民)2
摘要:
The genetic algorithm (GA) is a nature-inspired evolutionary algorithm to find optima in search space via the interaction of individuals. Recently, researchers demonstrated that the interaction topology plays an important role in information exchange among individuals of evolutionary algorithm. In this paper, we investigate the effect of different network topologies adopted to represent the interaction structures. It is found that GA with a high-density topology ends up more likely with an unsatisfactory solution, contrarily, a low-density topology can impede convergence. Consequently, we propose an improved GA with dynamic topology, named DT-GA, in which the topology structure varies dynamically along with the fitness evolution. Several experiments executed with 15 well-known test functions have illustrated that DT-GA outperforms other test GAs for making a balance of convergence speed and optimum quality. Our work may have implications in the combination of complex networks and computational intelligence.
中图分类号: (Networks and genealogical trees)