Improved Self-Adaptive Differential Evolution Based on Exponent Crossover
Yulong Xu, Jian‐an Fang, Zhongyong Liu, Cui Wenxia · ICISEM '13 Proceedings of the 2013 International Conference on Information System and Engineering Management · 2013
Differential Evolution (DE) is well known as a simple and one of the most powerful intelligence algorithms in current. Based on exponent crossover, this paper address to enhance the performance of jDE, which is a famous variant of original DE with self-adaptive control parameter. Firstly, we introduce a new method to measure population diversity by the value of object function. Then we provide a crossover scheme according to aforementioned diversity degree, when above degree is smaller than a threshold value our scheme deploys part of superior target vector and inferior mutant vector for crossing, otherwise, the scheme controls part of superior target and superior mutant vector for crossing. Furthermore, combining this scheme with exponent crossover we can distinctly improve performance of original jDE algorithm. Finally, the comparative study indicates that our new scheme can enhance the performance of conventional jDE on a suite of 6 numerical optimization problems.