MjSO: A modified differential evolution with a probability selection mechanism and a directed mutation strategy
Yintong Li, Tong Han, Xiaofei Wang, Huan Zhou, Shangqin Tang, Changqiang Huang, Bo Han
Abstract
To further enhance the performance of jSO, the most representative variant of differential evolution, a modified differential devolution with a probability selection mechanism and a directed mutation strategy named MjSO is developed in this study. There are three main modifications for jSO. Firstly, a novel probability selection mechanism is utilized in mutation process to fully take advantage of dominant individuals. Specifically, the better dominant individual has a higher probability of being selected. Then, a directed mutation strategy is adopted to ensure the evolution towards a better direction during the optimization. Moreover, the archive scheme is no longer adopted in MjSO to avoid the excessive use of historically dominant individuals. MjSO is evaluated using the CEC 2018 test suite in comparison with six state-of-the-art DE variants, called jSO, mL-SHADE, ELSHADE-SPACMA, EB-LSHADE, iL-SHADE, and MadDE. The numerical experimental results certify that MjSO outperforms the six comparison algorithms. Moreover, MjSO significantly enhances the performance of jSO while reducing its time complexity. In conclusion, the probability selection mechanism and directed mutation strategy proposed in this study efficiently enhance the performance of jSO, and MjSO is a promising algorithm.