Adaptive social mobility-restructuring differential evolution for global optimization

Abstract

Differential evolution (DE) is widely recognized as a highly effective algorithm for global optimization. As a simple yet powerful strategy, the greedy selection operator is employed in most existing DE algorithms. However, this strategy focuses solely on individual-level evolution while neglecting population-level evolution. Consequently, it fails to fully exploit the holistic information of the population during the evolutionary process, thereby limiting the performance of DE. To address the issue above, this study proposes a novel DE variant called adaptive social mobility-restructuring DE (ASMRDE). Specifically, a social generation selection mechanism (social restructuring) is proposed enabling population-level evolution to occur alongside individual-level evolution with the iteration of social generations. This dual-level evolution effectively addresses the aforementioned limitation. Additionally, an external archive is constructed to preserve historical populations eliminated during the social restructuring process. This archived information is subsequently utilized in both social restructuring and individual evolution. Furthermore, this study also develops a parameter adaptation strategy based on narrow-sense diversity and a new mutation strategy that guides individual-level evolution using historical population information. Finally, the proposed variant was thoroughly assessed by using the Congress on Evolutionary Computation (CEC) 2017, 2014 and 2013 benchmark functions, with results showing ASMRDE’s remarkably robust performance and distinct advantages for complex problems when compared to other efficient DE variants, including some winning algorithms in the CEC competition. Besides, the sensitivity of ASMRDE to parameter changes was explored, and ablation experiments and visual analyses were performed on ASMRDE, so as to demonstrate the effectiveness of the above strategies.

Topics & Concepts