An adaptive differential evolution algorithm based on archive reuse

Abstract

In recent years, differential evolution algorithms based on archives have achieved significant success because archives can increase population diversity and balance the exploration and exploitation of algorithms. However, insufficient utilisation of archives has led to an imbalance between exploration and exploitation. Herein, a new archive-reuse-based adaptive differential evolution (AR-aDE) algorithm framework is proposed that can be applied to (L)SHADE and its variants. It comprises three main strategies. First, a new external archive update method based on a cache mechanism is proposed, in which the archive size is the same as the population size, eliminating the need to adjust its size. Second, influenced by knowledge transfer in multitasking optimisation, we designed a new method of reusing the archive to better utilise the information in it. Finally, the classic parameter adaptation method was improved. The experimental results for the CEC2020 and CEC2021 competition problem sets show that the KR-aDE has a strong competitive advantage.