A comprehensive survey of adaptive strategies in differential evolutionary algorithms

Abstract

Classical differential evolution (DE) encounters premature convergence when dealing with diverse optimization problems. This challenge has encouraged extensive research efforts aimed at improving and enhancing the original methodologies. Among the various improvement techniques, adaptive strategies have been universally employed. However, there is a lack of systematic research on the adaptation mechanisms. This work comprehensively investigates the adaptive strategies adopted in DE algorithms. Typical adaptation strategies employed in DE algorithms are refined and summarized, highlighting their characteristics. A new taxonomy of adaptation strategies is proposed, categorizing them based on their primary properties, which include adaptations of control parameters, mutation strategies, population size, search space, learning schemes, and composite adaptations. The advantages and disadvantages of these adaptation strategies are summarized, elucidating their unique characteristics. Additionally, a general framework with an adaptive updating engine is proposed, which can serve as a reference for developing new DE algorithms or improving existing ones. The paper also highlights the challenges and open issues of adaptive strategies, suggesting several promising research directions.