A novel adaptive optimization scheme for advancing metaheuristics and global optimization

Abstract

Metaheuristics have been the dominant approach for tackling complex optimization challenges across diverse disciplines. Numerous studies have sought to enhance the performance of existing metaheuristics by identifying their limitations and modifying their frameworks. Despite these efforts, many resulting strategies remain overly complex, often narrowly focused on a single algorithm and a specific problem domain. In this study, we introduce a novel adaptive optimization scheme (AOS) designed as an algorithm-independent mechanism for enhancing the performance of metaheuristics by addressing various optimization challenges. This scheme is developed through a comprehensive integration of three substructures, each aimed at mitigating common deficiencies in metaheuristics across three optimization pillars: high exploration capabilities, effective avoidance of local optima, and strong exploitation capabilities. Three prominent approaches—Lévy Flights, Chaotic Local Search, and Opposition-based Learning—are skillfully combined to overcome these shortcomings in various metaheuristic algorithms, establishing a straightforward unit. Through rigorous testing on 50 diverse mathematical benchmark functions, we assessed the performance of original metaheuristics and their AOS-upgraded versions. The results confirm that the proposed AOS consistently elevates algorithmic effectiveness across multiple optimization metrics. Notably, four AOS-upgraded algorithms—EO-AOS, HBA-AOS, DBO-AOS, and PSO-AOS—emerge as the leading performers among the 16 algorithms under evaluation. Comparisons between the upgraded and baseline metaheuristics further reveal the substantial impact of AOS, as each upgraded variant demonstrably surpasses its original algorithm in various optimization capabilities.