Adaptive Fusion Optimization (AFO): A Hyper-Adaptive Metaheuristic with Learning-Driven Control for Global Minimization

Introduction: Global minimization in complex, nonlinear search spaces is challenging due to premature convergence and high parameter sensitivity in classical metaheuristics.Objective: This study proposes Adaptive Fusion Optimization (AFO), a hyper-adaptive metaheuristic designed to balance exploration and exploitation through dynamic operator control and online feedback.Method: AFO integrates a compact fusion pool of three operators with a self-evolving exploration factor and a lightweight learning-driven controller for operator selection. The algorithm adapts its search behavior using fitness improvement and population diversity, avoiding fixed switching rules. Performance is evaluated on standard benchmark functions under a fixed function-evaluation budget and compared with GA, PSO, DE, and ACO.Results:AFO demonstrates superior accuracy, faster convergence, and stronger robustness, especially on multimodal functions. In 30 dimensions, it achieves mean final fitness of 3.1×10⁻² on Rastrigin and 6.2×10⁻³ on Ackley. In 50 dimensions, robustness remains high, with Ackley showing a standard deviation of about 4.1×10⁻³ over 30 runs. Statistical tests at 95% confidence confirm the improvement, with Friedman ranking placing AFO first (mean rank 1.20, p = 2.1×10⁻⁶), supported by Wilcoxon pairwise tests.Conclusions: AFO provides a reliable framework for global minimization and shows promise for extension to constrained, multi-objective, and real-world optimization tasks such as energy-efficient scheduling and power dispatch.