An adaptive hyper-heuristic expert system for data-driven logistics optimization

Logistics optimization in dynamic and uncertain environments remains a challenging task due to the complexity of decision-making processes and the limitations of fixed-strategy optimization methods. Traditional metaheuristic approaches, while effective, often lack adaptability and fail to adjust their search behavior in response to evolving problem conditions, particularly in large-scale, data-intensive logistics systems. This paper addresses this limitation by proposing a novel hyper-heuristic expert system (HH-ES) that shifts the optimization process from a static solution-level paradigm to a meta-level adaptive reasoning framework. The main contribution lies in integrating hyper-heuristic control within an expert system architecture, enabling dynamic and context-aware selection of search strategies. The proposed framework is designed to operate efficiently in distributed and data-rich environments, where real-time data streams and large-scale operational datasets require scalable and adaptive optimization mechanisms. The framework is built upon a structured low-level heuristic space categorized into constructive, improvement, perturbation, and reconstructive operators. An adaptive selection mechanism, combined with multiple acceptance criteria—including improving-only, simulated annealing, and threshold acceptance—guides the search process through a feedback-driven learning loop, allowing the system to continuously refine its heuristic selection policy using data-driven insights. Numerical experiments conducted on logistics scenarios inspired by Australia Post operations demonstrate that the proposed HH-ES consistently outperforms traditional metaheuristics such as Genetic Algorithm and Scatter Search. In particular, the HH-ES achieves an average cost reduction of approximately 5–7% compared to baseline methods, while reducing performance variance by up to 40%. Furthermore, the convergence rate is improved by nearly 20%, highlighting the efficiency of adaptive heuristic selection and acceptance control. These results confirm that embedding hyper-heuristic mechanisms within expert systems provides a powerful approach for addressing complex optimization problems, paving the way for scalable and data-driven intelligent decision-support systems.