Machine Learning-Empowered Benders Decomposition for Flow Hub Location in E-Commerce.

This article focuses on the flow hub location problem (FHLP), an extension of classical hub location problems motivated by the flexible and dynamic logistics networks of e-commerce businesses. Unlike traditional models where origins and destinations are fixed, the FHLP simultaneously determines origins, destinations, hub locations, and flow routing, reflecting e-commerce retailers’ agility in leasing warehouse spaces and adapting supplier and customer zone decisions. To address the computational challenges of large-scale FHLPs, the authors develop a novel optimization algorithm combining Lagrangian relaxation, Benders decomposition, and machine learning techniques, including clustering-empowered reformulations and learning-empowered elimination tests and variable reduction. Extensive computational experiments demonstrate that this learning-empowered Benders decomposition (LEBD) method outperforms benchmark algorithms in solution quality and computational efficiency, particularly for large instances with complex capacity constraints. The paper also discusses potential future research directions, such as incorporating uncertainty and profit maximization into the FHLP framework.