Inventory routing problem of automotive parts considering time-varying demands: A machine learning enhanced branch-and-price approach

With the rise of mass customization and smart manufacturing, the automotive industry is rapidly transitioning to improve responsiveness, manage highly diversified customer orders, and reduce inbound logistics costs. To address this challenge, this paper proposes a new variant of the multi-period inventory routing problem, which focuses on coordinating discrete, time-varying demands for auto parts on the assembly line with predetermined packages at suppliers over a finite short-term time horizon (e.g., on an hourly basis). The objective is to minimize the total transportation and inventory cost by making aperiodic decisions on collection quantities and traveling routes simultaneously for an inbound warehouse near the assembly plant. An integer programming (IP) formulation with time-indexed variables is tailored for the problem to analyze the feasibility conditions. Then, a reformulation is designed to make the problem more tractable, based on which a novel machine learning enhanced branch-and-price algorithm (BPL) is proposed, where prediction-based cuts are embedded to accelerate the pricing procedure. Experiments on real-scale instances demonstrate that the algorithm consistently achieves near-optimal solutions, with a gap of 4.42% on average from the best-found lower bound, and reduces computation time by over 90% compared to directly solving the IP model by CPLEX. The proposed learning technique is computationally efficient, capable of shortening the total calculation time by an average of 13%. This work facilitates timely decision-making and offers new insights into multi-period inventory routing for inbound logistics.