Generative AI–enhanced sales forecasting and optimization for the multi-period multi-product inventory routing problem with procurement
针对自营电商的多周期多产品库存路径与采购问题，提出结合生成式与判别式AI的预测方法，并用混合启发式算法优化采购、运输和仓储决策，案例显示能显著降低成本。
In the current e-commerce domain, rising customer demands for diversity, responsiveness, and service quality create major challenges in inventory management and logistics optimisation. To address these, this paper introduces the multi-period and multi-product inventory routeing problem with procurement decisions (MIRP-PD) in self-operated e-commerce, supported by a generative AI and discriminative AI–based forecasting method. The goal is to optimise (i) procurement from geographically dispersed suppliers, (ii) transportation to a central warehouse, and (iii) product pickup from suppliers to the warehouse. Based on AI-generated forecasts, an integer programming model for MIRP-PD is developed. To solve medium- and large-scale problems, a hybrid bi-level heuristic is proposed, combining genetic algorithms (GA) for procurement planning and ant colony optimisation (ACO) for routeing, enhanced by a Lagrangian constraint–based repair operator. A rolling-horizon framework is further applied to mitigate forecast errors. A real-life case study with 15 scenarios demonstrates that the proposed GA–ACO achieves superior performance compared with Gurobi and a GA-only baseline. Comparative execution tests confirm that AI-based forecasting substantially reduces excess holding, transportation, and stockout costs. Sensitivity analyses provide managerial insights into transport strategies, warehousing–transport trade-offs, and service-level penalties, highlighting the role of generative and discriminative AI in enabling robust replenishment decisions.