Deep learning and multi-agent simulation for spatiotemporal inventory management in e-commerce

E-commerce inventory management faces the dual challenge of handling complex spatiotemporal demand variations and coordinating decisions across distributed warehouse networks.To address this, we propose an integrated simulation framework that combines deep learning for demand forecasting with multi-agent reinforcement learning for inventory optimisation.Our approach employs a spatiotemporal graph network to capture demand dependencies and a value-decomposition network for collaborative decision-making.Validated on the real-world M5 dataset, the framework demonstrates significant improvements in both forecast accuracy and operational efficiency compared to traditional and state-of-the-art methods.Notably, it achieves a substantial reduction in total costs while maintaining high service levels.This work contributes a novel, scalable solution for intelligent inventory management, with direct implications for enhancing supply chain resilience in e-commerce.