A distributed supply chain inventory collaborative replenishment strategy based on edge-agent federated learning
With the continuous expansion of the supply chain scale and the widespread adoption of distributed architectures, how to achieve cross-node inventory collaborative optimization while protecting the data privacy of all participants has become a key challenge. To address this challenge, this paper proposes a distributed supply chain inventory collaborative replenishment strategy based on mean-field approximation and hierarchical federated reinforcement learning. This algorithm constructs a three-layer computing architecture of "cloud-edge-end", combining the macroscopic laws of global group interaction with the regional personalized knowledge learning. At the terminal layer, each warehouse acts as an agent and uses local data for reinforcement learning to optimize its own replenishment decisions; at the edge layer, the intelligent agents in the same region aggregate model parameters through the federated averaging mechanism to form regional shared knowledge; at the cloud layer, by integrating information from each region, a dynamic mean-field representing the global state of the system is generated and issued to guide the collaborative decisions of all agents. This method transforms the complex multi-agent collaborative problem into an interaction problem between individuals and the macroscopic meanfield, reducing the decision dimension while ensuring data privacy and security through the hierarchical federated mechanism. Experimental results show that, compared with traditional independent learning, standard federated learning, and centralized methods, the proposed algorithm demonstrates significant advantages in terms of system total cost, robustness to data heterogeneity, large-scale scalability, and dynamic environment adaptability, and can effectively approach the global optimal collaborative replenishment state while protecting data privacy.