Enhancing Supply Chain Demand Forecasting Through Gated Graph Neural Networks and Federated Learning Systems

Abstract

In the context of increasingly complex global supply chains, accurate demand forecasting is crucial for companies to optimize inventory and reduce costs. However, traditional forecasting methods are often difficult to effectively capture complex interactions in the supply chain, and data privacy protection has also become a major challenge. Therefore, this study proposes a demand forecasting method for enterprise supply chains based on gated graph neural and federated learning. By constructing a gated graph neural network, the dynamic relationship of each link in the supply chain is successfully captured. The experimental results show that the prediction accuracy of this model is improved by 12% compared with the traditional method. In order to further strengthen data privacy protection, we have introduced a federated learning mechanism to realize model training without data leaving the local area. Experiments show that the performance of the model under the federated learning framework is only 4% lower than that of centralized training. Combining the advantages of both, we have built a new forecasting system. When processing large-scale and complex supply chain data, the forecasting accuracy rate is 8% higher than that of a single model while effectively protecting data privacy. This study not only provides a new technical path for modern enterprise supply chain management but also lays a solid foundation for intelligent and efficient supply chain management in the future.