Intelligent Network-Level Energy Saving Strategy With STGNN-Driven Traffic Prediction and Path Optimization in Transport Networks and Field Trial

With the evolution of green communication networks, device-level energy saving approaches face diminishing returns due to fundamental hardware limitations, while persistent traffic imbalances in metropolitan transport networks create untapped optimization potential. This work reveals that the inherent flexibility of switch fabrics in live network enables dynamic traffic redistribution, presenting an opportunity for multi-node collaborative optimization to achieve deeper energy savings. This paper proposes a novel network-level energy saving strategy that leverages spatiotemporal graph neural networks (STGNN) and dynamic path optimization to overcome the limitations of conventional device-level approaches. The proposed approach integrates AI-driven traffic prediction with intelligent resource allocation, where the STGNN model captures complex spatiotemporal traffic patterns over joint port-tunnel traffic modeling across large-scale network elements, achieving prediction accuracy with a 19.44% improvement over traditional LSTM models. Combined with K-shortest-path optimization algorithm, the system dynamically reallocates traffic to minimize active switch fabrics while maintaining strict service quality guarantees. Validated through a large-scale field trial involving 1966 network elements, the strategy demonstrates a 7.7% total network energy reduction with up to 20.8% for individual nodes without service disruption. The results highlight the effectiveness of STGNN in capturing complex spatiotemporal traffic dependencies and optimizing network resources. It establishes a new paradigm for green and sustainable transport networks, demonstrating how network-wide coordination can unlock energy savings beyond device-level approaches.