STGNNM: Spatial-Temporal Graph Neural Network with Mamba for Cellular Traffic Prediction

Accurate long-term prediction of cellular traffic is a critical task in the rapidly developing field of intelligent communications. However, due to the high mobility of users and the complex scheduling mechanism within the network, cellular traffic data presents significant spatial-temporal dependencies, which pose a considerable challenge to long-term cellular traffic prediction. Although Transformer-based models perform well in dealing with long-term dependencies, their quadratic computational complexity leads to low efficiency and high overhead. Moreover, existing studies are often insufficient in dealing with the spatial-temporal correlation of cellular network traffic, limiting the further improvement of prediction accuracy. To overcome these challenges, we propose a novel deep learning model, Spatial-Temporal Graph Neural Network with Mamba (STGNNM). First, we introduce a bidirectional Mamba module to capture the dynamic characteristics of the time series. Second, we apply a double-view graph learning module. The Graph Convolutional Network (GCN) captures the characteristics of neighboring base stations, while the Graph Attention Network (GAT) records the relationships between distant base stations. Finally, the bidirectional Mamba module processes spatial-temporal features comprehensively. We conduct extensive experimental evaluations on a real-world cellular traffic dataset. The results show that STGNNM outperforms the current state-of-the-art methods in all evaluation metrics, demonstrating its superior performance and effectiveness in cellular network traffic prediction.