Wireless Traffic Data Completion Using Irregular Spatio-Temporal Structure for Internet of Vehicles

The Internet of Vehicles (IoV) has experienced exponential growth in communication traffic fueled by 5G/6G advancements and emerging automotive applications. The substantial communication traffic data offers valuable insights into user demand and serves as a foundation for downstream network optimization. However, challenges including signal degradation and data anonymization result in significant data loss, hindering effective network management. Existing methods for wireless traffic data completion struggle to adapt to the irregular spatio-temporal patterns of IoV, especially in the temporal domain. To address these limitations, we propose a novel graph learning approach that adaptively captures irregular spatio-temporal structures from incomplete data. Building upon this, we develop the joint Graph Learning and Traffic Recovery (GLTR) method, and establish sufficient conditions for its convergence. Experiments on real-world datasets demonstrate that GLTR achieves superior accuracy and robustness compared to state-of-the-art methods, underscoring the benefits of leveraging irregular temporal topologies for traffic data completion in IoV systems.