Federated Generative Artificial Intelligence Empowered Traffic Flow Prediction Under Vehicular Computing Power Networks
Yujie Ye, Z.X. Zhao, Lei Liu, Jie Feng, Jun Du, Qingqi Pei
Abstract
Traffic flow prediction holds great promise in prompting the rapid development of intelligent transportation systems. The key challenge for traffic flow prediction lies in effectively modeling the complicated spatiotemporal dependencies of traffic data while considering privacy and cost concerns. Existing methods based on neural networks exhibit limitations, particularly in handling dynamic data and long-distance dependencies. To address these challenges, we have proposed a novel distributed traffic flow prediction architecture that makes the integration of generative artificial intelligence (AI) and hierarchical federated learning. This architecture makes the prediction of traffic flow by incorporating spatial self-attention module and traffic delay-aware feature transformation module, which achieves a better balance between communication and computation costs, enhances training efficiency and guarantees data privacy and security. Next, we have introduced the important characteristics and key technologies used for this devised architecture. Finally, several open issues are given with the aim to attract more attentions for further investigation.