Secure Aggregation for Federated Learning in VANETs with Anonymous Handover Authentication

In vehicular ad hoc networks (VANETs), federated learning enables vehicles to collaboratively train global models for intelligent transportation without exposing their local data, thereby providing a crucial foundation for applications such as traffic prediction and cooperative driving. However, existing federated learning schemes in vehicular networks often fail to address the challenges posed by dynamic cross Roadside Unit (RSU) switching and the frequent joining and departure of vehicles, limiting their deployment in real scenarios. Therefore, developing a secure and efficient federated learning framework that supports dynamic RSU switching and flexible user participation remains a significant challenge. To address this, we propose a secure aggregation federated learning scheme tailored for VANETs. Compared to existing approaches, our scheme offers several key advantages. It employs an efficient anonymous and switchable identity authentication protocol to ensure both vehicle anonymity and seamless mobility across RSUs. In addition, our scheme allows users to join or quit during training. Security analysis demonstrates that our scheme significantly enhances the confidentiality and integrity of intermediate model parameters. The experimental results show that the efficiency of our scheme is improved by about $36 \%$ on average in the case of no user dropping and by about $\mathbf{4 3 \%}$ in the case of some users dropping.