A Hierarchical Asynchronous Federated Learning Privacy-Preserving Framework for IoVs

Data sharing plays a crucial role in the Internet of Vehicles, as it greatly enhances the driving experience for users. Federated Learning (FL) has shown good advantages and efficiency in knowledge sharing among vehicles. However, due to the uncertainty of the IoVs, the existing federated learning frameworks cannot meet the high-precision, fast convergence, and high fault tolerance requirements in the learning process. To address these issues, this paper proposes a hierarchical federated learning framework for IoVs environment that combines synchronous and asynchronous methods to improve machine learning performance in the Internet of Vehicles environment. The proposed asynchronous algorithm can improve the accuracy of the global model via controlling the proportion of parameters submitted by users. In addition, to improve the reliability of the parameters, our framework provides a malicious node exclusion algorithm to improve the reliability of the parameters. It effectively reduces the adverse impact of malicious parameters on the global model. Finally, lightweight pseudonym is used in the proposed framework to ensure the privacy of participants' identities. The experimental results demonstrate that the proposed framework achieves high learning accuracy and fast convergence speed. Additionally, it effectively defends against poisoning attacks and ensures the protection of participants' identity privacy.