Blockchain-Based Federated Learning Algorithm for Secure Data Sharing in Internet of Vehicles

Internet of Vehicles (IoV) can effectively improve driving experience and enhance in-vehicle services through data sharing. However, how to ensure security and privacy during data sharing is a challenge. In this paper, a hierarchical federated learning algorithm with adaptive weights (AW-FL) is proposed, where vehicles utilize distributed machine learning to learn from environmental sensing data and share their learning models with each other. AW-FL employs a hybrid blockchain framework based on Practical Byzantine Fault Tolerance (PBFT) and Directed Acyclic Graph (DAG) blockchain, aiming to meet IoV scenarios and security and privacy requirements. In addition, in order to achieve effective detection of malicious models uploaded by abnormal vehicles, a Tip Selection Algorithm (TSA) based on accuracy and historical contribution is presented for the DAG blockchain. Simulation results show that AW-FL is capable of improving the sharing efficiency and learning quality. Moreover, the proposed TSA can effectively deal with certain malicious attacks.