On-chain federated learning approach for Internet of Vehicles

In the field of telematics, the nature of traffic data leads to its scattering, involving numerous traffic objects with individual sensitive information. Consequently, sharing a large amount of data becomes difficult, resulting in the existence of "data silos". To address this issue, this paper proposes a scheme that combines coalition learning and blockchain for the purpose of data sharing and privacy protection. The scheme involves modeling multi-source vehicle data using federated learning and storing the trained model parameters and reputation values of participating vehicles on the blockchain. Furthermore, a reputation value calculation method based on the double subjective logic model is proposed, which analyzes the impact of data source quality on the performance of the federated learning algorithm. This calculation method helps in the selection of the client for federated learning, ensuring efficient screening of data sources, improving sharing efficiency, and achieving privacy protection in data sharing. Finally, a simulation analysis is conducted to evaluate the proposed scheme, and the results demonstrate its capability to filter high-quality data sources in real-time dynamic data exchange scenarios of telematics, thereby enhancing the accuracy of federated learning training.