Research on Decentralized Federated Learning System for Vehicle Data Privacy Protection Based on Blockchain

This paper proposes an edge computing vehicle data privacy protection method based on blockchain and federated learning. Using blockchain to give edge computing tamper-proof and anti-single point of failure attacks and other characteristics. Integrate gradient verification and incentive mechanism into consensus protocol. By constructing master-slave chain architecture among vehicles, roadside units and base stations, distributed model security sharing is realized. An asynchronous federated learning algorithm based on incentive mechanism is proposed to motivate vehicles and roadside units to participate in the optimization process. The framework is verified on real locus data sets. LSTM self-coding is used as embedded learning network. The prototype of the proposed architecture is implemented in the experiment, and its feasibility, accuracy and performance are evaluated. The results show that this method maintains the integrity of data and has good prediction accuracy and performance.