Federated Learning for VANET Based on Homomorphic Encryption

Abstract

VANET is a key technology to realize intelligent transportation services in smart cities. The traditional VANET cloud intelligent model has the risk of user privacy disclosure. In this paper, a federated learning (FL) method for VANET based on homomorphic encryption is proposed. Node functions are reasonably designed according to VANET scenarios, and local training node algorithm, node selection algorithm and global model update algorithm are designed. The analysis shows that the scheme can improve the security of the distributed training model based on FL. The user privacy of vehicle nodes could be protected, and the common malicious node attacks could be resisted, such as routing spoofing attacks, witch attacks, wormhole attacks and black hole attacks.