面向高动态车载网的可验证联邦学习方案OA

With the rapid development of intelligent transportation systems,federated learning(FL)has made remarkable progress in vehicular ad-hoc networks(VANETs),establishing itself as a crucial approach to improve data sharing efficiency and privacy protection.However,the existing solutions face significant challenges arising from the dynamic nature of the networks,the rapid mobility of vehicles,and frequent environmental changes,particularly in terms of privacy preservation,malicious vehicle management,verification mechanisms,and so on.To address these challenges,a novel and verifiable FL scheme for VANETs is proposed in this paper.First,an enhanced Paillier homomorphic encryption algorithm is integrated into the framework,which not only strengthens privacy protection but also reduces computational overhead,thus meeting the frequent data processing requirements of VANETs.Second,a malicious vehicle revocation and tracking mechanism,based on group keys and the cosine similarity,is introduced to manage the dynamic joining and leaving of malicious vehicles,thereby enhancing the overall security of the system.Finally,a lightweight aggregate signature verification mechanism is implemented using elliptic curve-based certificate-free signatures,effectively mitigating the high overhead typically associated with traditional certificate management while reducing the storage and communication burdens on vehicles,thereby improving the efficiency of message authentication.Meanwhile,the communication environment and vehicle dynamic information processing mechanism are optimized to enhance the scheme's adaptability to the highly dynamic VANET environment.Experimental results demonstrate that the proposed scheme significantly outperforms the existing methods in terms of both privacy protection and computational efficiency,offering a practical solution for the deployment of FL in vehicular network environments.