The Research On Privacy-Preserving Scheme Of Federated Learning In Internet Of Vehicles

With the development of communication technology and computer technology,the Internet of Vehicles(IoV)has been widely used in the traffic system.Meanwhile,vehicle sensors,roadside units(RSUs),and other devices have collected massive user data.The emerging federated learning technology can train local models with these data,and aggregate local models into global models via a central server.In the scenario of IoV,federated learning requires a large amount of user private data to develop applications such as autonomous driving and vehicle energy demand analysis,which will increase the risk of privacy leakage.Therefore,it is of great practical significance to preserve the data and identity privacy of federated learning users in IoV.Federated learning in IoV faces internal attacks and "single point of failure" caused by centralization.Meanwhile,the privacy of user data and identity would be revealed under the attacks such as membership inference and model inversion.Blockchain can provide a decentralized solution for federated learning in IoV,effectively resisting malicious attacks from the central server.Therefore,based on blockchain and cryptography,this dissertation studies the data and identity privacy-preserving scheme of federated learning users in IoV,establishes the mutual trust mechanism in the application scenario of IoV,and preserves the privacy and security of users.The specific researches of this dissertation are as follows:(1)Data privacy-preserving scheme of federated learning in IoV.Aiming at the issue of data privacy leakage faced by IoV users in the federated learning process,a data privacy-preserving scheme of federated learning in IoV was proposed based on sharding blockchain,homomorphic encryption,and differential privacy.First of all,by scenario-based sharding technology and proof of time-sensitive contribution,this scheme improves the throughput and scalability of traditional blockchain while retaining decentralization and immutability.Secondly,a three-layer federated learning architecture is designed to provide private models,local models,and global models according to users’ personalized model requirements.Finally,to avoid the leakage of user data privacy in the federated learning process,CKKS fully homomorphic encryption is used to preserve the data privacy of IoV users in the computation offloading stage of private models,and differential privacy technology based on the Laplace mechanism is used to resist the attacks of membership inference and model inversion in the aggregation stage of local models and global models.The experimental analysis and scheme comparisons show that this scheme can effectively preserve the data privacy of federated learning users in IoV and improve scalability.(2)Identity privacy-preserving scheme for energy trading in the IoV.Aiming at the issue of that user identity information is used to realize the sample alignment of vertical federated learning in the energy trading scenario of IoV,this dissertation proposes an energy trading scheme with user identity anonymity based on blockchain technology and identitycommittable signature technology to preserve user identity privacy in the data collection stage of federated learning.First of all,the vehicles and charging service providers are incorporated into the trading network based on the consortium-sharding blockchain,and the decentralized energy trading process is realized through the identity verification and trading confirmation processes performed by smart contracts.Then,based on the identity-committable signature,trading anonymity is achieved.This scheme avoids the problem that the lengths of signatures increase linearly with the number of group members using traditional ring signature technology.Security analysis and scheme comparison show that this scheme can preserve the users’ privacy of identity while providing high scalability.