Research On Vehicular Information Privacy Protection For Internet Of Smart Vehicles

With the development of communication technology and vehicle industry,Internet of Vehicles(IoV)has gradually evolved towards the direction of Internet of smart vehicles.With the trends of network scale expansion,layered heterogeneous network architecture,multi-dimensional vehicular information privacy,and highly dynamic vehicle connection.These trends make the privacy and security design of Internet of smart vehicles more challenging than traditional IoV,specifically in terms of the potential information leakage problem brought by the diversification and intelligence of vehicle business,and the adaptation problem of privacy protection mechanism brought by the heterogeneity and high dynamics of Internet of smart vehicles.This dissertation combines Federated Learning(FL)architecture,Machine Learning(ML)algorithm and cryptography theory to study the privacy protection scheme design of vehicular information for Internet of smart vehicles from two dimensions:spontaneous privacy protection and passive privacy protection.The dissertation aims to solve the privacy leakage problem of Internet of smart vehicles and seek the balance between vehicular privacy and vehicular data availability.The main research results are summarized as follows.1)Research on spontaneous privacy protection of IoV based on single vehicle intelligence:For autonomous driving business in IoV,we design an autonomous driving experience sharing model based on FL architecture to solve the model deviation problem caused by limited vehicle computing resources.A traceable model information protection scheme and an anonymous identity information protection scheme is proposed with consideration of model information and identity information characteristics,respectively.Then,we analyze the reliability of privacy protection from the vehicle perspective and establish the autonomous driving model deviation minimization problem on the basis of proposed improved homomorphic encryption and zero knowledge proof(ZKP)data transfer methods.We use cryptographic theory and probability theory to analyze the effective privacy protection probability of the proposed scheme and prove the feasibility of the scheme.Simulation results show that the proposed algorithm can protect the privacy of raw data and can improve the accuracy of autonomous driving by 5.55%compared to single-vehicle autonomous driving model training.2)Research on spontaneous privacy protection of IoV based on group vehicle intelligence:Based on research 1),we pay attention to the object recognition module of autonomous driving business with further consideration of the limited computing capacity of single vehicle and the low latency demand of IoV.We establish an optimal model of vehicle object recognition based on swarm intelligence with a FL-like architecture.An improved model classifier and its allocation scheme are designed based on the characteristics of FL model aggregation.Considering the requirements of high dynamic,low latency and strong privacy of vehicle swarm intelligence,we propose a multi-objective multi-agent vehicle intelligence clustering scheme based on deep Q-learning(DQL)to establish the group reward maximization problem and analyze the success probability of clustering.Considering the diversity of input variables and high dynamic changes,we adopt the Sharpley Value in game theory to find the kernel of the cluster optimization problem,accelerate the algorithm,and solve the training latency problem that traditional DQL brings.Simulation results show that the proposed scheme can protect the privacy of raw data and can reduce the training time by 50%compared to the centralized object recognition at the expense of only 0.7%recognition accuracy.3)Passive privacy protection research of Social Internet of Vehicles(SIoV):To address the contradiction between the demand for large-scale and high-quality data in SIoV and the problem of data isolated island problem caused by vehicular information privacy protection,we propose a passive privacy protection scheme facing cross-platform vehicular data utilization.Considering the characteristics of multi-dimensional information and diversified privacy demandss in SIoV,we transform the redundant vehicular information into a social relationship graph through graph representation learning to solve the problem of large-scale multi-dimensional information storage and transmission.Then,we use graph differential privacy(GDP)method to analyze the similarity of the obtained social graph structure and realize an node-edge combined GDP.Further,we reuse the data after GDP processing to simulate the business of vehicle dispatching platform.With the combination of real-world dispatching data,the impact of GDP on data availability and business efficiency are analyzed.Simulation results show that with data dimension increase,the proposed algorithm can meet the dynamic adjustment of data availability and privacy in cross-platform utilization,on the premise of ensuring the vehicles get effective passive privacy protection.4)Research on passive privacy protection of vehicles under decentralized architecture:With further consideration on the decentralized characteristics of smart vehicle networks,we design a traffic violation supervision platform based on SIoV for the problem of passive privacy leakage among distributed vehicles.For the necessity and urgency of passive privacy protection among distributed vehicles,we propose a mutually supervision reward and punishment mechanism through the privacy game among smart vehicles,achieving intrinsically driven passive privacy protection.Then,by establishing historical behaviors and reputation values,the degree of required protection for supervision object is determined intelligently.Meanwhile,considering the distributed confidence problem,we design an information dissemination scheme based on directed acyclic graph(DAG),and construct a blockchain-based information sharing system.The information sharing latency and graph structure stability of the system under different traffic environments are analyzed by probabilistic theory.Simulation results show that the proposed algorithm can achieve multi-dimensional passive information privacy protection of distributed intelligent vehicle networks under different vehicle density assumptions.Moreover,the stability of the proposed system can be maintained at low extent of latency through the adjustment of system parameters.