Federated Learning Oriented Security And Resource Allocation For Vehicle Network

Internet of vehicles(Io V)is a classic application of Internet of things technology in the field of intelligent transportation.Among them,Io V realizes the information exchange between vehicles and infrastructures through wireless network communication technology,and provides the foundation for traffic management,driving safety and entertainment information services.However,due to the cloud computing centric network architecture adopted by the traditional Io V,the user data in Io V business needs to be transmitted in the core network,which may bring the problems of user data leakage and a lot of energy consumption,seriously affecting the user experience.In this thesis,the data communication problem of automatic driving decision model in the training process is studied,and the data security and energy consumption are focused.The main work of this thesis is as follows:1.Aiming at the problems of data security and resource consumption in the centralized network architecture of Io V,combined with the unique distributed computing characteristics of federated learning,a training framework of autonomous driving decision model based on federated learning is proposed,and the key to solve the problems of data security and resource consumption under this framework is analyzed.2.Aiming at the threat of reconstruction attacks that the federated learning training framework may face in the process of parameter upload,resulting in the leakage of user data,a federated learning training scheme based on k anonymity is designed.Firstly,a machine learning algorithm is used to build a decision-making model for autonomous vehicles,and then after the model is trained locally,the k-anonymity algorithm is introduced to anonymize model parameters to enhance data security;for the problem that the amount of model parameter data is small,the k-anonymity algorithm cannot be applied,propose an improved k anonymous algorithm.Finally,simulation experiments verify the effectiveness of the k-anonymous federated learning training scheme proposed in this thesis,and test the performance of this scheme through comparative experiments.3.In the framework of federated learning and training,edge computing technology is introduced to solve the problem of resource shortage when vehicle nodes are trained locally.Aiming at the mobility of vehicle nodes and the heterogeneity of edge node resources,a dynamic resource allocation strategy based on the state prediction is designed.First,the service time is divided into continuous time intervals,the optimal resource allocation problem is transformed into a convex optimization problem in each time interval,and the Lagrange multiplier algorithm and KKT conditions(Karush–Kuhn–Tucker conditions)are used to solve the problem.Considering the additional energy consumption caused by changes in the state of edge nodes,a linear regression algorithm is used to predict resource requests,determine the state of edge nodes,reduce the number of state changes,and reduce the additional energy consumption of the system.Finally,simulation experiments are compared to verify the effectiveness and performance of the dynamic resource allocation strategy based on state prediction proposed in this thesis.