Research Of Fedrated Learning In Data Sharing And Privacy Protection Technology Od Internet Of Vehicles

The transmission and sharing of massive data between various nodes in the Internet of Vehicles makes ensuring the efficient and secure data transmission and the information privacy of vehicle users a problem that cannot be ignored.To solve these problems,part of the existing research on the Internet of vehicles is based on data encryption verification or trust assessment of vehicle users,and the other part is based on model training of machine learning.The process of encryption,decryption and evaluation often involves a lot of computing,while the vehicle nodes on the edge often have limited resources.Therefore,it is difficult for the vehicle nodes to effectively calculate,analyze and share data while considering data privacy.Machine learning can extract key information from data and share it in the form of model,but centralized processing of data will lead to high cost of single node resources and risks of privacy disclosure of vehicle users.When attackers use normal vehicle users to join the Internet of Vehicles and launch attacks,the validity and security of data will be threatened.When faced with a dishonest central server,there is a risk of leakage of vehicle user information privacy.In order to solve these defects,this thesis is devoted to studying how to combine federated learning technology and secret sharing technology to design the data sharing architecture of the Internet of Vehicles,so as to improve the efficiency of data sharing of the Internet of Vehicles and ensure the data security and information privacy of vehicle users.The main research contents of this thesis are as follows:（1）In the current Internet of vehicles,it is difficult for vehicle nodes to effectively calculate,analyze and share data while protecting data privacy.this thesis proposes a data sharing architecture for the Internet of Vehicles based on federated learning,including the vehicle user layer,the edge computing layer and the center layer.The scheme refers to the characteristics of the network and each entity in the Internet of Vehicles,and designs a training participation strategy in the federated learning architecture for each layer of entities to ensure the reliability of data transmission and sharing in the Internet of Vehicles.The vehicle user is responsible for the local training of the model,and the edge server is responsible for the transfer and aggregation of the model,designing a data set distribution scheme,and enhancing the performance of the vehicle local training model.Design a weighted allocation scheme for aggregation to improve the convergence speed and accuracy of the global model.At the central server layer,a global aggregation scheme is designed to accelerate the synchronization of the global model of each node.The final model trained is used as the carrier of data information transmission to reduce the information redundancy of the original data and enhance the application efficiency of the key information of the receiving node.（2）After the introduction of federated learning technology into the Internet of Vehicles,in the scenario where the central server is dishonest and malicious users participate and launch attacks,data security and vehicle user privacy protection issues,this thesis proposes a secret sharing-based attack mitigation and Privacy Protection Program.In this scheme,for vehicle users,privacy protection training is designed to strengthen the privacy protection of vehicle user information.For the central server,design a mitigation plan for poisoning attacks,detect and filter out malicious attackers,and improve data security and accuracy.In addition,the computational cost of the detection algorithm in the poisoning attack mitigation scheme is analyzed,and an optimization algorithm is proposed to further reduce the computational cost of the system.