Research On An Efficient Data Sharing Mechanism Based On Federated Learning And Blockchain Dynamic Partitioning Consensus

In recent years,the Internet of Vehicle(Io V)has emerged with the aim of improving the quality of user experience,driving safety and comfort through the design of complex applications.To achieve these goals,data needs to be transferred/exchanged between the roadside unit(RSU)and the vehicle via various smart devices in a real-time scenario.However,There are still many potential challenges in storage and communication between vehicles and ubiquitous devices roadside,such as data tampering,data integrity,cyber attacks and the risk of information leakage or forgery.In addition,without corresponding constraints,the quality of data provided by vehicles varies and it is not only impractical but also wasteful of vehicle resources to let all vehicles provide the data.Finally,in a complex and dynamic Io V environment,transaction latency must be reduced and communication overheads must be reduced to ensure the timeliness and reliability of data sharing.Federated Learning(FL)is widely used in the Io V context for its ability to protect sensitive and private data,and blockchain ledgers are favoured for their ability to solve distributed trust issues for FL users and to quickly update the status of vehicles.Therefore,this paper combines blockchain and FL technologies to address the issue of how to improve the security and efficiency of data sharing in the Io V scenario,with the following main research elements:(1)In order to solve the problems of imperfect privacy protection,uneven quality of data sources and poor quality of shared models in the process of data sharing in Telematics,a data sharing and privacy protection scheme combining federal learning and blockchain is proposed.First,the multisource vehicle data is modelled by federation learning,and the trained model parameters and the reputation values of the participating vehicles are stored on the blockchain.Then,by analysing the impact of data source quality on the performance of the federated learning algorithm,a reputation value calculation method based on a dual subjective logic model is proposed to select clients for federated learning to ensure efficient screening of data sources in data sharing,improve sharing efficiency and achieve privacy protection.Finally,simulation analysis is conducted and the results show that the designed scheme is able to filter high quality data sources in the scenario of real-time dynamic vehicular data exchange,which in turn improves the accuracy of federation learning training.(2)In order to solve the problems of slow consensus speed and high communication complexity in the process of data sharing in vehicle networks,an optimization scheme of PBFT consensus algorithm based on elastic partitioning protocol is proposed.Firstly,since the traditional PBFT consensus strategy of using randomly selected master nodes to lead the consensus process is not suitable for real-time consensus management in dynamic environments,a master node selection strategy is designed to reduce the probability of malicious nodes becoming master nodes,reduce view switching and improve consensus efficiency.Secondly,a dynamic trust partitioning protocol for the Telematics environment is designed to flexibly deploy the number of slices according to the movement characteristics of nodes to meet the requirements of the Telematics scenario where realtime transaction transmission is required,and to improve the ability,security of node behaviour changes in a malicious environment.Finally,security analysis and simulation analysis are conducted,and the results show that the designed scheme can improve the scalability of PBFT consensus in the Telematics scenario,reduce the communication overhead between nodes through consensus slices,and improve the block transaction efficiency.