A Node Trust Evaluation Method Of Vehicle-road-cloud Collaborative System Based On Federated Learning

As the vehicle-road-cloud collaboration system develops rapidly,it is accompanied by serious information security problems while solving the data transmission issues.For constructing secure transmission of data,trust is recommended as a relevant way to accomplish network security;that is,developing a trust model that can be used by sensor nodes to determine the reliability of another node is crucial.However,on the one hand,the heterogeneity of the network has different functional requirements for trust evaluation,and the openness of the network makes nodes more vulnerable to attacks.The existing trust evaluation cannot be directly transferred to the Vehicle-road cloud collaborative system,so the research of trust evaluation model in the vehicle-road cloud collaborative system has more challenges than the traditional network.On the other hand,the openness of the cooperative system makes the nodes more vulnerable to security threats such as calls,instruction eavesdropping,tampering and replay by malicious nodes.Moreover,many nodes in the collaborative system are deployed in an unguarded environment,and the nodes are powered by dry batteries.There is no fixed power supply and resources are limited,and the status of nodes is dynamic.Even under strong security protection,any node in the vehicle road cloud collaborative system may produce unreliable information.For example,due to hardware or software defects,node resources in the vehicle road cloud collaborative system are limited during massive data interaction.When data is transferred between nodes,traditional encryption technology cannot guarantee node energy consumption,and nodes are vulnerable to attacks by malicious nodes.Aiming at the two major problems of network performance and information security existing in the vehicle-road cloud collaborative system,this paper comprehensively analyzes and summarizes the differences and difficulties existing in the existing solutions to the above problems,aiming at studying an effective and reliable trust evaluation scheme of the vehicleroad cloud collaborative system,which can effectively prevent the spread of malicious information while improving the network performance.To further improve the effective use of reliable information.Specifically,the main research work of this paper can be summarized as follows:(1)Based on the characteristics and architecture of car-road cloud cooperative system,a layered trust management architecture is proposed to effectively solve the problems of data collection,storage and management in the process of data transmission and interaction in the network.The proposed framework is analyzed based on trust factors and divided into different levels of management,including data trust management,network trust management and node trust management.Each level of management is responsible for managing and maintaining corresponding on-board data,and the framework clearly defines the internal connection and mutual support among all levels.(2)A new hierarchical trust evaluation model is proposed to solve the information security problem of vehicle-road cloud cooperative system.The model takes into account the influence of each node in the vehicle-road cloud collaboration system.Based on the characteristics of nodes and networks,each node’s trust value will be considered according to different weights in the evaluation process.Complex models are simplified into ordered hierarchies and evaluated through different levels of trust indicators.The model is divided into multiple aspects of data transmission forwarding amount and success rate,network reliability,real-time performance,node behavior and state,and the total trust value of the node is finally obtained.Compared with the traditional trust evaluation mechanism,hierarchical trust evaluation mechanism is applicable to more different scenarios and the evaluation of trust degree is more comprehensive.(3)This paper proposes a federated learning algorithm based on trust evaluation,and innovatively introduces personalized federated learning into heterogeneous networks when the trust evaluation scheme is effectively deployed and the trust values of the next hop forwarding node and neighboring nodes are considered in the routing process.It solves the problem of limited resources of edge nodes caused by node heterogeneity,reduces the cost of trust evaluation,and carries out personalized processing at the device,data and model levels.In order to reduce heterogeneity and obtain high quality personalized model for each device,the trust value calculation can be updated adaptively with the real-time changes of the network topology.