Research On Reliable Communication Technology Of Vehicle Network Based On Machine Learning

The vehicle network supports vehicles to communicate with other network equipment,which can provide relevant driving information,traffic conditions,and multimedia services for vehicles,reduce the probability of traffic accidents,and improve the driving experience.This paper focuses on the reliable transmission of security alarm information between vehicles.The specific research results and contributions are summarized as follows:Compared with ordinary users,the vehicle communication environment is complex and variable,and the communication reliability between vehicles is difficult to guarantee.However,the vehicle to the workshop usually transmits safety warning information,which provides an important reference for safe driving,so the communication delay between vehicles is strict and the reliability requirements are high.Experiments show that in the urban traffic environment,the large probability of loss of data transmission between vehicles is caused by non-line-of-sight factors such as occlusion.In the case where the link between vehicles is non-line-of-sight,the communication between the vehicle and the vehicle requires the auxiliary forwarding of other vehicles or infrastructure to ensure the reliability of communication.Therefore,it is especially important to accurately identify non-line-of-sight links.Based on the convolutional neural network algorithm,this paper proposes to use channel state information,vehicle position information,and signal reception strength indication as feature vectors to determine the line of sight and non-line of sight of the link.In the intersection environment,the signal characteristics of real line-of-sight and non-line-of-sight links are collected as the basis to verify the accuracy of the proposed link non-line-of-sight recognition algorithm.At the same time,the link identification mechanism is used as the reference basis for signal transmission between vehicles.Based on NS3,the vehicle network system simulation platform is built to verify the reliability of non-line-of-sight link recognition for inter-vehicle communication.Secondly,this paper studies the problem of vehicle node abnormal node behavior detection.The car network environment is complex,and the abnormal behavior of nodes may be affected by many aspects,such as network attacks and node failures.This paper defines the abnormal behavior of nodes,which mainly includes physical anomalies and network anomalies.Physical anomalies are manifested in the fact that vehicle nodes are not able to drive normally due to external environmental factors.Network anomalies are manifested in DOS attacks or black holes in vehicle nodes.Attacks cannot communicate normally and reliably.These abnormal behaviors cause some security alarm information to be reliably transmitted to the vehicle nodes,thereby threatening network communication security and vehicle driving safety.In order to eliminate the impact of abnormal nodes on network reliability,this paper proposes an abnormal node detection mechanism.Considering that there may be unknown types of electronic attacks and types of faults in the environment,this paper needs to use unsupervised machine learning algorithms.The variable-divided self-encoder is an unsupervised learning algorithm that does not need to acquire more system prior information in advance,and automatically learns behavior information in a normal environment.This paper uses a variational auto-encoder to detect abnormal nodes.Considering the limited number of training data samples in the real environment,this paper manually generates the number of training data by means of the variational auto-encoder.In addition,when the formation passes through the central node coverage area,in order to reduce the system occupied bandwidth,this paper uses the joint average algorithm to achieve the calibration of the detection model,which improves the detection accuracy.