Intrusion Detection Study Of Pure Electric Vehicle On-board Network Security

With the development of pure electric vehicle intelligence and interconnection,Controller Area Network(CAN)network is widely used in automotive electronic control system.However,CAN networks have many problems in communication security,such as information leakage,denial-of-service attacks and remote execution code attacks,which pose a great threat to the automotive industry and the lives and properties of vehicle owners.Therefore,the study of CAN network security technology detection has become increasingly important.In recent years,with the rapid development of deep learning technology,its wide application in the fields of computer vision,natural language processing and speech recognition has been proven.Deep learning has also shown great potential in the detection of CAN network security technologies.Deep learning techniques enable automatic feature extraction and learning of CAN network data,thus allowing network security attack detection and classification without human intervention.This paper analyzes the CAN bus as a reference point,and starts from three aspects:the basic concept of CAN bus,message format,communication mechanism and parsing method.Summarized a number of CAN bus vulnerabilities,hackers for the vulnerability as an attack portal and the implementation of eavesdropping,replay,injection,tampering and the most serious and common Dos attacks and other intrusion methods.In this paper,the collected data set is processed based on the READ algorithm,and the collected attack messages are first classified into four categories: Dos attack,fuzzy attack,simulated attack,and no attack,and the Dos attack,which has the highest hazard and is very likely to occur,is detected and identified.Due to the Dos attack characteristics,the processed data is out of balance and thus SMOTE technique is used to prevent the overfitting phenomenon.In this paper,the data after processing is detected by a classification algorithm that combines three classical classification algorithms to identify normal data messages with those of Dos attacks.From the subsequent simulation experiments,all three classification algorithms have good detection accuracy,which are above 90%,among which the random forest classification algorithm and XGBoost classification algorithm are optimal.However,the detection rate of the traditional machine algorithm is still low,so the LSTM algorithm model is designed based on deep learning after continuous research,and the detection rate of LSTM algorithm reaches 97% after verification by experimental algorithm.Therefore,the purpose of this paper is to explore the CAN network security technology detection based on deep learning,to build an efficient and accurate network security detection model through feature extraction and learning of CAN network data,and to improve the security performance and reliability of CAN networks.The research results of this thesis will provide an effective CAN network security technology solution for the automotive industry and make some contribution to the development of the automotive network security field.