Vehicle Network Intrusion Detection Based On Transfer Learning

In recent years,the Internet of Things technology has become a research hotspot,which has led to the development of the field of Internet of Vehicles.Smarter and more complete Internet of Vehicles has gradually replaced the traditional Internet of Vehicles.At the same time,in computer vision,natural language processing,text audio systems,etc.,there are more and more applications of machine learning and deep learning.Combining new scenarios and new environments,machine learning and deep learning can explore new problems.As more and more artificial intelligence technologies are applied to the smart car networking,while enriching the experience,the car networking system is also facing more security issues.Through the research on the current status of intrusion detection for the existing Internet of Vehicles,the existing Intrusion Detection System(IDS)is generally installed in the gateway of each vehicle,so that it can detect the message threat and the CAN inside the vehicle.Cyber ??threats from outside the vehicle.IDS can detect whether the vehicle is under attack by analyzing the messages received by the vehicle,and issue an alarm to normal vehicles in the network after other vehicles have been hijacked.Traditional machine learning is no longer suitable for this scenario,because the traffic characteristics in the Internet of Vehicles are not necessarily independent and distributed.Then how to use the knowledge of transfer learning to design an intrusion detection system suitable for the special environment of the Internet of Vehicles to help the safe driving of vehicles has become New research direction.Through the study of the above issues,the main research work of this article includes:(1)Found two problems in the existing intrusion detection schemes for the Internet of Vehicles.1)The high-speed mobility of vehicles,switching between different networks,and a fixed detection model cannot accurately detect attacks with different characteristics of the underlying traffic on different networks.2)The security function of the Internet of Vehicles network is limited by resources,which affects the speed of security detection.Such as the on-board computer processing performance is not high,the vehicle storage capacity is insufficient and other problems.(2)Aiming at the deficiencies of the existing car networking intrusion detection schemes,we propose a heterogeneous adaptive intrusion detection scheme.For different network(heterogeneous network)data,the heterogeneous migration learning idea is adopted,and the heterogeneous problem is first transformed into a homogeneous problem,that is,the heterogeneous network data sets of different feature spaces are mapped to the common subspace through the designed data transformation matrix Form a new network data set,and then use the new network data set to train the classification model.The final model can be used for intrusion detection in different networks.Experiments show that the IDS scheme proposed in this paper is superior to other available IDS in its accuracy,stability,and processing speed.(3)Aiming at the characteristics of limited computing capacity and small storage capacity of connected car terminal equipment.Research transfer learning and deep learning models,and propose a vehicle network intrusion detection scheme based on deep transfer learning.In the model building stage,the network will generate a large amount of network data,and the resource constraints of the vehicle cannot train the model on the vehicle side.We can transfer the data to the cloud,perform data preprocessing in the cloud,and use different network data sets for training.Draw out the corresponding neural network model.Next,we consider how many parameters in one network model can be transferred to another network model.In this way,the new network model after fusing different neural network models only needs to change the parameters Distributed to the vehicle terminal can be detected,without increasing the hardware capabilities of the vehicle,it also improves the classification capabilities of the vehicle network intrusion detection model.