Design And Implementation Of Vehicle Network Intrusion Detection System For Unbalanced Samples

With the rapid development of automobile industry and Internet industry,the Internet of vehicles has been more and more closely combined with human modern life.By invading the Internet of vehicles system,hackers can interfere with the normal operation of vehicles,directly threaten the safety of users’ lives and property,and cause great harm to the society.Therefore,the CAN bus intrusion detection technology for the Internet of vehicles has important application value in the field of Internet of vehicles security and people’s life and property security.Based on the analysis of the existing models and aiming at the shortcomings,the current more advanced algorithms are used to fuse and improve them respectively,and the main technology of CAN bus intrusion detection based on deep learning is realized.The main research contents of each part are as follows:Firstly,research on lightweight can bus intrusion detection method for known attacks.The existing deep learning methods have the contradiction between accuracy and real-time.In view of the lack of real-time due to the heavy application of the current deep learning model in the on-board network can bus intrusion detection system,the adaptive optimization and real-time simplification of the existing concept RESNET model can process 3492 can messages per second,which is higher than the sending rate of CAN bus data packets,so it can meet the needs of real-time detection.Then,the experiment is compared with other advanced methods to prove that the method is also effective.Second,research on small sample can bus intrusion detection method for unknown attack.Aiming at the problem that it is difficult to obtain a large amount of data to support the training of deep learning model,a migration learning intrusion detection model(CANTransfer + H)integrating heuristics is designed in the direction of small sample detection.Firstly,the migration learning model is trained through known attack samples,and the CAN bus intrusion is detected in combination with heuristic rules,which can improve the detection ability of the model under the condition of limited known samples.Compared with other transfer learning models and other in-depth learning models integrating heuristic rules,it is proved that the performance of this method is improved.Thirdly,the research on the technology of multi can bus intrusion data generation for unbalanced samples.Thirdly,taking the lack of data samples as the starting point,taking the diversity expansion of data samples as the research direction,and further aiming at the problem of unstable data generated by the existing Gan model,a diversity data generation model integrating evolutionary computing theory and multiple generation countermeasure networks is proposed,and the detection effects of the first two models are verified by generating data,The performance of the model is tested and verified on the existing public data set by using is and FID.Finally,based on the above models and algorithms,a CAN bus intrusion detection system based on deep learning is designed and implemented.The system can detect the intrusion behavior in the CAN bus of the Internet of vehicles in real time under the condition of insufficient data samples,and can also generate diversified can bus intrusion data for model learning and comparison of unknown attacks by relevant practitioners.