融合BERT与迁移学习的车载CAN网络自适应入侵检测
Integrated BERT and Transfer Learning for Adaptive Intrusion Detection in Vehicle CAN Network
李向荣 1张运胜2
作者信息
- 1. 北京工业大学,北京 100102
- 2. 湖北经济学院,湖北物流发展研究中心,武汉 430205
- 折叠
摘要
为了提高分布式车载电子控制单元(ECU)的通信安全,提出了基于BERT和迁移学习的无监督自适应入侵检测系统.利用自然语言模型BERT学习控制器局域网(CAN)总线中标识符序列特征,将掩码语言模型作为训练目标,进行无监督异常检测,并结合迁移学习完成不同车型的数据训练.在大规模CAN数据集上的试验结果表明:该检测系统对拒绝服务(DoS)、模糊和欺骗攻击的检测精度分别达到100%、99.99%和99.97%,能够通过模型微调实现对新车型的自适应检测.
Abstract
In order to improve the communication security of distributed on-board Electronic Control Unit(ECU),an unsupervised adaptive intrusion detection system based on BERT and transfer learning is proposed.The natural language model BERT is used to learn the identifier sequence features in the Controller Area Network(CAN)bus,and the mask language model is used as the training target for unsupervised anomaly detection.Combined with transfer learning,the data training of different vehicle models is completed.Experimental results on large-scale CAN datasets show that the detection accuracy of the detection system for Denial of Service(DoS),fuzzy and spoofing attacks reaches 100%,99.99%and 99.97%respectively,and the adaptive detection of new vehicle models can be realized through model fine-tuning.
关键词
车载网络/控制器局域网/车载电控单元/入侵检测系统/迁移学习Key words
Vehicle network/Controller Area Network(CAN)/Electronic Control Unit(ECU)/Intrusion detection system/Transfer learning引用本文复制引用
出版年
2024
NETL
NSTL
万方数据