Intrusion Detection for Intelligent Connected Vehicles Based on Bidirectional Temporal Convolutional Network

Intelligent Connected Vehicles (ICVs) are increasingly prevalent, with various applications and systems operating in complex network environments. Consequently, detecting and preventing network intrusion is increasingly important. The Controller Area Network (CAN) is presently the predominant network in vehicles, capturing communication among Electronic Control Units. Such data can facilitate the analysis of system anomalies and enhance security measures. In response, a novel intrusion detection framework is proposed, leveraging the additive fusion Bidirectional Temporal Convolutional Network (BiTCN) model. The model treats the CAN message sequence as a natural language sequence, extracting features through bidirectional sliding windows and stacked Temporal Convolutional Network residual blocks. An additive fusion strategy is employed for feature fusion to detect system anomalies more effectively. Experimental results demonstrate that the proposed framework outperforms other models in both detection efficiency and performance. Overall, this framework provides a promising solution for detecting and preventing network intrusion in ICVs.