Multi-Attack Intrusion Detection for In-Vehicle CAN-FD Messages.

Area of Science:

• Cybersecurity
• Automotive Networking
• Intrusion Detection Systems

Background:

• Controller Area Network with Flexible Data (CAN-FD) lacks robust information security, making it vulnerable to attacks.
• Existing anomaly detection methods for CAN-FD require improved accuracy to effectively prevent cyber threats.

Purpose of the Study:

• To propose a novel, two-sub-model intrusion detection system for CAN-FD bus security.
• To enhance the accuracy and classification capabilities of anomaly detection in CAN-FD networks.

Main Methods:

• Developed a two-stage model: Anomaly Data Detection Model (ADDM) using Long Short-Term Memory (LSTM) for temporal pattern recognition.
• Integrated an attention mechanism into the Anomaly Classification Detection Model (ACDM) to refine sequence-based relationship identification.
• Evaluated the model on real-vehicle and established CAN-FD intrusion datasets.

Main Results:

• The proposed model demonstrated superior performance in detecting anomalies on CAN-FD data.
• Achieved accuracy improvements of 1.44% over LSTM-based methods and 1.01% over CNN-LSTM-based methods.
• Showcased broader applicability and more refined multi-attack classification capabilities.

Conclusions:

• The novel intrusion detection model significantly enhances the security of CAN-FD networks.
• The combination of LSTM and attention mechanisms offers a promising approach for advanced cyber threat detection in automotive systems.