Dual-Branch Graph Learning with Frequency Gating for Industrial Sensor Anomaly and Cyberattack Detection

Area of Science:

• Industrial IoT
• Cybersecurity
• Machine Learning

Background:

• Industrial sensor systems face increasing threats from physical anomalies and cyberattacks.
• Sensor data exhibits complex periodic, non-stationary characteristics, and dynamic spatial dependencies.

Purpose of the Study:

• To propose a dual-branch graph learning framework for simultaneous anomaly and cyberattack detection in industrial sensors.
• To address the challenges posed by complex time series data in industrial environments.

Main Methods:

• A dual-branch graph learning framework with frequency gating is proposed.
• Time series data is decomposed into periodic and non-stationary components.
• Periodic GIN (P-GIN) and Non-stationary GIN (NS-GIN) model spatial dependencies separately.
• Frequency gating enhances feature representation for abnormal and attack-related data.
• Hierarchical temporal encoding captures local and long-range temporal dependencies.

Main Results:

• The proposed framework achieves superior performance in both anomaly and cyberattack detection.
• Experimental results on real-world industrial sensor datasets validate the method's effectiveness.
• The approach outperforms existing state-of-the-art methods.

Conclusions:

• The dual-branch graph learning framework offers a robust solution for industrial sensor security.
• The method effectively handles complex data characteristics for improved detection.
• This research contributes to enhancing the security and reliability of industrial sensor systems.