Real-Time Intrusion Detection Based on Decision Fusion in Industrial Control Systems
Yawen Xue,Jie Pan,Yangyang Geng,Zeyu Yang,Mengxiang Liu,Ruilong Deng
DOI: https://doi.org/10.1109/ticps.2024.3406505
2024-01-01
Abstract:Industrial control systems (ICSs) are becoming increasingly interconnected as the rapid convergence of information technology (IT) and operation technology (OT) networks, and meanwhile massive attack surfaces have been exposed. However, traditional intrusion detection systems (IDSs) are difficult to be directly deployed in ICSs due to the hard real-time requirement and rare patching chance. Besides, the design of effective and practical IDSs is hampered by the lack of benchmarking ICS cybersecurity datasets. To bridge the gaps, this paper makes the first attempt by open-sourcing the developed ICS cybersecurity datasets and proposing a decision fusion based real-time IDS. Firstly, we design a customized cybersecurity dataset in a full-hardware and high-fidelity platform, including 7 types of cyber threats tailored for ICSs. The collected dataset includes network traffic, sensor readings, actuator status, and system parameters, providing the state-of-the-art benchmark dataset for ICSs consisting of cross-layer characteristics. Furthermore, we design an online decision fusion-based IDS by strategically integrating 4 widely-used machine learning models. The proposed IDS is deployed on a real-time running ethanol distillation, surpassing the performance of single detection models in terms of precision and F1-score, which substantially enhances intrusion detection accuracy and cybersecurity of ICS.