Measurement data intrusion detection in industrial control systems based on unsupervised learning

Abstract

Anomaly detection strategies in industrial control systems mainly investigate the transmitting network traffic called network intrusion detection system. However, The measurement intrusion detection system inspects the sensors data integrated into the supervisory control and data acquisition center to find any abnormal behavior. An approach to detect anomalies in the measurement data is training supervised learning models that can learn to classify normal and abnormal data. But, a labeled dataset consisting of abnormal behavior, such as attacks, or malfunctions is extremely hard to achieve. Therefore, the unsupervised learning strategy that does not require labeled data for being trained can be helpful to tackle this problem. This study evaluates the performance of unsupervised learning strategies in anomaly detection using measurement data in control systems. The most accurate algorithms are selected to train unsupervised learning models, and the results show an accuracy of 98% in stealthy attack detection.