Cyber Intrusion Detection and Sensor Failure Diagnosis using Statistical Hypothesis-Based Correlation: Application to FACTS AC Filter Banks

This paper introduces a statistical hypothesis-based correlation coefficient method using harmonics for diagnosing sensor and equipment failures in a cyber-physical system of FACTS devices. The diagnosis assists in determining root cause of failures such as caused by cyber intrusions, sensing device malfunction, equipment degradation, or communication delays or failures. We applied the method to detect anomalies in the AC filter circuits of FACTS devices. We extracted and decomposed the harmonic information from a central measuring device and correlated it with each harmonic filter circuit measurement. A weak cross-correlation indicates a mismatch between these harmonic signals, the pattern of which could be used to diagnose the type of failures, such as harmonic filter degradation, communication delays, and intrusion. The key innovation is based on the assertion that the harmonic information can be observed at both harmonic filter and at the power electronic device that generates the harmonics, and that the signatures of these observed harmonics is comparable or identical under normal conditions. Numerical results show that the method can detect signal manipulation due to cyber intrusion in the FACTS communication network.