Cross-Domain Learning for Power Grid Security: Harnessing Human Physiology for Anomaly Detection

This work presents a novel approach to tackle the problem of real-time detection of cyber intrusions at power electronics dominated grids (PEDG) inverters by establishing an analogy between electrical and medical fields. The methodology leverages the significant similarities between electric grid and human body and posits the feasibility of utilizing the large body of knowledge available in medicine, particularly in the area of disease diagnosis, for the benefit of detecting intrusions occurring to the grid, considering the tools introduced by tremendous advances in artificial intelligence (AI). The theoretical foundation of the analogy is discussed and as a first validation step, the grid inverter voltage is classified into three types of normal, disturbed and anomalous by associating it with three classes of heart rate electrocardiogram (ECG) signals of normal sinus rhythm, mild arrhythmia (supraventricular), and severe arrhythmia (ventricular), respectively. This is done via AI-based time series analyzers which establish a mapping between the two signal categories and enable classifying grid voltage based on ECG using the trained analyzer block. Achieved results demonstrate validity of the approach and promise addressing one of the main challenges in developing intrusion detection systems, that is, data scarcity, by tapping at the enormous knowledge base in the medical field.