Anomaly Detection in Operational Technology Systems Using Non-Intrusive Load Monitoring Based on Supervised Learning

When using Operational Technology (OT) across critical infrastructures, including Naval installations, it is important to implement robust anomaly detection mechanisms. Using non-intrusive load monitoring (NILM) techniques and methods, this paper leverages multiple supervised machine learning models to identify faults and anomalies in operational technology power systems. We build a novel small-scale OT and sensor system to generate a NILM data set into which we introduce four distinct single point faults and anomalies. We use three supervised ML algorithms (CNN, SVM and LSTM) to facilitate detection and classification of these various anomalies. Through vigorous evaluation, our results show the potential of integrating NILM technology with ML to enhance OT security and resiliency. This research not only fills a critical gap in OT anomaly detection but also sets a precedent for future advancements in the field.