Unsupervised Learning for Cyberattack Detection in Distribution Systems: Leveraging Spatiotemporal Patterns

Modern distribution systems are more vulnerable to cyberattacks due to the usage of communication networks and networked equipment. Labelled datasets, which are necessary for supervised learning approaches to detection, can be difficult to get in the actual world. This study proposes a method for detecting cyberattacks in distribution systems that makes use of unsupervised learning and spatiotemporal pattern recognition. In order to identify malicious events without any labelled data, we develop a framework that combines spatial and temporal data and makes use of clustering and anomaly detection techniques. The technique proved successful in identifying FDI, DoS, and reconnaissance assaults when evaluated on a simulated distribution network. The results highlight the potential for using machine learning techniques to address the issue of unsupervised in distribution systems security.