Telling Apart: ML Framework Towards Cyber Attack and Fault Differentiation in Microgrids

Microgrids have revolutionized the field of power systems as they provide higher reliability and power quality over traditional power systems by possessing the ability to self-supply, distribute, and regulate electrical flow to local customers. However, the presence of anomalies in microgrid data can lead to inefficient performance and can put system security at risk. Unfortunately, discerning between malicious anomalies (MCAs) and malfunction anomalies (MFAs) is an open research area that needs attention. In this paper, we propose a multi-staged machine learning (ML) framework geared toward discerning between MFAs and MCAs in microgrids using Explainable AI. Performance evaluation yields that $N=3$ provides us with optimal performance with high $P_{\mu}, R_{\mu}$, and $F_{\mu}$ scores of 0.99, 0.99, and 0.99, respectively. This means that, by using this framework, the system can differentiate effectively between the anomaly types using a very low number of influential features. The proposed framework can help the system autonomously and effectively discern between MFAs and MCAs leading to more effective responses.