A COMPUTATIONAL FRAMEWORK FOR SMART GRID TAMPERING DETECTION USING UNSUPERVISED MACHINE LEARNING

Abstract

This paper presents a computational framework for detecting tampering in smart grids using unsupervised machine learning algorithms. Leveraging Advanced Metering Infrastructure (AMI) and integrating algorithms such as K-Nearest Neighbourhood with Isolation Forest (KNN + IF), Local Outlier Factor (LOF), and Lightweight Online Detector of Anomaly (LODA), the framework achieves robust tampering detection. Experimental evaluations on a smart grid IoT Coordinating Infrastructure (SG_IoT_CI) showed that the proposed system significantly improved tampering detection accuracy, achieving up to 31.11% accuracy with KNN + IF compared to lower-performing methods. Additionally, the framework demonstrated enhanced service rates, with KNN + IF yielding a 35.09% improvement and LODA achieving the highest throughput at 28.12%. Latency was reduced by 11.98% for KNN + IF and 8.98% for LODA. These results underscore the efficiency of the proposed approach in addressing zero-day attacks and reducing false positives, paving the way for scalable and secure smart grid ecosystems.