An Empirical Evaluation of Anomaly Detection Techniques in Smart Grid Systems Using Real-Time Operational Data

S M Shakil; Alamgir Hossain; Md Mijanur Rahman

doi:10.63125/2xcry064

Authors

S M Shakil Department of Electrical and Computer Engineering; Florida Polytechnic University, FL, USA Author
Alamgir Hossain School of Engineering; University of Tasmania, Australia Author
Md Mijanur Rahman Department of Electrical and Electronic Engineering; Dhaka University of Engineering and Technology, Bangladesh Author

DOI:

https://doi.org/10.63125/2xcry064

Keywords:

Smart grid, Anomaly detection, Explainability, False alarm burden, Technology Organization Environment (TOE)

Abstract

This study addresses the problem that smart grid operators increasingly rely on enterprise and cloud enabled monitoring platforms, yet anomaly detection outputs are often difficult to trust due to data quality variability, limited alarm explainability, weak workflow integration, and false alarm overload. The purpose was to quantify which technical and organizational factors most strongly predict perceived detection effectiveness and adoption readiness for anomaly detection techniques in a real smart grid case. A quantitative cross sectional, case-based design was applied using a structured 5-point Likert survey administered to N = 182 stakeholders drawn from enterprise smart grid environments spanning AMI and smart metering, SCADA and EMS or DMS, PMU or WAMS, and supporting IT and cybersecurity functions (operations 28.6%, metering 20.3%, data and IT 20.9%, protection 15.9%, cybersecurity 14.3%). Key variables included Data Quality Adequacy (DQ), Robustness and Adaptability (RB), Explainability of Alarms (EX), System Integration Capability (SI), Perceived Detection Effectiveness (PDE), Adoption Readiness and Trust (ART), and False Alarm Burden and Operational Impact Index (FABOI). The analysis plan used descriptive statistics, scale reliability (Cronbach alpha .81 to .88), Pearson correlations, and multiple regression models. Headline findings showed moderately high perceptions of DQ (M = 3.92, SD = 0.61) and PDE (M = 3.81, SD = 0.62), while false alarm burden was nontrivial (FABOI M = 3.41, SD = 0.73). PDE correlated with DQ (r = .52) and RB (r = .48), while ART correlated most strongly with EX (r = .54) and PDE (r = .58), and decreased with FABOI (r = −.49), all p < .01. Regression confirmed PDE was predicted by DQ (β = .31, p < .001) and RB (β = .27, p < .001), R² = .39, whereas ART was driven by EX (β = .33, p < .001), PDE (β = .29, p < .001), SI (β = .21, p = .002) and reduced by FABOI (β = −.24, p < .001), R² = .47. Implications suggest utilities should prioritize data governance and robustness for effectiveness, but invest in explainable alarms, workflow integration, and alarm burden management to increase operational trust and sustained adoption.