Intelligent Fault Detection and Diagnostic Systems for Electric Smart Meters Using ML and AI Algorithms: A Comprehensive Review

Abstract

Smart meters have become the focus of electricity distribution management as the global deployment of Advanced Metering Infrastructure (AMI) has surpassed 1.7 billion installations worldwide, with projections reaching 3.4 billion units by 2033 [1]. As meter populations scale, the volume and complexity of fault detection, anomaly detection, and diagnostic classification tasks increase proportionally. Threshold-based and rule-based fault detection techniques exhibit well-documented limitations when applied to high-dimensional, non-stationary, and class-imbalanced data streams characteristic of modern AMI networks. This paper presents a systematic and comprehensive review of peer-reviewed literature published between 2020 and 2026 on machine learning (ML) and artificial intelligence (AI)-based fault detection and diagnostic systems for electric smart meters. A structured search of IEEE Xplore, ScienceDirect, Scopus, and Web of Science identified 28 articles satisfying stringent inclusion criteria. The resulting taxonomy classifies approaches into five categories: (1) supervised learning with classical ML classifiers; (2) deep learning models including convolutional neural networks (CNNs) and long short-term memory (LSTM) networks; (3) hybrid CNN-LSTM architectures; (4) unsupervised and semi-supervised anomaly detectors; and (5) federated and privacy-preserving learning frameworks. Systematic comparison of key performance measures—detection accuracy, F1-score, area under the ROC curve (AUC), and false-positive rate (FPR)—reveals that CNN-LSTM hybrid architectures achieve peak detection accuracies of 98.5%, while classical models such as support vector machines (SVM) and extreme gradient boosting (XGBoost) offer consistent performance with lower computational overhead. Critical open challenges—including the absence of standardised benchmark datasets, class imbalance, adversarial robustness, edge deployment constraints, and model interpretability—are identified and discussed. Directions for future explainable, privacy-aware, and computationally efficient fault diagnostic systems are proposed.

Keywords