Document Type : Research Paper
Authors
Computer Department, Faculty of Statistics, Mathematics and Computer Science, Allameh Tabatabaei University, Tehran, Iran
Abstract
Unauthorized electricity consumption and electricity theft are among the most important forms of non-technical losses in electricity distribution networks. They may appear in consumption data as deviations from normal usage patterns, abnormal load variations, or discrepancies between delivered and recorded energy. In addition to imposing economic losses on power utilities, these phenomena affect grid stability, consumption monitoring, and energy policy. This paper presents a structured review of data-driven approaches for consumption anomaly detection in smart power grids, with an emphasis on unauthorized electricity consumption and the Iranian context. Statistical methods, machine learning, deep learning, federated learning, and hybrid approaches based on smart meter data are reviewed and compared in terms of required data, detection capability, scalability, privacy, implementation cost, and practical limitations. The reviewed studies indicate that hybrid approaches, particularly those combining smart meter data with machine learning and deep learning algorithms, have shown greater potential for reducing detection errors and identifying complex patterns of unauthorized consumption. However, their effective use in Iran depends on the development of smart metering infrastructure, access to reliable data, attention to security and privacy, and the design of corrective policies consistent with energy justice.
Keywords
Main Subjects
)