Federated Learning Solutions for Energy Theft Detection: A Focused Survey

Federated Learning (FL) offers transformative potential for Smart Grids (SGs), particularly in Energy Theft (ET) detection, a critical challenge threatening global power grid stability and sustainability. FL enables decentralized model training across distributed grids, preserving user data privacy while identifying fraudulent consumption patterns. Despite growing interest in FL for SGs, existing literature lacks a comprehensive survey on its specific application to ET detection. This paper bridges this gap by systematically reviewing FL-based ET detection frameworks, evaluating datasets, methodologies, and performance metrics. Our analysis highlights FL’s ability to enhance model generalizability across diverse energy networks without centralized data sharing, addressing privacy concerns inherent in traditional approaches while promoting sustainable grid management. Key findings demonstrate FL’s efficacy in improving detection accuracy and reliability, though challenges like communication overhead and model heterogeneity persist. By synthesizing current advancements and limitations, this paper provides insights for future research directions, emphasizing scalable, adaptive FL solutions to secure smart grid infrastructures against evolving ET threats.