Smart Fault Detection, Classification, and Localization in Distribution Networks: AI-Driven Approaches and Emerging Technologies

Distribution networks play a vital role in bridging transmission systems and end users, offering enhanced flexibility, decentralization, and the capacity to integrate distributed generation. However, with nations worldwide actively pursuing carbon neutrality and emission peak goals, sustainable energy sources such as solar and wind are increasingly penetrating distribution networks, posing significant challenges to conventional fault detection, classification, and localization techniques due to bidirectional power flows, dynamic fault currents, and rising network complexity. These challenges manifest as reduced sensitivity of protection systems in distribution networks, increased difficulty in identifying high impedance faults, and frequent misclassification or mislocation of faults under dynamic network conditions. To address these limitations, this paper presents a comprehensive review of artificial intelligence-driven approaches and emerging technologies that are specifically tailored for fault analysis in distribution networks, to enhance diagnostic accuracy, adaptability, and real-time decision-making efficiency. Following the chronological development of artificial intelligence, the review systematically investigates smart fault detection methods applied to fault scenarios in distribution networks, with a particular emphasis on presenting fault type classification and fault localization separately to facilitate a logically structured understanding. In addition, common types of distribution network faults are examined, and the impact of distributed generation on fault behavior, electrical characteristics, and protection coordination is critically assessed. The review further distinguishes between artificial intelligence-based smart approaches that directly process raw distribution networks signal data and those that rely on advanced feature extraction techniques to enhance functional performance. This review also explores the emerging potential of large language models to enhance the explainability of diagnostics, support multi-agent coordination, and enable natural language-based fault reasoning. The insights offered herein are expected to provide practical guidance for engineers and researchers for selecting and deploying intelligent fault diagnosis strategies in future distribution networks with high distributed generation penetration.