Photovoltaic (PV) systems are subject to operational failures due to aging and external environmental conditions. These faults may affect PV modules, connection lines, and converters/inverters, resulting in reduced efficiency, performance deterioration, and potential system instability. Effective fault detection is therefore essential to ensure operational safety, maximize energy yield, and minimize maintenance costs. Deep learning (DL) has become a powerful paradigm for PV fault detection due to its ability to automatically learn discriminative representations from large-scale and heterogeneous data sources. This paper presents a comprehensive review of deep learning–based approaches for fault detection in photovoltaic systems. Major DL architectures are examined, including convolutional neural networks, recurrent neural networks, transformer-based models, autoencoder-based methods, energy-based models, and transfer-learning frameworks. Recent studies are analyzed to highlight their strengths, limitations, and practical applicability in real-world PV monitoring scenarios. Key challenges and open research directions are also discussed, including data scarcity and imbalance, cross-site generalization, real-time deployment constraints, and the integration of physics-informed and self-supervised learning strategies. This review provides researchers and practitioners with a structured overview of current advances and practical guidance for designing robust, intelligent PV fault-detection systems.
Harrou et al. (Mon,) studied this question.