Prognostics and health management of photovoltaic systems based on deep learning: A state-of-the-art review and future perspectives

As global photovoltaic (PV) power generation capacity rapidly expands, efficient and effective health management of PV systems has emerged as a critical focal point. With the evolution of the Internet of Things (IoT), massive heterogeneous data has been generated in PV systems, enabling the widespread application of deep learning, a powerful data-driven modeling tool, in prognostics and health management (PHM) of PV systems. However, comprehensive reviews specifically focused on deep learning in PV system PHM are scarce. To bridge the gap, core concerns in PV system PHM, including condition monitoring, fault diagnosis, and prognostics, are emphasized. Through a summary of five hundred and six articles published from 2016 to September 2023, an overview of common PV signals, prevalent PV faults, and primary degradation patterns is given. Additionally, an abstract of eight open-source data resources on PV faults is further provided. Significantly, this work compiles cases of the application of deep learning models, led by CNN, in PHM of PV systems, discussing their characteristics and applicability to various types of PV signals. Extra attention has also been paid to the degree of adaptation of these deep models to specific PV PHM tasks. Additionally, this research addresses challenges faced by PV system PHM in the deep learning context, offering guidance for future research. In the future, deep learning will remain indispensable in PV system PHM, and this work aims to provide comprehensive information on deep learning methods, research, and engineering applications to researchers in the field.