An I-V-Curves based deep learning framework for rapid fault diagnosis of curved building integrated photovoltaic system

Curved photovoltaic (PV) systems hold significant potential for use in buildings with complex forms, while their curved geometry and varied environments increase fault risks. I–V curve analysis is a widely used method for rapid fault diagnosis in PV systems. However, the I–V characteristics of curved PV systems under normal operating conditions often resemble those under fault conditions, making accurate fault diagnosis particularly challenging. Therefore, this study proposes an intelligent fault diagnosis framework for curved building-integrated photovoltaic (BIPV) systems based on I–V curve analysis. To validate the effectiveness of the proposed framework, a curved PV system with a curvature of 120° was employed as a case study in this research, with consideration of two common electrical topologies of systems. The I–V characteristics of the system under normal operation and 11 fault conditions were then computed for each topology using an experimentally validated simulation model and meteorological data from a typical day in Xiamen. For each electrical topology, 53,388 I–V curve samples were used for training and 22,872 for testing, with the data processed using deep learning models for training and evaluation. The results demonstrate that the proposed framework achieves accurate classification of normal and 11 fault conditions across both topologies, with an average accuracy exceeding 99.83%. This study provides new technical approaches and useful insights for fault diagnosis and safe operation of curved BIPV systems.