Data-Driven Two-Stage Fault Detection and Diagnosis Method for Photovoltaic Power Generation

Detection of abnormal photovoltaic (PV) system operation is essential to ensure safe and uninterrupted performance. In this study, the authors present a data-driven two-stage method for PV fault detection and diagnosis (FDD). We exploit an inherent characteristic of PV systems, i.e., voltage and current changes at maximum power point (MPP) caused by faults. In the first stage, fault occurrences are detected using predefined criteria based on the MPP values. The second stage employs ${I}$ – ${V}$ characteristic curve data and a densely connected convolutional network (DenseNet) model to diagnose the fault type. The DenseNet model is rigorously trained using a very large dataset of ${I}$ – ${V}$ curves; this ensures precise and efficient fault diagnosis. We validate our approach via simulations and hardware analyses employing a $5\times3$ PV array that initially operates normally, but then develops line-to-line faults (LLFs), open-circuit faults (OCFs), degradation faults (DFs), and partial shading faults (PSFs). We compare our DenseNet-based PV FDD model to the latest PV FDD models. The results confirmed that the new method accurately detect and diagnose PV faults.