Fault Diagnosis of Photovoltaic Arrays using Inverter Data and Classification Models

The expanding deployment of photovoltaic (PV) systems is increasing the demand for diagnostic technologies that accurately assess plant performance and identify faults.In Korea, the limited availability of areas for installing these technologies has led to the growing adoption of PV systems that vary with site, where access for operation and maintenance (O&M) personnel is often restricted.Data-driven and AI-based diagnostic technologies have been studied, but most research has focused on predicting power output.The accuracy of fault diagnosis can be increased by inferring voltage and current characteristics.In this study, a method for diagnosing the faults of DC PV arrays was developed using inverter-collected data and classification models.Normal and faulty conditions were discriminated by estimating the normal operating state using a regression model based on a previously developed mathematical model.The classification model was trained using a dataset consisting of approximately 1,800 I -V curves at their maximum power points and additional data obtained from simulated fault experiments.Five classification models were evaluated, and the validation accuracy of the k-nearest neighbor (KNN) model was the highest, at 99.8%.The trained KNN model was tested with data collected from two simulated fault types, achieving an average classification accuracy of 98.25%.The proposed diagnostic method more accurately and reliably detects faults in PV power plants and could contribute to increasing the efficiency of O&M activities.