Data-driven methods for detecting under-performing inverters in large-scale photovoltaic plants

In view of the increasing number of large-scale PV plants, the need for effective data management and techniques for ensuring the reliability of these systems has grown. In this context, a comparison of techniques for identifying underperforming inverters based on inverter yield is presented. The k-means technique was applied at two levels to ensure the consistency of the analysis and improve cluster segmentation, while the histogram technique was applied as a baseline for comparison. The results show that the two-level k-means approach yields promising results, with benefits such as being an unsupervised technique. This two-level k-means approach can be particularly useful for the rapid identification of issues in large-scale PV generation systems.