Detection and Prediction of Faults in Solar Photovoltaic Arrays Using a Gradient Boosting Decision Tree Model

The possible impacts of solar PV array malfunctions on system performance and safety render them a significant concern. The risks of DC arcing, fire, and damage to modules and wiring are exacerbated in PV systems that lack adequate safety measures. Occurrences such as line-line failures in large photovoltaic installations are plausible and may occur. To prevent power outages and ensure optimal operation, fault investigation is essential. This study examines the impact of various pre-processing approaches, such as median and mean filters that normalise data, on the accuracy of photovoltaic array fault detection and delineation. In the training and testing phases of numerous supervised machine learning algorithms, PCA is employed to identify the most pertinent subsets of lowdimensional information. A novel predictive model, GBDT, is employed to anticipate interactions between PV and targets in order to address class imbalance. This model utilises negative data. Different categories of PV-to-target routes are identified, and their unique characteristics are delineated by the GBDT model. The experimental results indicate that the GBDT-based method attains an accuracy rate of 97.34 %, surpassing the performance of existing models. The findings indicate that intelligent defect detection systems enhance the safety and reliability of photovoltaic arrays.