Enhancing the Reliability of PV Fault Diagnosis through a Hybrid CNN–DT Approach

Abstract

This paper presents a hybrid diagnostic approach combining a one-dimensional convolutional neural network (1D-CNN) and a Decision Tree (DT) for automatic fault detection in photovoltaic (PV) systems. The CNN performs feature extraction and preliminary classification using normalized electrical and environmental data, while the DT refines predictions based on confidence scores, irradiance, and temperature. Experimental results show that the proposed CNN–DT framework improves diagnostic accuracy, interpretability, and robustness compared to conventional methods, ensuring reliable operation of PV installations under varying conditions.