Fault detection in photovoltaic modules using I–V curves and machine learning

Abstract

Considering growing global demand for electricity, it is essential to invest in clean and renewable energy production. Investing in renewable energy production is crucial to creating a sustainable matrix that can efficiently meet consumption needs. Reducing costs for solar energy operation and maintenance (O&M) is critical to ensure the economic viability of photovoltaic (PV) projects. Efficient management of these costs can make the difference between a successful project and an unprofitable installation. This work proposes the structuring of a dataset using real measurements of current–voltage (I–V) curves from a solar plant. Based on this, two classifications were performed to detect shading and soiling in PV modules. Two different methods were used for feature processing. One dataset was created using complete I–V curves, while the other relied solely on key parameters from a characteristic curve. The dataset used incorporates data on current, voltage, irradiance, and module temperature at the time of measurements. Different machine learning models were compared, achieving 100% accuracy in soiling classification and 93.2% in shading classification.