Fault Detection in Photovoltaic Systems Using a Machine Learning Approach

Research and development of intelligent fault monitoring in photovoltaic systems are crucial for efficient energy generation. In response to the industry's demand for innovative solutions to enhance energy output and reduce maintenance costs, this study explores machine-learning approaches for the autonomous detection and classification of faults caused by partial shading and dirt accumulation in photovoltaic modules. The proposed fault detection solutions rely on analyzing different algorithms, including Support Vector Machine, Artificial Neural Network, Random Forest, Decision Tree, and Logistic Regression. The research explored data collected from two real photovoltaic systems, each with distinct module characteristics and power ratings. Data were gathered for systems without faults, with faults simulated by partial shading, and faults simulated by dirt accumulation. Crucial information, including voltage, current, ambient temperature, and irradiance, was recorded to assess and classify these kinds of faults. This study presents three main contributions: the implementation and comparison of multiple machine learning models for fault detection, an investigation into the feasibility of identifying these faults using only electrical and environmental data, and an analysis of model performance in a photovoltaic system different from the one used for training. The results indicate that models trained on a specific system achieve high accuracy but face challenges when applied to systems with different characteristics, suggesting that each new photovoltaic system to be monitored should be included in the training phase to enhance classification performance. Noteworthy results were obtained with the Artificial Neural Network model, achieving precision values exceeding 98%.