Real-time detection of photovoltaic module faults using a hybrid machine learning model

This paper presents an optimized machine learning framework for the real-time detection of partial shading and module faults in photovoltaic (PV) systems. Using real-world data from the Kaggle platform, we analyze voltage, current, power, and irradiance parameters and compare failure scenarios with normal conditions. The methodology includes data preprocessing, feature selection, model training, and performance evaluation. We compared Hist Gradient Boosting, K-Neighbors, Decision Tree, and Random Forest regression models and optimized their hyperparameters by combining LightGBM with Bayesian optimization. The proposed model demonstrated superior performance, achieving 99.01% accuracy, an R 2 value of 98.9%, and an F1 score of 0.99. The negative effects of partial shading and module failure on power generation are presented in detail, showing that these conditions lead to significant power generation losses. Class imbalance is addressed using data replication and class weighting techniques to improve rare fault detection. This study aims to enhance energy efficiency and promote the sustainability of renewable energy systems by optimizing PV system maintenance processes. This framework is a significant advancement in the development of proactive maintenance strategies for large-scale solar power plants and will guide future research.