Performance Analysis of Machine Learning Algorithms for Fault Detection in Solar Panels

Abstract

Nowadays, with the continuous increase in energy demand, renewable energy sources, especially solar and wind energy, are gaining great importance. This situation increases the importance of providing uninterrupted, continuous and high-quality energy, and due to the many factors affecting the efficiency of solar energy systems, it is necessary to accurately detect the failures that may occur in these systems and to use innovative methods. In this study, the performance of different machine learning algorithms in classifying 3 different types of faults, namely short circuit, open circuit and shadowing, which are frequently encountered in solar panels, is examined and detailed performance reporting is made. Thus, it is aimed to minimize the interruptions in energy production, to plan maintenance and repair operations more effectively and thus to increase system efficiency and reliability. A dataset of more than 6 million data, each with 19 measurements, was used to train and test the models, and performance is reported in terms of accuracy and F1 scores. Accuracies of 96.4%, 96.7% and 60.7% were achieved in the tests performed with Decision Tree, Random Forest and Bayesian algorithms, respectively. To further analyses, feature selection was applied, reducing the number of input attributes from 19 to 8. This led to a training time improvement of up to 40% while preserving model accuracy. Additionally, the effects of different cell connection configurations on fault detection performance are analyzed through comprehensive reporting. © 2025 Elsevier B.V., All rights reserved.