Efficient Fault Detection in Photovoltaic Systems Using Machine Learning: A Comparative Analysis of Tree-Based Models
[ X ]
Tarih
2025
Yazarlar
Dergi Başlığı
Dergi ISSN
Cilt Başlığı
Yayıncı
Institute of Electrical and Electronics Engineers Inc.
Erişim Hakkı
info:eu-repo/semantics/closedAccess
Özet
Using the most recent machine learning techniques, this article examines several defect detection algorithms that may be utilised in photovoltaic (PV) systems. Maintaining energy efficiency and economic stability requires accurate defect detection, which is becoming increasingly important as the use of renewable energy and large-scale photovoltaic systems continue to expand. Faults that are common in photovoltaic systems, such as string, string-to-ground, and string-to-string faults, can have a significant negative influence on production and result in financial losses. The research presents a flaw detection technique that makes use of classifiers such as Decision Tree, Random Forest, XGBoost, Gradient Boosting, and Extra Trees. We used data from a simulated photovoltaic power plant with a capacity of 250 kW. This data included 600 training instances, 50 testing instances, and 30 characteristics. Among the defects that were examined were those that occurred during normal operation (16.67%), string faults (25.5%), string-to-ground faults (24.83%), and string-to-string mistakes (33%). Accuracy, F1 score, recall, precision, and training duration were the aspects that were considered while evaluating the models. Due to the fact that the Decision Tree model earned the maximum accuracy of 95%, an F1-score of 0.949, recall of 0.95, and precision of 0.958, all while requiring just 0.0083 seconds for training, it is an excellent choice for applications that need real-time processing. The next best model was Random Forest, which achieved an accuracy of 88%, an F1-score of 0.879, and a training time of 0.4856 seconds. The accuracies of XGBoost and Gradient Boosting were below average, coming in at 79% and 78%, respectively. Additionally, the training durations for Gradient Boosting were significantly longer, requiring 4.732 seconds. Extra Trees demonstrated the lowest accuracy, with a score of 74%, an F1-score of 0.723, and a training time of 0.3199 seconds.
Açıklama
Conference name : 15th International Conference on Electrical Engineering, ICEENG 2025.
Conference city : Cairo
Conference date : 12 May 2025 - 15 May 2025
Conference code : 209654
Anahtar Kelimeler
Fault detection, Machine learning Classifiers, Performance evaluation, Photovoltaic (PV) systems
Kaynak
2025 15th International Conference on Electrical Engineering, ICEENG 2025
WoS Q DeÄŸeri
Scopus Q DeÄŸeri
Cilt
Sayı
Künye
Alqaraghuli, O., Ibrahim, A. (2025). Efficient Fault Detection in Photovoltaic Systems Using Machine Learning: A Comparative Analysis of Tree-Based Models. 2025 15th International Conference on Electrical Engineering, ICEENG 2025. 10.1109/ICEENG64546.2025.11031355
