Anomaly Detection in PV Modules: A Comparative Study of DBSCAN, k-means, Isolation Forest, and LOF

Abstract

The increasing adoption of photovoltaic (PV) modules for renewable energy generation highlights the importance of maintaining their performance and efficiency. Anomalies in PV modules can lead to energy losses and reduced system reliability. Classical approaches for detecting PV module anomalies include current-voltage (IV) curve analysis, visual inspection, infrared thermography, and electroluminescence imaging. However, these methods often require employees, manual intervention, and may lack scalability and accuracy.In this study, we explore unsupervised machine learning techniques for detecting anomalies in PV modules using data from a PV plant, aiming to develop more efficient and accurate solutions compared to classical approaches. We evaluate four unsupervised machine learning algorithms—Isolation Forest, Local Outlier Factor (LOF), K-Means and DBSCAN, and compare their effectiveness based on metrics such as accuracy, precision, recall, and F1 score.Our findings reveal that the K-Means and DBSCAN algorithms, when properly tuned, demonstrate superior performance in detecting anomalies, while other algorithms show promise with further optimization. This study provides insights into the potential of unsupervised machine learning algorithms as an alternative to classical approaches for anomaly detection in PV modules, contributing to the development of efficient, accurate, and scalable anomaly detection systems in the solar energy sector and enhancing PV system performance and reliability.