Researchers tested eight standalone deep learning methods for PV cell fault detection and found that their accuracy was as high as 73%. All methods were trained and tested on the ELPV benchmark dataset, which contains 2,624 electroluminescence (EL) images of PV cells.
Scientists from King Fahd University of Petroleum & Minerals in Saudi Arabia have analyzed the benefits of an ensemble-based deep learning framework for PV cell defect classification. Ensemble deep learning combines multiple deep learning models to improve prediction accuracy.
The group tested eight advanced standalone models and compared their performance to that of two ensemble techniques known as tuning and bagging.
“For the voting technique, we have eight trained voice ensemble models, each with unique performance values. The voice aggregation techniques are applied to improve the overall performance. In this paper, we used the soft voting technique, which assumes a majority vote based on the average performance values of each model,” the team explains. “Wrapping ensemble methods involves sampling the training dataset and distributing it to the different models, using soft voting aggregation for the performance metric.”
All methods were trained and tested on the ELPV benchmark dataset, which contains 2,624 electroluminescence (EL) images of PV cells. The dataset was divided into four classes: functional, moderate, mild and severe defect, and the models were asked to place them in the correct category. In addition, a binary test was also performed, where the functional and moderate classes were considered non-defective and mild and severe were considered defective.
Image: King Fahd University of Petroleum & Minerals, Case Studies in Thermal Engineering, CC BY 4.0
“This study systematically evaluates the performance of popular computer vision architectures – AlexNet, SENet, GoogleNet (Inception V1), Xception, Vision Transformer (ViT), Darknet53, ResNet18 and Squeeze Net – in classifying defects in photovoltaic panels,” said the team. . “This study addresses a significant gap in photovoltaic system research by integrating advanced defect detection techniques with machine learning ensemble methods, improving the reliability and efficiency of solar energy systems under adverse environmental conditions.”
According to the results, when analyzing four classes of defects, the voting ensemble achieved an accuracy of 68.36%, while wrapping had an accuracy of 68.31%. The worst performing individual model was YOLOv3, with an accuracy of 51.27%, while AlexNet had the best individual model results, with 67.62%.
According to the binary test results, ResNet18 achieved the highest accuracy of 73.02%, bagging both the 72.17% and 72.06% votes. The lowest single model accuracy among these settings was that of ViT, at 39.68%.
The methods were presented in “Classification of cell defects in photovoltaic solar panels using deep learning ensemble methods”, published in Case studies in thermal engineering.
This content is copyrighted and may not be reused. If you would like to collaborate with us and reuse some of our content, please contact: editors@pv-magazine.com.
Popular content
