Intelligent monitoring of photovoltaic panels based on infrared detection

• DOI: 10.1016/j.egyr.2022.03.173
• Publication Date: 2022-04-09T02:55:24Z
• Authors: Xing Wang; Wenxian Yang; Bo Qin; Kexiang Wei; Yunyu Ma; Daibing Zhang
• Keywords: Infrared image; Image processing; 0202 electrical engineering, electronic engineering, information engineering; Photovoltaic panel; Electrical engineering. Electronics. Nuclear engineering; 02 engineering and technology; Neural network; Condition monitoring; TK1-9971

With the continuously increasing application of photovoltaic (PV) panels, how to effectively manage these valuable facilities has become an issue concern. To date, some methods have been developed meet this purpose. However, a satisfactory solution not achieved for managing large-scale solar PV power plants. address issue, new panel condition monitoring and fault diagnosis technique is in paper. The uses U-Net neural network classifier combination intelligently analyse panel's infrared thermal images taken by drones or other kinds remote operating systems. novelties research include: (1) U-net trained carry out image segmentation, thereby significantly improving efficiency processing; (2) based on contour features 'mask' true colour images. As compared images, their mask little interference information, reliability accuracy can be guaranteed large extent. In research, 295 were first from panels different health states, then 'masks' manually created using software LabelMe. Secondly, enlarge number samples three expansion (i.e. mirroring left right, flipping up down, cropping zooming in) establish sample database training testing network. Thirdly, use 1852 stored train until create as accurately LabelMe does. will guarantee that segmentation high 95.2%. Finally, four potential criteria (i.e., Contour area, Perimeter, Aspect ratio Ratio area outer rectangle) are proposed characterise calculation results criteria, types faults diagnosed with aid classifiers. classifiers Decision tree, K-Nearest Neighbours algorithm (KNN), Support-vector machine (SVM). shown combined well-trained tree diagnose 99.8% accuracy. Therefore, it may arguably provide promising intelligent tool panels.