There is a significant growth in the use of solar photovoltaics (PV) for clean and sustainable renewable energy. The ease of installation and expansion of PV power plants makes them suitable for grid-connected operation. However, increasing the size of the PV power plant will make it difficult to detect and classify faults, which leads to lower system efficiency and reliability. Therefore, this article proposes a method for detecting and classifying faults in PV systems in order to fix them more quickly and efficiently. For this purpose, the data of current and voltage are collected from every panel for a period of time in normal case and for three common fault types (open circuit, short circuit, and shading), and afterwards converted into wavelet transform images. Then a deep learning approach is used to detect, classify, and locate defective panels in a PV system. The obtained results show that the deep learning approach based on Resnet50, and voltage images was 100% accurate compared to the deep learning approach based on current images. In this methodology there is no need for additional sensors since they are embedded with panels and sent to main computer. Since the voltage parameter has the best accuracy scenario, it is converted to images, then ResNet-50 is applied. So, every panel can be detected and classified, and as each panel is numbered, the fault location can be readily determined.
Key words: Convolutional neural network; Deep learning; Photovoltaic systems; Faults detection and classification; Real-time monitoring.
|
