PV panel fault detection based on improved ResNet
Ligang Wu, Changxin Zhao, Zushan Ding, Xiao Zhang, Yiding Wang, Fei Sun, Anming He · 2023
At a time when photovoltaic power generation is becoming more and more mature, the maintenance of photovoltaic modules has also become a major problem. In order to avoid the environmental impact of photovoltaic hot spot damage photovoltaic panels, this leads to a decrease in power generation efficiency, as well as the current detection means small hot spots are often be missed, combined with the need for speed and accuracy in the UAV inspection and cleaning, for the traditional two-level detection method, a ResNet residual network-based full convolutional network model, and further use grayscale, filtering and edge gradient fitting and other measure to enhance the model evaluation conditions, achieving accurate hot spot location on PV panels.The experimental results show that the improved full convolutional network model is very useful for small photovoltaic panels. The experimental results show that the improved full convolutional network model is also effective in detecting small photovoltaic hot spots.The application of this model not only satisfies the requirements of PV inspection for real time, but also has a lower cost, stronger portability, can be applied to most scenarios and greatly reduces the requirements on the environment and the working state of PV modules.