Photovoltaic Fault Diagnosis Based on Residual Network

Abstract

Photovoltaic fault diagnosis plays an important role in photovoltaic operation and maintenance. Traditional fault diagnosis methods have low accuracy and slow recognition speed and are greatly affected by external factors. In this paper, the residual network is used to diagnose faults in the open dataset of electroluminescence. Build based on Pytorch framework. Firstly, the data set is preprocessed to avoid the unnecessary influence of external factors on the experiment. By comparing the influence of different depths of the network on performance, ResNet34 with the best performance is selected. The number of data sets is too small to train the network, so the ResNet34 model pre-trained on ImageNet is used for classification. The effects of different Dropout probability and different gradient descent algorithms on the performance of the test set are analyzed. The experimental results show that the final accuracy is 94.25%, which can effectively diagnose photovoltaic faults.