Fast Detection And Identification Of Transmission Line Faults Based On Convolutional Neural Networks

In the power system,the patrol and protection of transmission line is an important part to ensure the normal operation.In most cases,image processing technology is used to recognize and detect massive images collected by UAV.However,due to the complexity and time-consuming of traditional machine learning steps,it is no longer suitable for large number of data sets recognition and detection.With the development of computer technology,deep learning arithmetic has emerged.It can automatically extract image features,so that images can be better recognized,and it has stronger generalization ability than traditional networks.However,in the deep learning algorithm,the shallow neural network is not ideal for image recognition under complex background,which results in the image data sets with large amount of data and relatively complex background in the deep neural network.In terms of the particularity of transmission line fault image,this paper chooses deep neural network to recognize and detect the data set.This paper takes TensorFlow as a platform to identify and detect nine types of transmission line faults.It is proposed to use Faster-RCNN network based on VGG16 to identify and detect transmission line faults.By adjusting the batch_size of RPN network,the positive and negative sample ratios in the network are determined,so as to achieve the goal of network optimization,and finally to form the transmission line fault detection network model.In the process of network training,firstly,the transmission line fault images collected by UAV are segmented and preprocessed,and then the data set is expanded by image enhancement technology.3300 images are obtained and divided into training set and test set.Then,the appropriate basic convolutional neural network is selected,and the original Faster-RCNN network is ameliorated by improving the RPN network and using softened non-maximum suppression as the detection network of data sets.Finally,the prediction block diagram and its prediction score are obtained through the network,and various kinds of MAP are obtained through statistical operation.The total MAP of the network test set is 93.68%.The experiment shows that in the complex background of transmission line fault image data centralization,the improved detection network can be used to identify anddetect the transmission line fault,which can make a good judgment.The improved network effectively boost the operation efficiency of the network and the accuracy of fault identification of transmission lines.It provides a new idea to boost the efficiency of power grid patrol and ensure the normal operation of power grid.