Research and application of target detection algorithm for live operation in substation

Abstract

The substation is an important node connecting the backbone network and distribution network and is the core hub of the smart grid. With the continuous increase of electrical equipment, the substation plays a vital role in the whole power system. The structural principle of deep learning convolutional neural network is deeply studied, which lays a theoretical foundation for substation fault detection and recognition based on the object detection algorithm. This paper focuses on fault detection algorithms based on deep learning. First, two mainstream target detection algorithms, Faster R-CNN and YOLO V5, are applied to the self-made substation fault data set. The experimental comparison shows that while the YOLO V5 algorithm has a slightly higher recognition speed, the accuracy of fault target recognition and similar background image recognition is higher than that of the Faster R-CNN algorithm. Then, for the benchmark network YOLO V5, an improved method is proposed. The innovative idea is to delete large branches from the network detection part for the detection of small faults in the station, so as to reduce the number of model parameters and improve the speed of model detection.