Online substation equipment recognition technology

Abstract

Fast and reliable identification on transformer substation devices is the prerequisite for AR system to perform virtual information display and virtual-real fusion. Hence the author proposes to establish a transformer substation equipment recognition model relying on deep-learning technology, and deploy it on edge devices such as AR, etc. Firstly, collect the images and videos of transformer substation devices, obtain the dataset of transformer substation devices, and use the mark labeling software to build the dataset. Secondly, apply the Faster RCNN object identification algorithm to establish the transformer substation devices identification model on the basis of VGG16 convolutional network. Then, improve the precision of the model through data migration model training, parameter optimization, and dataset enhancement methods such as image transformation. Finally, deploy the algorithm to Intel Neural Compute Stick 2, realizing the online identification of major devices in transformer substation such as the main transformer, breaker, voltage transformer, current transformer and control cabinet, and providing basis for the application of AR system on the training, practical inspection, and operation and maintenance.