A Method to Identify Inrush Current Based on Deep Convolutional Neural Networks

Guo-xing WU, Jing-zhuo ZHANG, Zhen-yu LAI, Shu-liang HE, Xi-shan WEN

Abstract


In this paper, the deep convolutional neural network (CNN) is proposed as the core classifier to discriminate between the magnetizing inrush and the internal fault of power transformers. Two novel CNN structures, A8-Net and A5-Net, are proposed considering the influence of the features and quantities of the inrush current and the internal fault current waveform image data on the classification effect. Parameters of the CNN structures can be automatically trained from the dataset. Relaying signals for various operating conditions, consisting of internal faults and magnetizing inrush, have been obtained by modelling the three-phase transformer in PSCAD/EMTDC. Half of the dataset is used to train the CNN classifier and the rest are used to evaluate the performance of the proposed algorithm. Experimental results at various testing configurations indicates the efficiency and robustness of the proposed CNN classifier. Even with heavy interferences, the proposed method can still remain stable and functional.

Keywords


Transformer, Inrush current, Internal fault, Deep learning, CNN


DOI
10.12783/dtetr/aemce2019/29483

Refbacks

  • There are currently no refbacks.