A deep recognition network of capacitor voltage transformer based on dilated convolution and Bi-LSTM. (9th January 2023)

• Record Type: Journal Article
• Authors: Wu, Jie; Li, Shilong; Chang, Zhengwei; Chen, Mingju; Xiong, Xingzhong; Duan, Zhengxu
• Is Part Of: International journal of power and energy conversion. Volume 13:Number 2(2023)
• Page Start: 131
• Page End: 143
• Publication Date: 2023-01-09
• Subjects: bi-directional long-short term memory network -- capacitor voltage transformer -- self-attention mechanism -- dilated convolution -- fault identification; Energy conversion -- Periodicals; Power resources -- Periodicals; Electric power production -- Periodicals; 621.31205
• Languages: English
• ISSNs: 1757-1154

Abstract

In this paper, a novel deep network is proposed to recognise weak faults of capacitive voltage transformer (CVT). The network takes the supervisory control and data acquisition data (SCADA) of CVT as the analysis and identification object. Spatial features of voltage data are first extracted by dilated convolution and self-attention mechanism. Then time characteristics of SCADA are extracted from both forward and backward directions, using bidirectional long-term and short-term memory network. Finally, the normalised mean square error of the spatio-temporal characteristic information is calculated and compared with the threshold value, so as to discern faults of the capacitive voltage transformer. The comparative experiments show that the proposed network is sensitive to value change of the capacitive voltage transformer, and can efficiently recognise the weak faults of the capacitive voltage transformer.