利用深度度量学习提取差动电流非涌流特征的变压器保护方法

打开文本图片集

中图分类号：TM77 文献标志码：A

DOI： 10.7652/xjtuxb202605019 文章编号：0253-987X（2026）05-0194-12

A Transformer Protection Method Based on Non-Inrush Feature Extraction of Differential Current Using Deep Metric Learning

HE Anyang，JIAO Zaibin （School of Electrical Engineering，Xi'an Jiaotong University，Xi'an 71oo49，China）

Abstract： To avoid the adverse effects of magnetizing inrush current on transformer fault identification， an intelligent transformer protection method based on non-inrush feature extraction of differential current using improved deep metric learning is proposed. First，inspired by the ability of power experts to distinguish between the inrush and non-inrush components of differential current through waveform comparison， the proposed method utilized an improved deep metric learning framework that takes paired diferential current images as input and minimizes the feature differences between normal differential current and magnetizing inrush current， thereby achieving non-inrush feature extraction. The improved deep metric learning framework introduced a similarity discrimination loss to overcome the difficulty of manually selecting parameters in classical deep metric learning loss functions. Subsequently， based on the extracted non-inrush features of the diferential current，a fault identification module was trained to construct the intelligent transformer protection method. Finall，the method was validated through simulations and experimental data. The results show that the proposed protection method can accurately identify internal faults while ignoring magnetizing inrush interference， demonstrating good adaptability in scenarios such as noise， transformer energization under fault conditions， current transformer saturation，and over-excitation. The method achieves an accuracy of 100% in all simulated scenarios and 99.20% in experimental scenarios，while also meeting the action time requirements for transformer protection.

Keywords： transformer protection; differential current; inrush current; deep metric learning

电力变压器是电力系统中电能转换和传输的重要元件。（剩余16141字）