Fault diagnosis of converter transformers based on time-frequency maps and parallel CNN-LSTM

CUI Peng; WU Mingkai; LU Hongjian; AN Yanjie; CAO Zhi; FENG Xiaowei; MA Runze

doi:10.3969/j.issn.1672-3643.2025.04.008

English Version Reader LoginJournal Login

您当前的位置：

首页 >

文章列表页 >

Fault diagnosis of converter transformers based on time-frequency maps and parallel CNN-LSTM

更新时间：2025-11-08

- Fault diagnosis of converter transformers based on time-frequency maps and parallel CNN-LSTM
- Ningxia Electric Power Issue 4, (2025)
- 作者机构：
  
  1. 国网宁夏电力有限公司超高压公司,宁夏,银川,750011
  2. 国网宁夏电力有限公司,宁夏,银川,750001
  3. 中国科学院上海微系统与信息技术研究所,上海,201800
- 作者简介：
- 基金信息：
- DOI：10.3969/j.issn.1672-3643.2025.04.008
  CLC：
- Published Online：20 August 2025，
  
  Published：20 August 2025
- 稿件说明：
移动端阅览
CUI Peng, WU Mingkai, LU Hongjian, et al. Fault diagnosis of converter transformers based on time-frequency maps and parallel CNN-LSTM[J]. Ningxia Electric Power, 2025, (4).
DOI：

CUI Peng, WU Mingkai, LU Hongjian, et al. Fault diagnosis of converter transformers based on time-frequency maps and parallel CNN-LSTM[J]. Ningxia Electric Power, 2025, (4). DOI： 10.3969/j.issn.1672-3643.2025.04.008.

摘要

换流变压器是特高压直流输电系统的核心组件，其状态评估对保障电网稳定运行至关重要。针对换流变压器故障特征提取难和诊断精度不足的问题，提出了一种基于时频图与并行卷积神经网络（convolutional neural network，CNN）和长短期记忆网络（long short-term memory network，LSTM）的换流变压器故障诊断方法。该方法首先通过连续小波变换（continuous wavelet transform，CWT）将一维时序信号转换为时频图；随后，一维时序信号与二维时频图像被同时送入双分支并行CNN进行特征提取；其次，对两种模态的特征进行拼接融合，利用压缩和激励（squeeze-and-excitation，SE）通道注意力增强关键特征；最终，通过LSTM进一步挖掘时间特征信息，进行故障诊断。实验结果表明，所提方法在换流站实际数据上的诊断准确率显著优于现有先进方法，能够实现更有效的特征提取，验证了其在故障识别方面的优越性。

Abstract

The converter transformer is a core component in ultra-high voltage direct current(UHVDC) transmission systems

and its condition monitoring is crucial to ensuring the stable operation of the power grid.To address the challenges of poor fault feature extraction and limited diagnostic accuracy in converter transformers

this paper proposes a fault diagnosis method for converter transformers based on time-frequency maps and a parallel CNN-LSTM architecture.First

one-dimensional time-series signals are converted into time-frequency maps using continuous wavelet transform(CWT).〖JP+1〗Subsequently

both the one-dimensional time-series signals and two-dimensional time-frequency images are simulta-neously input into a dual-branch parallel convolutional neural network(CNN) for feature extraction.Next

the extracted features from both modalities are subsequently concatenated and fused

with important features further enhanced using a squeeze-and-excitation(SE) channel attention mechanism.Finally

a long short-term memory(LSTM) network is used to capture temporal feature information for fault diagnosis.Experimental results on real-world data from converter stations show that the proposed method significantly outperforms existing advanced methods in terms of diagnostic accuracy

demonstrating improved feature representation and superior fault identification capabilities.

关键词

Keywords

references

Views

下载量

CSCD

Alert me when the article has been cited

提交

Tools

Publicity Resources

Composite fault diagnosis method of photovoltaic modules based on feature extraction and machine learning

Mechanical fault diagnosis method for high voltage circuit breakers based on SS FA ELM algorithm

Research on the development process of partial discharge in GIS due to insulator gap defects

Design and application of the insulation online monitoring system for high voltage bushing in the network side of the converter transformers

Research on operation reliability and fault diagnosis technology of pre-insertion resistors for circuit breakers

Related Author

DONG Jinlong

HAN Xiang

JIAO Pingyang

XIA Kun

WANG Qiming

HE Yu

YIN Xiangguo

BAI Xue

Related Institution

Ultra High Voltage Company,State Grid Ningxia Electric Power Co.,Ltd.

Nanjing ODES Intelligent Control System Co.,Ltd.

State Grid Changsha Power Supply Company

State Grid Hunan Electric Power Co.,Ltd.

AI问答

⁰