Scholarlink

With the increasing of power system scale and complexity, transformer monitoring and maintenance is particularly important. How to use multi-dimensional and multi-source transformer monitoring data efficiently is an important proposition in transformer fault diagnosis. However, it is difficult to obtain transformer abnormal state data. Compared with traditional machine learning algorithms, deep learning has a strong ability to extract data features, and generative adversarial network (GAN) is a deep learning model. The adversarial training between generator (G) and discriminator (D) can achieve high-quality data generation and realistic simulation, which can well help solve the pain point of insufficient abnormal data of transformer. This paper first introduces the working principle and characteristics of GAN, then introduces a variety of derivative models widely used in transformer monitoring data enhancement from the perspective of application, and then reviews the application status of GAN in transformer fault diagnosis in detail. Finally, the problems faced by the further application of GAN in power transformer fault diagnosis are summarized, and the future application prospect is forecasted.

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

A Review of the Application of Generative Adversarial Networks in Transformer Fault Diagnosis

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Tao Wang,
  • Hua Ge,
  • Yong Zhao,
  • Jingwen Bai,
  • Xiangdong Zhang,
  • Wanhao Wu,
  • Qingquan Li

摘要

With the increasing of power system scale and complexity, transformer monitoring and maintenance is particularly important. How to use multi-dimensional and multi-source transformer monitoring data efficiently is an important proposition in transformer fault diagnosis. However, it is difficult to obtain transformer abnormal state data. Compared with traditional machine learning algorithms, deep learning has a strong ability to extract data features, and generative adversarial network (GAN) is a deep learning model. The adversarial training between generator (G) and discriminator (D) can achieve high-quality data generation and realistic simulation, which can well help solve the pain point of insufficient abnormal data of transformer. This paper first introduces the working principle and characteristics of GAN, then introduces a variety of derivative models widely used in transformer monitoring data enhancement from the perspective of application, and then reviews the application status of GAN in transformer fault diagnosis in detail. Finally, the problems faced by the further application of GAN in power transformer fault diagnosis are summarized, and the future application prospect is forecasted.