Scholarlink

The HVDC converter transformer, a crucial component in ultra-high-voltage direct current (UHVDC) transmission systems, is prone to various typical faults, including gas generation, circuit faults, bushing heating and flashover, oil leakage, misoperation of non-electrical components, and chromatograph malfunctions, with gas generation being the most prevalent. Among these gas generation issues, faults due to hot spots in winding end joints, loosened bolts causing local discharge, and poor welding quality leading to winding short circuits are particularly common, which often result in severe local overheating of the converter transformer. In order to enhance the accuracy of fault diagnosis under local overheating conditions, this paper proposes a fault diagnosis scheme employing a Convolutional Neural Network (CNN) combined with a Long Short-Term Memory (LSTM) network, termed as CNN-CNN-LSTM (CCL), based on shell temperature characteristics and gas chromatograph features. Compared to CNN-LSTM diagnostic models based on oil chromatography and CNN-based models using shell temperature, the CCL model significantly improves fault diagnosis accuracy and further enhances fault recognition under local overheating conditions of converter transformers, offering a new approach for fault diagnosis of these transformers.

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

An Improved CNN-LSTM Converter Transformer Fault Diagnosis Method Based On Enclosure Thermal

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Sunjing Gong,
  • Dezhi Chen,
  • Shiyu Liu,
  • Ziyuan Xin,
  • Weifeng Liu,
  • Xiu Zhou

摘要

The HVDC converter transformer, a crucial component in ultra-high-voltage direct current (UHVDC) transmission systems, is prone to various typical faults, including gas generation, circuit faults, bushing heating and flashover, oil leakage, misoperation of non-electrical components, and chromatograph malfunctions, with gas generation being the most prevalent. Among these gas generation issues, faults due to hot spots in winding end joints, loosened bolts causing local discharge, and poor welding quality leading to winding short circuits are particularly common, which often result in severe local overheating of the converter transformer. In order to enhance the accuracy of fault diagnosis under local overheating conditions, this paper proposes a fault diagnosis scheme employing a Convolutional Neural Network (CNN) combined with a Long Short-Term Memory (LSTM) network, termed as CNN-CNN-LSTM (CCL), based on shell temperature characteristics and gas chromatograph features. Compared to CNN-LSTM diagnostic models based on oil chromatography and CNN-based models using shell temperature, the CCL model significantly improves fault diagnosis accuracy and further enhances fault recognition under local overheating conditions of converter transformers, offering a new approach for fault diagnosis of these transformers.