Scholarlink

<p>Partial discharge (PD) poses a significant risk to the reliable operation of high-voltage cables, necessitating swift and precise detection to ensure the stability of power systems. The availability of comprehensive high-frequency voltage signal datasets facilitates data-driven approaches to analyzing PD patterns. Nonetheless, challenges such as the complexity of discharge characterization and the skewed distribution of data remain prevalent. To mitigate noise and interference, advanced signal processing techniques, including signal filtering and flattening, have been applied. For feature extraction, the KL-means method has been employed, enabling effective differentiation of pulse types and robust signal representation. Furthermore, a novel deep semi-supervised learning network, DSL-Net, that synergizes deep neural networks with semi-supervised learning methodologies is proposed in this paper. This approach capitalizes on unlabeled data to enhance model accuracy and address data imbalance issues. Evaluation of publicly available datasets demonstrates a 3.7% improvement in the Matthews correlation coefficient (MCC), highlighting the enhanced effectiveness of the proposed method in PD identification.</p>

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

Fault detection in cable systems via deep semi-supervised learning for partial discharge

  • 期刊论文

  • 发表日期：

    2025年4月23日

查看全文
  • Ang Li,
  • Shilong Li,
  • Jianlei Zhang,
  • Chunyan Zhang

摘要

Partial discharge (PD) poses a significant risk to the reliable operation of high-voltage cables, necessitating swift and precise detection to ensure the stability of power systems. The availability of comprehensive high-frequency voltage signal datasets facilitates data-driven approaches to analyzing PD patterns. Nonetheless, challenges such as the complexity of discharge characterization and the skewed distribution of data remain prevalent. To mitigate noise and interference, advanced signal processing techniques, including signal filtering and flattening, have been applied. For feature extraction, the KL-means method has been employed, enabling effective differentiation of pulse types and robust signal representation. Furthermore, a novel deep semi-supervised learning network, DSL-Net, that synergizes deep neural networks with semi-supervised learning methodologies is proposed in this paper. This approach capitalizes on unlabeled data to enhance model accuracy and address data imbalance issues. Evaluation of publicly available datasets demonstrates a 3.7% improvement in the Matthews correlation coefficient (MCC), highlighting the enhanced effectiveness of the proposed method in PD identification.