Scholarlink

<p>The failure of capacitive voltage transformers (CVTs) can severely impact the safety of the power grid. Existing detection technologies lack the capability to monitor the operational status of CVTs in real-time, which hinders the early identification and resolution of equipment faults. This paper proposes a time series analysis method that integrates MiniRocket with a Gradient Boosting Decision Tree (GBDT) classifier for the early detection of breakdown faults in CVT capacitor units. First, the structure of the CVT and the characteristics exhibited by different fault types are analyzed, and a basic model is established using Simulink. Next, the secondary side voltage data of the CVT is segmented in the time domain using a sliding time window. The features extracted specifically for step signals are combined with those obtained from the MiniRocket algorithm, and classification is performed using a GBDT. Finally, based on the classification results of the time series windows, the corresponding fault type is successfully diagnosed, and the time interval of the fault occurrence is identified. The results demonstrate that this method effectively reduces the false alarm rate compared to the commonly used MiniRocket algorithm, providing a feasible solution for early fault warning without the need for shutdown.</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

CVT capacitor breakdown fault diagnosis integrating MiniRocket and gradient boosting decision trees

  • 期刊论文

  • 发表日期：

    2026年2月23日

查看全文
  • Yuntao Zhao,
  • Xuhui Chen

摘要

The failure of capacitive voltage transformers (CVTs) can severely impact the safety of the power grid. Existing detection technologies lack the capability to monitor the operational status of CVTs in real-time, which hinders the early identification and resolution of equipment faults. This paper proposes a time series analysis method that integrates MiniRocket with a Gradient Boosting Decision Tree (GBDT) classifier for the early detection of breakdown faults in CVT capacitor units. First, the structure of the CVT and the characteristics exhibited by different fault types are analyzed, and a basic model is established using Simulink. Next, the secondary side voltage data of the CVT is segmented in the time domain using a sliding time window. The features extracted specifically for step signals are combined with those obtained from the MiniRocket algorithm, and classification is performed using a GBDT. Finally, based on the classification results of the time series windows, the corresponding fault type is successfully diagnosed, and the time interval of the fault occurrence is identified. The results demonstrate that this method effectively reduces the false alarm rate compared to the commonly used MiniRocket algorithm, providing a feasible solution for early fault warning without the need for shutdown.