Scholarlink

<p>Accurately and effectively identifying cable insulation faults is crucial for ensuring the reliability of the power system. Due to the complex and non-stationary characteristics of fault signals, traditional identification methods often suffer from problems such as low accuracy and cumbersome processing. In order to achieve excellent multi-scale feature extraction and classification performance, this study proposes a cable insulation fault identification model that integrates variational mode decomposition (VMD), composite multi-scale dispersion entropy (CMDE), and multi-scale cascaded deep belief network (MCDBN). The model first decomposes cable fault signals using VMD and reconstructs key modal components based on kurtosis values. Subsequently, CMDE is calculated based on the reconstructed signal to obtain robust multi-scale feature vectors. These features are input into the MCDBN network, which integrates an improved multi-scale coarse-grained mechanism with parallel deep feature learning functionality. The model is trained and validated using a 35 kV cross-linked polyethylene power cable fault dataset, which contains 40,000 samples covering four types of faults. The accuracy of the proposed VMD–CMDE–MCDBN model is 98.72%, with accuracy, recall, and F1 score of 98.45%, 98.58%, and 98.51%, respectively. The comparative experiments with mainstream models such as VMD–MCDBN, CMDE–MCDBN, and CNN–SVM have verified the superior performance and stability of the proposed method.</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

Cable insulation fault identification model based on VMD–CMDE and multi-scale cascaded deep belief network

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年6月28日

查看全文
  • Hui Jia,
  • Lixin Yu,
  • Xiaohua Zhang,
  • Hongxue Li,
  • Minghai Zheng,
  • Weichao Li,
  • Wei Chen

摘要

Accurately and effectively identifying cable insulation faults is crucial for ensuring the reliability of the power system. Due to the complex and non-stationary characteristics of fault signals, traditional identification methods often suffer from problems such as low accuracy and cumbersome processing. In order to achieve excellent multi-scale feature extraction and classification performance, this study proposes a cable insulation fault identification model that integrates variational mode decomposition (VMD), composite multi-scale dispersion entropy (CMDE), and multi-scale cascaded deep belief network (MCDBN). The model first decomposes cable fault signals using VMD and reconstructs key modal components based on kurtosis values. Subsequently, CMDE is calculated based on the reconstructed signal to obtain robust multi-scale feature vectors. These features are input into the MCDBN network, which integrates an improved multi-scale coarse-grained mechanism with parallel deep feature learning functionality. The model is trained and validated using a 35 kV cross-linked polyethylene power cable fault dataset, which contains 40,000 samples covering four types of faults. The accuracy of the proposed VMD–CMDE–MCDBN model is 98.72%, with accuracy, recall, and F1 score of 98.45%, 98.58%, and 98.51%, respectively. The comparative experiments with mainstream models such as VMD–MCDBN, CMDE–MCDBN, and CNN–SVM have verified the superior performance and stability of the proposed method.