Scholarlink

<p>In order to accurately detect series arc fault, this paper proposes a series arc fault detection method based on voltage signal which introduces inception with multi-scale parallel convolution operation, and combines bidirectional long short-term memory recurrent network (BiLSTM) with attention mechanism. Firstly, a household experimental platform was built, and the line voltage signal obtained by the experiment was subjected to wavelet transform and principal component analysis (PCA) dimensionality reduction to construct a dataset. Secondly, Inception is introduced to extract the multi-level features of the samples, and the parallel output is input into BiLSTM after global max pooling layer. Then, self-attention is used to perform reinforcement learning on the hidden state vector. Finally, the output results are classified by the fully connected layer. Compared with the detection results of various algorithms, it is verified that this method has more advantages in the identification of series arc fault. In addition, additional experiments at different sampling frequencies show that the method has good adaptability, and the identification accuracy has better performance when the sampling frequency is 10&#xa0;KHZ, which has certain theoretical guiding significance for the development of the series arc fault detection device in the next step.</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

Research on series arc fault detection method household loads based on voltage signals

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年7月27日

查看全文
  • Bin Li,
  • Jiahui Shu,
  • Feifan Cui

摘要

In order to accurately detect series arc fault, this paper proposes a series arc fault detection method based on voltage signal which introduces inception with multi-scale parallel convolution operation, and combines bidirectional long short-term memory recurrent network (BiLSTM) with attention mechanism. Firstly, a household experimental platform was built, and the line voltage signal obtained by the experiment was subjected to wavelet transform and principal component analysis (PCA) dimensionality reduction to construct a dataset. Secondly, Inception is introduced to extract the multi-level features of the samples, and the parallel output is input into BiLSTM after global max pooling layer. Then, self-attention is used to perform reinforcement learning on the hidden state vector. Finally, the output results are classified by the fully connected layer. Compared with the detection results of various algorithms, it is verified that this method has more advantages in the identification of series arc fault. In addition, additional experiments at different sampling frequencies show that the method has good adaptability, and the identification accuracy has better performance when the sampling frequency is 10 KHZ, which has certain theoretical guiding significance for the development of the series arc fault detection device in the next step.