Scholarlink

<p>With the continuous increase in photovoltaic energy storage system (PESS), fire accidents caused by series arc fault (SAF) have become a frequent occurrence. Timely and accurate identification of SAF is the key to developing an arc fault detection device (AFDD) and preventing electrical fires. Unfortunately, there is still no effective approach to identify such faults. Thus, an identification method of SAF in PESS is proposed in this paper. First, SAF experiments are performed in the battery charging circuit of a PESS, and the time-domain features of the circuit current signal under different charging state conditions are analyzed. Second, the current signal is preprocessed by using singular value decomposition filtering and sigmoid normalization, and then the sample matrices are created. Third, the SAF identification model is established by improving the ConvNeXt network with group convolution and channel shuffle technology. Finally, the identification performance, generalization performance, and robust performance of the model are evaluated with the measured data. The identification accuracy is higher than 95%, and the average inference time per sample is less than 26&#xa0;ms. The method also has good generalization and robustness. It can be used to develop an AFDD suitable for PESSs.</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

Identification of DC series arc faults in PV energy storage systems

  • 期刊论文

  • 发表日期：

    2025年6月11日

查看全文
  • Zhiyong Wang,
  • Zhaohui Fang,
  • Hongxin Gao,
  • Yuezhe Gao,
  • Fengyi Guo

摘要

With the continuous increase in photovoltaic energy storage system (PESS), fire accidents caused by series arc fault (SAF) have become a frequent occurrence. Timely and accurate identification of SAF is the key to developing an arc fault detection device (AFDD) and preventing electrical fires. Unfortunately, there is still no effective approach to identify such faults. Thus, an identification method of SAF in PESS is proposed in this paper. First, SAF experiments are performed in the battery charging circuit of a PESS, and the time-domain features of the circuit current signal under different charging state conditions are analyzed. Second, the current signal is preprocessed by using singular value decomposition filtering and sigmoid normalization, and then the sample matrices are created. Third, the SAF identification model is established by improving the ConvNeXt network with group convolution and channel shuffle technology. Finally, the identification performance, generalization performance, and robust performance of the model are evaluated with the measured data. The identification accuracy is higher than 95%, and the average inference time per sample is less than 26 ms. The method also has good generalization and robustness. It can be used to develop an AFDD suitable for PESSs.