Scholarlink

<p>Identification and classification of power quality disturbances (PQDs) are essential for solving power quality problems and maintaining power network security. To classify PQDs effectively, we propose the CNN-Transformer model to identify different types of PQDs. The model uses the Convolutional Neural Network (CNN) for feature extraction of local information of PQDs. In contrast, the Transformer is used to make up for CNN’s defects in extracting global information on PQDs. In this study, 10 classes of single disturbances and 15 classes of combined disturbances are designed based on the IEEE 1159 standard. The experimental results show that the classification accuracy of the method is high, with 99.48% under noiseless conditions and 94.32%, 98.96%, 99.16%, and 99.20% at signal-to-noise ratios (SNR) of 20, 30, 40, and 50&#xa0;dB, respectively. The results demonstrate the proposed method’s correctness and effectiveness, proving its significance in power quality disturbance classification.</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 and classification of power quality disturbances using CNN-transformer

  • 期刊论文

  • 发表日期：

    2025年3月28日

查看全文
  • Gaofeng Wang,
  • Hao Zhang,
  • Man Gao,
  • Wuren Ding,
  • Yun Qian

摘要

Identification and classification of power quality disturbances (PQDs) are essential for solving power quality problems and maintaining power network security. To classify PQDs effectively, we propose the CNN-Transformer model to identify different types of PQDs. The model uses the Convolutional Neural Network (CNN) for feature extraction of local information of PQDs. In contrast, the Transformer is used to make up for CNN’s defects in extracting global information on PQDs. In this study, 10 classes of single disturbances and 15 classes of combined disturbances are designed based on the IEEE 1159 standard. The experimental results show that the classification accuracy of the method is high, with 99.48% under noiseless conditions and 94.32%, 98.96%, 99.16%, and 99.20% at signal-to-noise ratios (SNR) of 20, 30, 40, and 50 dB, respectively. The results demonstrate the proposed method’s correctness and effectiveness, proving its significance in power quality disturbance classification.