Scholarlink

<p>To address transformer DC bias issues caused by high-voltage direct current transmission systems, this paper proposes an anomaly detection scheme for transformer DC bias. It combines Long Short-Term Memory (LSTM) and Transformer models through multi-information interaction. First, the mutual relationships among various physical information during transformer DC bias anomalies are analyzed. Electromagnetic and mechanical characteristic data from transformer simulations and experiments are collected to form a sample set. Subsequently, the LSTM captures local transient distortion features during DC bias anomalies, while the Transformer model extracts global dependencies within long-term sequences. Subsequently, current and vibration data are fused and interacted to construct a recognition model integrating local and global features. Model accuracy is evaluated using a test dataset, demonstrating that the LSTM-Transformer hybrid model achieves 97.36% accuracy in DC bias anomaly detection. Furthermore, this model exhibits superior generalization capability in small-sample scenarios, overcoming limitations of traditional models.</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 Method of Transformer Bias Anomaly Based on Multi Information Interaction

  • 期刊论文

  • 发表日期：

    2026年3月12日

查看全文
  • Chuang Liu,
  • Guohui Mu

摘要

To address transformer DC bias issues caused by high-voltage direct current transmission systems, this paper proposes an anomaly detection scheme for transformer DC bias. It combines Long Short-Term Memory (LSTM) and Transformer models through multi-information interaction. First, the mutual relationships among various physical information during transformer DC bias anomalies are analyzed. Electromagnetic and mechanical characteristic data from transformer simulations and experiments are collected to form a sample set. Subsequently, the LSTM captures local transient distortion features during DC bias anomalies, while the Transformer model extracts global dependencies within long-term sequences. Subsequently, current and vibration data are fused and interacted to construct a recognition model integrating local and global features. Model accuracy is evaluated using a test dataset, demonstrating that the LSTM-Transformer hybrid model achieves 97.36% accuracy in DC bias anomaly detection. Furthermore, this model exhibits superior generalization capability in small-sample scenarios, overcoming limitations of traditional models.