Scholarlink

Abstract <p>With the widespread operation of high-speed trains, traction motors are susceptible to failures under high-temperature conditions and complex working scenarios. This leads to abnormal temperature increases, which pose significant threats to train safety. Therefore, researching the temperature prediction of traction motors is of great significance for ensuring the safe operation of trains. This paper proposes a temperature prediction method based on an improved bidirectional long short-term memory (Bi-LSTM) network. Initially, convolutional neural networks (CNN) are employed to extract temporal structures from the data. Simultaneously, the squeeze-and-excitation (SE) attention mechanism is integrated with the key-query-value (KQV) mechanism. Finally, Bi-LSTM is utilized for temperature prediction. The proposed SE-BiLSTM-CNN model exhibits significantly enhanced prediction performance compared with recurrent neural networks (RNN), gated recurrent units (GRU), LSTM, and Bi-LSTM models, with improvements of 35.51, 56.07, and 33.75% in the evaluation metrics of MAE, MSE, and RMSE, respectively.</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

Temperature Prediction Model of High-Speed Train Traction Motor Based on SE-BILSTM-CNN

  • 期刊论文

  • 发表日期：

    2025年12月1日

查看全文
  • Yunming Wang,
  • Jiacai Gao,
  • Xianwu Chu,
  • Xinping Wang,
  • Jinbo Tang

摘要

Abstract

With the widespread operation of high-speed trains, traction motors are susceptible to failures under high-temperature conditions and complex working scenarios. This leads to abnormal temperature increases, which pose significant threats to train safety. Therefore, researching the temperature prediction of traction motors is of great significance for ensuring the safe operation of trains. This paper proposes a temperature prediction method based on an improved bidirectional long short-term memory (Bi-LSTM) network. Initially, convolutional neural networks (CNN) are employed to extract temporal structures from the data. Simultaneously, the squeeze-and-excitation (SE) attention mechanism is integrated with the key-query-value (KQV) mechanism. Finally, Bi-LSTM is utilized for temperature prediction. The proposed SE-BiLSTM-CNN model exhibits significantly enhanced prediction performance compared with recurrent neural networks (RNN), gated recurrent units (GRU), LSTM, and Bi-LSTM models, with improvements of 35.51, 56.07, and 33.75% in the evaluation metrics of MAE, MSE, and RMSE, respectively.