Scholarlink

<p>In high-speed trains and rail transit, insulated gate bipolar transistors (IGBTs) are critical to traction systems, affecting efficiency and safety. Due to exposure to high voltage, large currents, and thermal cycling, IGBT modules are prone to degradation, making accurate lifetime prediction essential. Existing models struggle with complex degradation patterns and long-term forecasting.This paper proposes a deep learning framework combining convolutional neural network (CNN), long short-term memory (LSTM) network, and attention mechanism, optimized via the sparrow search algorithm (SSA). Using thermal resistance data from accelerated aging tests on two IGBT modules, the proposed model shows superior performance. On the 1474-#5 dataset, the root mean square error (RMSE) of the proposed model is reduced by 21.61 %, 21.90 %, 17.13 %, and 4.29 % compared to the CNN-LSTM-Attention, CNN-LSTM, SSA-CNN-LSTM, and Bayesian-CNN-LSTM-attention models, respectively. On the 1474-#6 dataset, RMSE reductions are 12.28 %, 24.20 %, 13.71 %, and 5.77 %, respectively. The model outperforms all baselines in RMSE, mean absolute error (MAE), and mean absolute percentage error (MAPE), while reducing long-term prediction drift, demonstrating its superior performance in IGBT lifetime prediction. This method is expected to enhance the accuracy of IGBT lifetime prediction, providing valuable support for maintenance and operational decision-making in highspeed rail transit systems.</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

An attention-guided CNN-LSTM framework optimized by sparrow search algorithm for IGBT lifetime prediction

  • 期刊论文

  • 发表日期：

    2026年6月3日

查看全文
  • Yang Li,
  • Chunjun Chen,
  • Xiang Li,
  • Min Zhang

摘要

In high-speed trains and rail transit, insulated gate bipolar transistors (IGBTs) are critical to traction systems, affecting efficiency and safety. Due to exposure to high voltage, large currents, and thermal cycling, IGBT modules are prone to degradation, making accurate lifetime prediction essential. Existing models struggle with complex degradation patterns and long-term forecasting.This paper proposes a deep learning framework combining convolutional neural network (CNN), long short-term memory (LSTM) network, and attention mechanism, optimized via the sparrow search algorithm (SSA). Using thermal resistance data from accelerated aging tests on two IGBT modules, the proposed model shows superior performance. On the 1474-#5 dataset, the root mean square error (RMSE) of the proposed model is reduced by 21.61 %, 21.90 %, 17.13 %, and 4.29 % compared to the CNN-LSTM-Attention, CNN-LSTM, SSA-CNN-LSTM, and Bayesian-CNN-LSTM-attention models, respectively. On the 1474-#6 dataset, RMSE reductions are 12.28 %, 24.20 %, 13.71 %, and 5.77 %, respectively. The model outperforms all baselines in RMSE, mean absolute error (MAE), and mean absolute percentage error (MAPE), while reducing long-term prediction drift, demonstrating its superior performance in IGBT lifetime prediction. This method is expected to enhance the accuracy of IGBT lifetime prediction, providing valuable support for maintenance and operational decision-making in highspeed rail transit systems.