Scholarlink

<p>The high-speed railway contact system provides electricity to trains, and its stability and reliability directly impact train safety and performance. To avoid economic losses from excessive upkeep and shutdowns due to unclear predictions of the contact system’s remaining useful life(RUL), an error-improved bidirectional long and short-term memory network (BiLSTM) early warning model is proposed, which firstly predicts the RUL of the contact system of high-speed railroads, and then combines the residuals of the predicted value to realize the fault warning of the contact system. The normalized contact system data are input into the PSO-BiLSTM model, and the relative error is corrected by using extreme gradient boosting (XGBoost). The research confirms that the RMSE of this prediction model decreases to 11.6293, and the MAE decreases to 7.9643; then reorder the maintenance data with pole number as index, perform feature parameter fusion based on the prediction residuals from the previous step, and use the kernel density estimation (KDE) method to determine the warning threshold, The results showed that the early warning model was able to predict contact system failures up to 30 days in advance, providing timely maintenance plans to identify potential failures and risks.</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

Remaining useful life prediction and early warning model for high-speed railway contact system based on error-improved PSO-BiLSTM

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2026年1月5日

查看全文
  • Li Liu,
  • Yuting Gan,
  • ZiJian Deng,
  • PeiWen Cheng

摘要

The high-speed railway contact system provides electricity to trains, and its stability and reliability directly impact train safety and performance. To avoid economic losses from excessive upkeep and shutdowns due to unclear predictions of the contact system’s remaining useful life(RUL), an error-improved bidirectional long and short-term memory network (BiLSTM) early warning model is proposed, which firstly predicts the RUL of the contact system of high-speed railroads, and then combines the residuals of the predicted value to realize the fault warning of the contact system. The normalized contact system data are input into the PSO-BiLSTM model, and the relative error is corrected by using extreme gradient boosting (XGBoost). The research confirms that the RMSE of this prediction model decreases to 11.6293, and the MAE decreases to 7.9643; then reorder the maintenance data with pole number as index, perform feature parameter fusion based on the prediction residuals from the previous step, and use the kernel density estimation (KDE) method to determine the warning threshold, The results showed that the early warning model was able to predict contact system failures up to 30 days in advance, providing timely maintenance plans to identify potential failures and risks.