Scholarlink

<p>High-speed trains are easily affected by extreme crosswinds during operation. Existing short-term wind speed prediction methods overlook the importance of data preprocessing and non-stationarity analysis. To address this, we designed HSR-Former, a method that combines data preprocessing and an improved Transformer for short-term wind speed prediction in high-speed rail systems. This model utilizes data processing and non-stationarity analysis to tackle two key issues. First, it identifies the main trend of wind speed through data preprocessing and encodes the change points of non-stationary sequences. Then, the reconstructed wind speed information is decomposed into trend components and seasonal components, with a multilayer perceptron predicting the trend components and Fourier attention encoding the key point positions to predict the seasonal components. Finally, the results from both parts are integrated to obtain the prediction. The study examines wind speed information from three sites with different geographical characteristics, and the results indicate: a) the data preprocessing module improves data accuracy and prediction performance; b) seasonal trend decomposition and the Fourier attention mechanism effectively capture wind speed change information; c) HSR-Former outperforms all comparison models in terms of prediction accuracy and stability.</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

Hybrid model for high-speed rail wind speed forecasting based on improved transformer

  • 期刊论文

  • 发表日期：

    2025年11月6日

查看全文
  • Xiaoling Ye,
  • Anchuan Gong,
  • Xin Chen,
  • Yingchao Zhang,
  • Yingyi Qi,
  • Xiong Xiong

摘要

High-speed trains are easily affected by extreme crosswinds during operation. Existing short-term wind speed prediction methods overlook the importance of data preprocessing and non-stationarity analysis. To address this, we designed HSR-Former, a method that combines data preprocessing and an improved Transformer for short-term wind speed prediction in high-speed rail systems. This model utilizes data processing and non-stationarity analysis to tackle two key issues. First, it identifies the main trend of wind speed through data preprocessing and encodes the change points of non-stationary sequences. Then, the reconstructed wind speed information is decomposed into trend components and seasonal components, with a multilayer perceptron predicting the trend components and Fourier attention encoding the key point positions to predict the seasonal components. Finally, the results from both parts are integrated to obtain the prediction. The study examines wind speed information from three sites with different geographical characteristics, and the results indicate: a) the data preprocessing module improves data accuracy and prediction performance; b) seasonal trend decomposition and the Fourier attention mechanism effectively capture wind speed change information; c) HSR-Former outperforms all comparison models in terms of prediction accuracy and stability.