铁道标准设计2025，Vol.69Issue(5)：73-78,94,7.DOI:10.13238/j.issn.1004-2954.202308250003
基于深度神经网络的高铁沿线风速风向联合预测研究
Research on Joint Prediction of Wind Speed and Direction Along High-speed Railways Based on Deep Neural Networks
摘要
Abstract
Wind speed and direction are critical factors influencing the safety of high-speed train operations.Effective prediction of speed and direction of strong winds along high-speed railways is essential for timely assessment and early warning of train operation conditions.Current research on strong winds along high-speed railways primarily focuses on wind speed prediction,with joint prediction of wind speed and direction yet to be explored.Based on a long short-term memory(LSTM)deep recurrent neural network,three methods for the joint prediction of wind speed and direction were proposed:independent prediction,component prediction,and multivariate prediction.Real monitoring data from multiple stations along the Lanzhou-Xinjiang high-speed railway were used to perform short-term joint prediction of wind speed and direction.First,the original wind speed and direction sequences were preprocessed and normalized,and control variate method was used to determine the optimal time step and model parameters.Then,backpropagation through time(BPTT)and Adam algorithms were employed for iterative training,with early stopping used to control convergence,resulting in an optimized network structure.Finally,the trained LSTM network was used to jointly predict wind speed and direction using the three proposed methods.Experimental results from the four stations showed that the optimized LSTM model could effectively extract long-term dependency features from the wind speed and direction time series,and joint prediction methods could achieve high-precision synchronous predictions for both wind speed and direction.All three joint prediction methods accurately predicted wind speed and direction within a relatively small range.Except for station 5520,the wind speed prediction error was within 15％,and the wind direction prediction error was within 20％.Multivariate prediction method exhibited the highest overall prediction accuracy,followed by independent prediction method.This study offers a new perspective for the joint prediction of wind speed and direction and provides valuable reference for ensuring the safety of high-speed train operations.关键词
高速铁路/风速风向联合预测/大风监测/控制变量法/深度神经网络Key words
high-speed railway/joint prediction of wind speed and direction/strong wind monitoring/control variate method/deep neural networks分类
交通运输引用本文复制引用
肖图刚,王涵玉,文旭光,洪彧,蒲黔辉..基于深度神经网络的高铁沿线风速风向联合预测研究[J].铁道标准设计,2025,69(5):73-78,94,7.基金项目
广西科技计划项目(AA21077011) （AA21077011）
