基于VMD-LSTM混合模型的城际高速铁路时变客流预测  

Forecast of time-dependent passenger flow of Intercity high-speed railway based on VMD-LSTM mixed model

作　　者：苏焕银 彭舒婷 曾琼芳 代慧子 SU Huanyin;PENG Shuting;ZENG Qiongfang;DAI Huizi(School of Railway Tracks and Transportation,Wuyi University,Jiangmen 529020,China;School of Tourism Management,Hunan University of Technology and Business,Changsha 410205,China;Guangzhou Metro Group Co.,Ltd.,Guangzhou 510330,China)

机构地区：[1]五邑大学轨道交通学院,广东江门529020 [2]湖南工商大学旅游管理学院,湖南长沙410205 [3]广州地铁集团有限公司,广东广州510330

出　　处：《铁道科学与工程学报》2023年第4期1200-1210,共11页Journal of Railway Science and Engineering

基　　金：五邑大学高层次人才科研启动项目(2017RC51);江门市基础与理论科学研究类科技计划项目(2022JC01004);国家自然科学基金资助项目(71901093)。

摘　　要：城际高速铁路1周内每天不同出发时段的旅客需求体现出较为稳定的波动规律特征,依据该特征,设计变分模态分解-长短时记忆神经网络(VMD-LSTM)混合模型对城际高速铁路的O-D对客流进行预测,获得1周内每天各时段的旅客需求。首先,依据广珠城际高速铁路的历史售票数据分析旅客出行需求的时间分布特征(时变特征),获取非平稳的客流时间序列;然后,采用VMD方法将非平稳的客流时间序列分解为若干个平稳的客流时间子序列,提取客流的波动特征,设计LSTM神经网络模型对分解后的客流时间子序列进行预测。设置不同的模型参数,选取广珠城际高速铁路的6个典型O-D对进行实验分析,结果表明:1)VMD-LSTM混合模型的隐藏神经元个数和迭代次数的有效增加可以降低预测误差,但是当两者增加到一定量时,误差反而会有增大的趋势,对预测效果影响较大。2)相比于单一的LSTM神经网络模型,VMDLSTM混合模型的预测误差明显降低,说明混合预测模型比单一预测模型具有较高的预测精度。3)VMD-LSTM混合模型获得的各时段预测值与实际值较为接近,分布特征整体一致,说明混合模型能够较好地拟合旅客出行需求的时变特征。4)VMD-LSTM混合模型的MAPE预测误差可控制在10%左右,对于时变特征较为规则的O-D对客流,整体预测效果较好。The passenger demands of intercity high-speed railway in different departure times of every day in a week show a relatively stable fluctuation law.According to this characteristic,the mixed model of variational mode decomposition and long short-term memory neural network(VMD-LSTM)was designed to predict the O-D passenger flow of intercity high-speed railway to obtain daily passenger demand at different times in each day of one week.First,based on the historical ticket sales data of Guangzhou-Zhuhai intercity high-speed railway,the time distribution characteristics(time-dependent characteristics)of passenger travel demand were analyzed and the non-stationary passenger flow time series were obtained.The non-stationary time series of passenger flow were decomposed into several stationary time series by VMD method,and the fluctuation characteristics of passenger flow were extracted,so is the LSTM neural network model designed to predict the decomposed time series.Six typical O-Ds of Guangzhou-Zhuhai intercity high-speed railway were selected and analyzed in the experiment based on different model parameters.The results show that:(1)increasing the number of hidden neurons and iteration times of the VMD-LSTM mixed model can reduce the prediction error;however,when both increase to a certain amount,the error increases and has a great influence on the prediction effect.(2)compared with the single LSTM model,the prediction error of VMD-LSTM mixed model is obviously lower,which shows that the mixed model has higher prediction precision than the single prediction model.(3)the VMDLSTM mixed model yields the forecast value of each time period close to the actual value,and the distribution characteristic is consistent.This shows that the mixed model can fit the time-dependent characteristic of the passenger travel demand well.(4)the MAPE prediction error of the VMD-LSTM mixed model can be controlled around 10%.For the O-D passenger flow with regular time-dependent characteristics,the overall prediction effect is better.

关 键 词：城际高速铁路 时变客流 VMD方法 LSTM神经网络 预测精度 

分 类 号：U293.5[交通运输工程—交通运输规划与管理]