基于改进LSTM模型的铁路客运站客流预测研究  

Research on Method for Prediction of Passenger Flow of Railway Station Based on Improved LSTM Model

作　　者：彭凯贝 白伟 伍柳伊 王小书 吕晓军 PENG Kaibei;BAI Wei;WU Liuyi;WANG Xiaoshu;LYU Xiaojun(Institute of Computing Technology,China Academy of Railway Sciences Corporation Limited,Beijing 100081,China)

机构地区：[1]中国铁道科学研究院集团有限公司电子计算技术研究所,北京100081

出　　处：《铁道运输与经济》2023年第4期53-60,共8页Railway Transport and Economy

基　　金：国家重点研发计划(2020YFF0304100);北京经纬信息技术有限公司科研项目(DZYF21-31)。

摘　　要：准确地预测旅客到达数量有助于缓解铁路客运站运营压力。为实现铁路客运站客流量预测,以铁路客站进站闸机数据为研究对象,分析不同时间维度下铁路客运站客流的时间分布特征,采用层次聚类算法和阈值聚类算法综合对客流量进行聚类分析。针对传统LSTM模型输入数据分割尺度较大导致网络层数深度不够的问题,构建了改进型LSTM客流预测模型。以北京西站实际客流数据进行方法验证,并将预测结果与其他传统预测模型进行比对分析。结果表明:改进LSTM客流模型有较好的预测结果,比其他传统预测模型预测精度高,预测指标中平均绝对误差(MAE)低10%。说明该方法能较好地刻画客流的时间相关性,深度挖掘客流变化的内在机理,预测性能有明显提升。Accurately predicting the number of passengers arrivals helps alleviate the operation pressure of railway passenger station.In order to realize the passenger flow prediction of railway passenger stations,the data of the entrance gate of railway passenger stations was taken as the research object.This paper analyzed the time distribution characteristics of passenger flow in railway passenger stations in different time dimensions,and hierarchical clustering algorithm and threshold clustering algorithm were taken to cluster the passenger flow,with the improved LSTM passenger flow prediction model built.Given inadequate depth of network layers due to large hierarchical segmentation scale of the input data of the LSTM model,the improved LSTM passenger flow prediction model was constructed.The method was verified by the actual passenger flow data of Beijingxi Railway Station,and the prediction results of traditional prediction model compared.It is shown that,the improved LSTM model reports better prediction results,and the mean absolute error(MAE)of the prediction indicators is 10%lower.This shows that this method can better describe the time correlation of passenger flow,and can deeply mine the internal mechanism of passenger flow change,which means significant improvement in the prediction performance.

关 键 词：铁路客运站 客流预测 改进LSTM模型 时序特征 层次聚类分析 

分 类 号：U293.13[交通运输工程—交通运输规划与管理]