基于ARIMA和LSTM的城市轨道交通延时客流预测方法比较  

Comparison of delayed passenger flow forecasting methods for urban rail transit based on ARIMA and LSTM

作　　者：李德奎[1] 杜书波[2,3] 张鹏 LI Dekui;DU Shubo;ZHANG Peng(School of Computer Science and Technology,Liaocheng University,Liaocheng 252000,China;School of Architecture and Civil Engineering,Liaocheng University,Liaocheng 252000,China;College of Architecture and Urban Planning,Tongji University,Shanghai 200092,China;School of Built Environment,Western Sydney University,Sydney NSW2751,Australia)

机构地区：[1]聊城大学计算机学院,聊城252000 [2]聊城大学建筑工程学院,聊城252000 [3]同济大学建筑与城市规划学院,上海200092 [4]西悉尼大学建筑环境学院,悉尼NSW2751

出　　处：《青岛理工大学学报》2021年第4期135-142,共8页Journal of Qingdao University of Technology

基　　金：山东省高校青年教师成长计划项目;山东省住房城乡建设科技计划项目(2021-K9-3);聊城大学博士基金资助项目(318051531)。

摘　　要：随着我国诸多城市的轨道交通发展从大扩张期过渡到运营期,提升运营效率成为下一阶段的发展主题。北上广深等国内一线城市日益增长的延时运营需求,使平衡城市轨道交通延时运营的时长、成本和运营效率成为精细化运营的巨大挑战。以上海地铁数日数据为例,对刷卡数据进行预处理后,建立基于ARIMA和LSTM的轨道交通延时运营的客流预测模型。然后分别利用全天数据和半天数据,对5和15 min不同时间粒度进行了预测分析。研究结果表明,从整体上看,半天数据相对全天数据普遍均方根误差较小,显示出模型的拟合度较高;从方法上看,LSTM方法比ARIMA方法的均方根误差较小,具有较好的预测效果。研究结果可为轨道交通延时运营中客流预测提供技术支持。With the transition of the development of the urban rail transit from expansion stage to operation stage in many Chinese cities,improving the operational efficiency has been considered as the development theme of the next stage.With the increasing demand for extending operation time in Chinese first-tier cities such as Beijing,Shanghai,Guangzhou and Shenzhen,how to balance the duration,cost and operational efficiency of time-extended operation of the urban rail transit has become a great challenge to refined operation.By using the data from Shanghai Metro and pre-processing the metro card data,delayed passenger flow forecast models for urban rail transit based on ARIMA and LSTM are developed.After conducting the predictive analysis for the 5 minutes intervals and 15 minute intervals by using full-day data and half-day data separately,this research finds that:1)the half-day data generally has a smaller root mean square deviation than the full-day data,which indicates that the model has a high fitting degree;2)LSTM has a smaller root mean square deviation than the ARIMA method and LSTM has a better prediction effect.The findings of this research can provide technical support for passenger flow prediction in the time extended operation of urban rail transit.

关 键 词：城轨交通 延时客流预测 ARIMA LSTM 

分 类 号：U293.1[交通运输工程—交通运输规划与管理]