基于深度学习的城市公交站点客流预测  

Passenger flow prediction of urban bus stops based on deep learning

作　　者：卢毅[1,2] 王宇阳 李媛[1] LU Yi;WANG Yuyang;LI Yuan(College of Transportation Engineering,Changsha University of Science&Technology,Changsha 410114,China;School of Intelligent Transportation and Logistics,Changsha University of Science&Technology,Changsha 410114,China)

机构地区：[1]长沙理工大学交通学院,湖南长沙410114 [2]长沙理工大学智能交通与现代物流研究院,湖南长沙410114

出　　处：《长沙理工大学学报(自然科学版)》2025年第1期154-162,共9页Journal of Changsha University of Science & Technology:Natural Science

基　　金：湖南省交通运输厅科技进步与创新项目(202030);长沙理工大学研究生科研创新项目(CX2021SS12)。

摘　　要：【目的】城市公交站点客流的变化趋势和时空特征密不可分。本文目的为捕捉公交站点客流的时空特征。【方法】首先,使用图卷积网络捕捉客流的空间特征;接着,借助门控递归单元捕捉客流的时间特征;然后,构建基于深度学习的公交站点客流预测模型,即门控图卷积网络(gated-graph convolutional network,G-GCN)模型;最后,将驻马店市内的512个公交站点的客流数据按照30、45和60 min三种时间粒度进行划分,利用G-GCN模型进行预测,并将该预测结果与基线模型的预测结果进行对比。【结果】在上述三种时间粒度划分下,G-GCN模型的三种均方根误差分别为2.35、3.00和3.57,分别比其他基线模型的平均降低了19.60%、24.40%和26.40%。【结论】本研究成果突破了以往只在规则区域内对公交客流进行预测的局限,为城市公交组织优化提供了技术参考。[Purpose]The changing trend of passenger flow at urban bus stops is inseparable from the temporal and spatial characteristics.This paper aims to capture the temporal and spatial characteristics in passenger flow of bus stops.[Methods]Firstly,the graph convolutional network(GCN)was used to capture the spatial characteristics of the passenger flow;secondly,the gated recurrent unit(GRU)was employed to capture the temporal characteristics of the passenger flow;thirdly,a passenger flow prediction model of bus stops based on deep learning was developed and named gated-graph convolutional network(G-GCN).Finally,the passenger flow data of 512 bus stops in Zhumadian City were divided into three time granularities of 30,45,and 60 min,and the G-GCN model was utilized for prediction.The prediction results were compared with those of the baseline model.[Findings]The root mean square errors of the G-GCN model are 2.35,3.00,and 3.57 at the three time granularities,reducing by 19.60%,24.40%,and 26.40% compared with those of the other baseline models.[Conclusions]The research results break through the limitations of predicting passenger flows in public transport within regular areas and provide technical support for the optimization of urban public transport organizations.

关 键 词：客流和客运量调查与预测 智能交通 深度学习 图卷积网络 门控递归单元 

分 类 号：U492.413[交通运输工程—交通运输规划与管理]