摘要
轨道交通客流预测是轨道交通线网规划的重要内容,是确定轨道交通系统的线网规模、设置轨道站点及布设线路基础。不同类型的轨道站点在城市中的功能定位和布局要求等方面均存在差异,进而导致站点的进出客流量呈现显著的时空分布不均衡性。为了挖掘各类型站点的客流变化规律,将站点自身特征和周边环境特征组成向量因子,运用K-means聚类方法对站点进行分类。在此基础上,将影响乘客出行的多源数据作为输入特征,分别构建了随机森林(RF)模型、门控制循环单元(GRU)模型以及RF-GRU组合模型,从而进行站点短时客流分类预测。利用杭州地铁站自动检票系统(AFC)采集的刷卡客流数据,对所构建的预测模型的有效性进行检验。研究结果表明:利用7个刻画站点自身特征和周边环境特征的参数作为聚类因子,并结合站点客流时间分布数据,可将杭州市地铁站点分为就业导向型车站、职住混合型车站和住宅偏远型车站;采用平均绝对误差以及均方根误差作为评价指标,参数化模型(ARIMA),非参数化模型(SVR),深度学习模型(LSTM,GRU,SAEs和GCN),组合模型(DCRNN,STGCN,STHGCN和DSTHGCN)的预测误差依次降低,其中RF-GRU组合模型的预测精度优于其他的组合模型;对站点进行分类之后,单一模型和组合模型预测结果的精度均有提高。
Rail passenger flow prediction is an important part of rail transit network planning,which is the basis for determining the scale of the rail transit system,setting up rail stations,and laying lines.Different types of rail stations are different in terms of functional positioning and layout requirements in the city,resulting in a significant spatial-temporal imbalance in the passenger flow in and out of the stations.To explore the passenger flow change pattern of each type of station,the station’s characteristics and the surrounding environment characteristics were composed of vector factors,and the K-means clustering method was applied to classify the stations.On this basis,the multiple sources of data affecting passenger travel were used as input features,and the Random Forest(RF)model,the Gated Recurrent Unit(GRU)model,and the RF-GRU combined model were constructed to carry out the station short-term passenger flow classification prediction,respectively.The validity of the prediction models was tested using the swiped passenger flow data collected from the automatic fare collection system of Hangzhou subway stations.Experimental results show thatby using seven parameters that portray the characteristics of the station itself and the characteristics of the surrounding environment as clustering factorsas well as combining with the station passenger flow time distribution data,Hangzhou subway stations can be classified into employment-oriented stations,mixed employment-residential stations,and residential remote stations.By using the average absolute error and the root mean square error as evaluation indices,the prediction errorsdecrease in the order of parametric model(ARIMA),nonparametric model(SVR),deep learning models(LSTM,GRU,SAEs,and GCN),and combined models(DCRNN,STGCN,STHGCN,and DSTHGCN)with the RF-GRU combined model outperforming the other combined models in terms of prediction accuracy.After classifying stations,the accuracy of prediction results of both single model and combined models improved.
作者
王金水
欧雪雯
陈俊岩
唐郑熠
WANG Jinshui;OU Xuewen;CHEN Junyan;TANG Zhengyi(School of Computer Science and Mathematics,Fujian University of Technology,Fuzhou 350118,China;Fujian Provincial Key Laboratory of Big Data Mining and Applications,Fuzhou 350118,China)
出处
《铁道科学与工程学报》
EI
CAS
CSCD
北大核心
2023年第6期2004-2012,共9页
Journal of Railway Science and Engineering
基金
福建省自然科学基金资助项目(2022J01933)
湖南工商大学移动商务智能湖南省重点实验室开放研究基金资助项目(2015TP1002)。
关键词
智能交通
短时客流量预测
组合预测模型
多源数据
随机森林
门控制循环单元
intelligent transportation
short-time passenger flow prediction
combined prediction model
multisource data
random forest
gated recurrent unit