地理与地理信息科学

2021, v.37(04) 1-9

[打印本页] [关闭]
本期目录(Current Issue) | 过刊浏览(Past Issue) | 高级检索(Advanced Search)

基于图结构的城市道路短时交通流量时空预测模型
Spatio-Temporal Prediction Model of Urban Short-Term Traffic Flow Based on Graph Structure

王海起;李留珂;陈海波;
WANG Hai-qi;LI Liu-ke;CHEN Hai-bo;College of Oceanography and Space Informatics,China University of Petroleum;

摘要(Abstract):

准确、实时的城市短时交通流量预测可为驾驶员提供实时的道路状况预警,是城市智能交通系统发展的重点之一。考虑交通流量数据的时空特征,该文提出一种基于注意力机制的GC-GRU时空预测模型(STGCGRU),模型输入根据交通流量时间特性划分为邻近片段、日周期片段、周周期片段3类,以嵌入图卷积(GC)计算的门控循环单元(GRU)作为基本单元搭建Encoder-Decoder模型框架。其中,GC用以捕捉城市道路图中的空间特征,GRU用以捕捉交通流量时序特征,注意力机制用以调节交通流量的趋势变动性。基于北京市出租车GPS轨迹数据集的实验结果表明,该模型适用于短时交通流量预测,预测精度随预测时长减少而升高;未添加周期性信息模型的预测精度优于常规基准模型,添加周期性信息后预测精度提升,并优于添加周期性信息的DeepST模型。对比不同交通情况,该模型可捕捉易堵路段交通流量的趋势变动性,晚高峰时期预测精度更高,但对交通流量的突增突减不敏感。
Accurate and real-time urban short-term traffic flow forecasting can provide real-time road conditions warning for drivers, which is an important part of urban intelligent transportation system.Considering the spatio-temporal characteristics of traffic flow data, a prediction model of spatial-temporal graph convolution gate recurrent unit network(STGCGRU) based on the attention mechanism is proposed, which takes the gate recurrent unit embedded with graph convolution operation as the basic unit to build the Encoder-Decoder model framework.In this model, the input is divided into adjacent segments, daily-periodic segments and weekly-periodic segments according to the temporal characteristics of traffic flow.Graph convolution is used to capture the spatial features of urban road map.GRU is used to capture the temporal features of traffic flow time series.The attention mechanism can make the model adjust the weights of different segments of the input automatically so that the proposed model can adapt to trend variability of traffic flow.The STGCGRU model is tested with the taxi GPS track data set of Beijing.It is found that the model is suitable for short-time traffic flow prediction, and the prediction accuracy improvement with the decrease of forecasting duration.The prediction accuracy of the STGCGRU model without periodic information is better than that of the conventional benchmark models, and the prediction accuracy is improved after adding periodic information, which is better than the existing DeepST model with periodic information.Compared with different traffic conditions, the prediction accuracy of the proposed model is higher in the evening rush hours.It is expert at capturing the trend variability of traffic flow in the easily blocked road section, but not sensitive to the sudden increase or decrease of traffic flow.

关键词(KeyWords): 短时交通流量预测;图卷积;城市路网;时空特征
short-term traffic flow forecasting;GCN;urban road network;spatio-temporal feature

Abstract:

Keywords:

基金项目(Foundation): 国家自然科学基金项目(41471322)

作者(Author): 王海起;李留珂;陈海波;
WANG Hai-qi;LI Liu-ke;CHEN Hai-bo;College of Oceanography and Space Informatics,China University of Petroleum;

Email:

DOI:

参考文献(References):

扩展功能
本文信息
服务与反馈
本文关键词相关文章
本文作者相关文章
中国知网
分享