Adaptive Spatiotemporal Dependence Learning for Multi-Mode Transportation Demand Prediction

Due to the increasing diversification of urban transportation modes, many urban areas have the problem of unbalanced traffic demand, which makes accurate prediction of traffic demand very important. However, most of the existing studies focus on improving the prediction accuracy of traffic demand on...

Full description

Saved in:
Bibliographic Details
Published in:IEEE transactions on intelligent transportation systems 2022-10, Vol.23 (10), p.18632-18642
Main Authors: Xu, Haihui, Zou, Tao, Liu, Mingzhe, Qiao, Yanan, Wang, Jingjing, Li, Xucheng
Format: Article
Language:English
Subjects:
Autoregressive processes
Convolution
Correlation
deep learning
Feature extraction
Heterogeneity
Public transportation
Semantics
spatial-temporal attention mechanism
Time series analysis
Traffic demand prediction
Transportation networks
Travel demand
Urban areas
Urban transportation
Citations: Items that this one cites
Items that cite this one
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Due to the increasing diversification of urban transportation modes, many urban areas have the problem of unbalanced traffic demand, which makes accurate prediction of traffic demand very important. However, most of the existing studies focus on improving the prediction accuracy of traffic demand on the single spatial relationship of a single traffic mode, ignoring the diversity of spatial relationships and the heterogeneity of transportation stations in the traffic network. In this paper, we propose a Co-Modal Graph Attention neTwork(CMGAT) framework to uncover the impact of different spatial relationships and traffic mode interactions on traffic demand. Specifically, we first utilize a feature embedding block to capture the semantic information from several features. Then, a multiple traffic graphs-based spatial attention mechanism and a multiple time periods-based temporal attention mechanism are proposed to capture spatial and temporal dependencies in multi-mode traffic demands. Moreover, an output layer is provided to incorporate the hidden states and raw time sequences to predict future traffic demand. Finally, we conduct experiments on two real-world datasets, NYC Bike and NYC Taxi, and the results not only demonstrate the superiority of our model, but also indicate the necessity of considering multiple spatial relationships and traffic modes.
ISSN:1524-9050
1558-0016
DOI:10.1109/TITS.2022.3155753