Spatio-Temporal Contextual Conditions Causality and Spread Delay-Aware Modeling for Traffic Flow Prediction

Mobility is essential for all of us, and the daily routine of the majority is impacted by vehicular transportation. Thus, the ability to predict traffic flow is a challenging task in the field of intelligent transportation systems. However, achieving precise predictions of the state of traffic is a...

Full description

Saved in:
Bibliographic Details
Published in:IEEE access 2024, Vol.12, p.21250-21261
Main Authors: Xiong, Yijun, Wang, Huajun
Format: Article
Language:English
Subjects:
Causality
Causality graph
contextual conditions
Convolutional neural networks
Correlation
Datasets
Heuristic algorithms
Intelligent transportation systems
multi-graph convolution
Predictive models
Roads
Spatial dependencies
spread time-delay
Time lag
Traffic control
Traffic flow
Traffic models
Citations: Items that this one cites
Online Access:Get full text
Tags: Add Tag
No Tags, Be the first to tag this record!
Description
Summary:Mobility is essential for all of us, and the daily routine of the majority is impacted by vehicular transportation. Thus, the ability to predict traffic flow is a challenging task in the field of intelligent transportation systems. However, achieving precise predictions of the state of traffic is a complex undertaking, there are two challenges: 1) Existing studies do not explicitly account for the causal influence of the "trigger effect" from contextual conditions on spatial dependencies. 2) Prior methods ignore the fact that there is a time delay in the spread of information in large-scale regions. To address these limitations, we present a novel Graph Structural Causality Spread Delay-aware Model (i.e., GSCSDM) for accurate traffic flow prediction. First, we develop a contextual causality graph that learns the spatial graph structure under the "triggering effect". Second, we present a spread time-delay module that captures the information spread delaying triggered by contextual conditions in global regions. Furthermore, we construct a multi-graph fusion matrix to extract spatial correlation from diverse perspectives, which enhances the understanding of regions' state interaction. Experiments on two real datasets demonstrate that GSCSDM significantly outperforms the state-of-the-art methods. Since the "trigger effect" widely exists in practical datasets, the proposed framework may also cast light on other spatial-temporal applications.
ISSN:2169-3536
2169-3536
DOI:10.1109/ACCESS.2024.3357783