.png){kind=icon}
.png){kind=icon}
Return
ASTE: adaptive spatial-temporal enhanced graph convolution network for traffic flow prediction
.png){kind=icon}
PRE
AI
DOI：10.1007/s10586-025-05559-8.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
As a crucial component of Intelligent Transportation Systems (ITS), traffic flow prediction plays a vital role in enhancing traffic efficiency, alleviating congestion, and optimizing traffic planning. The most challenging issue in traffic flow prediction lies in fully extracting the spatial and temporal correlations hidden in the historical traffic data of nodes. Traditional traffic prediction models typically rely on a single graph structure or a single temporal convolution module to capture spatial and temporal correlations, which limits the model’s ability to explore these correlations from multiple perspectives. To address these limitations, this paper proposes a Adaptive Spatial-Temporal Enhanced Graph Convolution Network (ASTEGCN) for traffic flow prediction. First, a local spatial-temporal graph convolution network and a node adaptive graph convolution network are designed to capture complex spatial correlations and extract the unique spatial pattern information of each node. The synergy between these two graphs enables the model to extract spatial correlations from dual perspectives. Second, two temporal extraction modules with distinct functions are designed: multi-resolution gated temporal convolutional network extracts temporal correlations across continuous time steps, while the jump-gated temporal convolution network captures temporal correlations across discontinuous time steps. These two modules complement each other, allowing the model to extract temporal correlations from two perspectives. Finally, on the whole model, the double residual stacking mechanism is used to improve the nonlinear and predictive ability of the model. Our experimental results on multiple public traffic datasets show that our approach consistently achieves the best performance compared to other baselines.
Keywords:
Traffic flow prediction
Tand spatial correlations
Graph convolutional network
Temporal convolution network
Journal
C
IF:
0
Papers:
691
Citations:
1