.png){kind=icon}
.png){kind=icon}
Return
Dynamic graph convolutional neural network based on multi-scale feature fusion for traffic flow prediction
.png){kind=icon}
PRE
AI
J
N
L
DOI：10.1007/s10489-026-07381-0.png)
.png){kind=icon}
Original request for help.png){kind=icon}
Share.png){kind=icon}
Save
Abstract
En 中文
Traffic flow prediction has the potential to enhance the efficiency of traffic planning and management, drive urban sustainable development, strengthen urban emergency response capabilities, thereby improving residents’ quality of life and enhancing overall urban safety. However, due to the complexity of urban environments and the variability of traffic conditions, effectively integrating long-term and short-term traffic flow characteristics while accurately capturing their complex interactions with periodicity have become key challenges in urban traffic flow prediction. This study addresses this challenge by proposing an innovative dynamic graph convolutional neural network model with multi-scale feature fusion. Our proposed model extracts features from traffic data at multiple time scales and performs multi-source fusion, then learns dynamic spatial features through coupled graph convolutional networks. Finally, a decoder module based on multi-layer CGRU is designed for traffic flow prediction. To validate the effectiveness of the proposed model, extensive experiments were conducted on two real traffic datasets obtained from New York city, namely NYCBike and NYCTaxi. The results demonstrate that our proposed model outperforms other baseline models in terms of MAE, RMSE metrics.
Keywords:
Multi-source feature fusion
Dynamic graph convolution
Traffic flow prediction
Journal
IF:
3.5
Papers:
7.5K
Citations:
1.7W
Organization
Cited Papers