Spatio-temporal graph neural network based on time series periodic feature fusion for traffic flow prediction.

Autor: Chen, Guihui, Wei, Yuli, Peng, Jiao, Zheng, Xinyu, Lu, Kai, Li, Zhongbing
Zdroj: Journal of Supercomputing; Jan2025, Vol. 81 Issue 1, p1-20, 20p
Abstrakt: Accurate prediction of the future traffic flow of a road section is helpful to disperse and guide vehicles in real-time, which can effectively improve the traffic efficiency of the road network. However, due to the randomness and temporal correlation of traffic flow data, it is easy to be affected by various factors. While previous work has made significant efforts to capture the temporal and spatial correlation of traffic flow, the existing methods still ignore the similarity of traffic flow in historical segments. To address this problem, this paper proposes a traffic flow prediction model based on Sequence-to-Sequence Time Graph Convolutional Network (Seq2Seq-TGCN). Three segments that are strongly correlated with the traffic flow to be predicted are automatically cut out and concatenated in turn on the time axis as the input of the model. Then, an innovative Seq2seq architecture based on temporal graph convolution network is designed to capture the spatio-temporal features of traffic flow. Experiments on two real data sets show that the Seq2Seq-TGCN model has higher prediction accuracy and much faster speed than other algorithms. It is helpful to improve the current increasingly complex traffic conditions. [ABSTRACT FROM AUTHOR]
Databáze: Complementary Index
Externí odkaz: