Short-term highway traffic speed prediction based on spatio-temporal analysis and attention mechanism

Short-term traffic speed prediction in Intelligent Transportation System (ITS) provides an important idea for solving traffic problems. To capture the spatio-temporal properties of traffic speed prediction, we proposes a Graph Convolutional Network-Gated Recurrent Units with Attention (GCN-GRUA) mode for expressway. The Graph Convolutional Network (GCN) and Gated Recurrent Unit (GRU) were used to extract the spatial and temporal features of traffic speed, and the attention mechanism was introduced to improve the prediction performance of the model. Experimental results from the real traffic data set of Qingyin Expressway show that the proposed model has a significant improvement in prediction accuracy compared with GCN, GRU and GCN-GRU models. In addition, the importance of speed characteristic variables and exogenous variables on the traffic speed prediction accuracy show that the speed data with the closest time interval has the greatest influence on the traffic speed prediction, followed by the daily cycle characteristics of traffic speed. As the prediction time increases, the relative importance of the velocity characteristic variable remains above 0.6, while the relative importance of the exogenous variable keeps rising.