Fig 1.
Spatio-temporal data.
Fig 2.
Overall architecture diagram.
Fig 3.
Heat map of node relationships.
Table 1.
The information of the PeMS dataset used.
Fig 4.
Effect of attentional head count on evaluation indicators.
(a)The effect of the number of attention heads on MAE. (b)The effect of the number of attention heads on RMSE.
Table 2.
Comparison of performance metrics for different models.
Fig 5.
MFSTBiSGAT predicts performance of results.
(a)MFSTBiSGAT is on node 101 of the 04 dataset. (b)MFSTBiSGAT is on node 108 of the 08 dataset.
Fig 6.
Prediction errors for each model at different time steps on the two datasets.
The MAE and RMSE errors on the PeMSD4 dataset are denoted as (a) and (b), respectively. The MAE and RMSE errors on the PeMSD8 dataset are denoted as (c) and (d), respectively.