Fig 1.
Overall framework of the SSML-Net prediction network.
Fig 2.
Illustration of the graph convolution process.
Fig 3.
State-aware dynamic graph learning mechanism.
Table 1.
Datasets.
Table 2.
Experimental environment.
Table 3.
Performance comparison on three traffic forecasting datasets.
Fig 4.
Comparison between SSML-Net traffic flow predictions and observed values over a single weekday.
Fig 5.
Comparison between LSTM traffic flow predictions and observed values over a single weekday.
Fig 6.
Comparison between DCRNN traffic flow predictions and observed values over a single weekday.
Fig 7.
A figure with three subplots: a) RMSE-based predictive performance comparison; b) MAE-based predictive performance comparison; c) MAPE-based predictive performance comparison.
Fig 8.
A figure with three subplots: a) Comparison of MAE values across different training data volumes; b) Comparison of MAPE values across different training data volumes; c) Comparison of RMSE values across different training data volumes.
Table 4.
Performance comparison of models in different attention headcounts.
Table 5.
Impact of the number of SSML-Net layer on model performance.
Fig 9.
Impact of batch size on model performance.
Table 6.
Time consumptions of training on the dataset METR-LA/ Beijing Traffic Data.
Fig 10.
a) Test RMSE versus the training time; b) Test MAE versus the number of training epochs. (Beijing Traffic Data).