Table 1.
Comparison of dynamic graph methods and proposed models.
Table 2.
Summary of the related works.
Fig 1.
The Proposed STF-GGRU.
Fig 2.
The spatial network of Sensors.
Fig 3.
Recurrent patterns in time series data.
Fig 4.
Integrated spatiotemporal feature extraction alignment.
Fig 5.
Flowchart of the ISTFA array.
Fig 6.
Graph Convolution Network Architecture [36].
Fig 7.
Gated Recurrent Unit Network Architecture.
Fig 8.
Fully connected layer for prediction module.
Fig 9.
Flowchart of STF-GGRU implementation.
Table 3.
PeMSD4 traffic flow prediction dataset.
Table 4.
PeMSD8 traffic flow prediction dataset.
Table 5.
Evaluation metrics of Benchmarks and STF-GGRU on PeMSD4.
Table 6.
Evaluation metrics of Benchmarks and STF-GGRU on PeMSD8.
Fig 10.
RMSE of STFFGRU and Baselines.
Fig 11.
MAE of STFFGRU and Baselines.
Fig 12.
MAPE of STFFGRU and Baseline.
Fig 13.
Prediction for STF-GGRU and benchmark on PeMSD4.
Fig 14.
Prediction for STF-GGRU and benchmark on PeMSD8.
Fig 15.
Performance for STF-GGRU and Benchmark on all sensors On PeMSD8.
Fig 16.
Performance for STF-GGRU and benchmark on all sensors On PeMSD8.
Table 7.
Scalability performance of the proposed model.
Fig 17.
Traffic flow prediction results for Sensor 16 on the PeMSD4.
Fig 18.
Traffic flow prediction results for Sensor 250 on the PeMSD4.
Fig 19.
Traffic flow prediction results for Sensor 14 on the PeMSD8.
Fig 20.
Traffic flow prediction results for Sensor 105 on the PeMSD8.
Fig 21.
Prediction result for Sensor 250 on the PeMSD4 dataset during the 15:00–16:00 rush hour.
Fig 22.
Prediction results for Sensor 160 on the PeMSD4 dataset during the rush hour period.
Fig 23.
Prediction results for Sensor 14 on the PeMSD8 dataset during the rush hour.
Fig 24.
Prediction result for Sensor 105 on the PeMSD8 dataset during the 14:00–15:00 period.
Fig 25.
The highest traffic data PeMSD4 on 2-9-2018 (rush hour).
Fig 26.
Model performance without Temporal Module.
Fig 27.
Model performance without Spatial Module.
Fig 28.
Model performance without ISTFA.
Fig 29.
Model performance without CKA Module.
Fig 30.
Model performance without DKNN Module.