DSTAGNN: Dynamic Spatial-Temporal Aware Graph Neural Network for Traffic Flow Forecasting

1. Quang-Huy Tran
Network Science Lab
Dept. of Artificial Intelligence
The Catholic University of Korea
E-mail: huytran1126@gmail.com
2024-05-20
DSTAGNN: Dynamic Spatial-Temporal
Aware Graph Neural Network for Traffic
Flow Forecasting
Shiyong Lan et al.
ICML: 2022 International conference on machine learning

2. 2
OUTLINE
• MOTIVATION
• INTRODUCTION
• METHODOLOGY
• EXPERIMENT & RESULT
• CONCLUSION

3. 3
MOTIVATION
• Due to the presence of complex dynamic spatial-temporal dependencies within a
road network, achieving highly accurate traffic flow prediction is a challenging task.
Overview

4. 4
MOTIVATION
• Spatially similar urban functional areas in the road network have remarkably similar
traffic flow patterns even if they are far away, which demands simultaneously
capturing wide-scale local and global spatial relevance.
• The interweaving effect of long-term dynamic similar patterns and short-term
random irregular patterns in the time dimension is bound to require adaptively
focusing on temporal dependence in a wide range.
Overview

5. 5
INTRODUCTION
• A novel graph to capture dynamic attributes of spatial association among nodes
• by mining from their historic traffic flow data directly, without using a predefined static adjacency
matrix.
• A new spatial-temporal attention module to exploit the dynamic spatial correlation
within multi-scale neighborhoods:
o multi-order Chebyshev polynomials in GCN.
o the wide range of temporal dependency is exploited by the multi-head self-attention.
• An improved gated convolution module, which can further enhance the awareness of
the model to dynamic temporal dependency within the road network.

6. 6
METHODOLOGY
PROBLEM SETTING
• For ST graph forecasting tasks, our goal is aim to predict the traffic volume by learning
a function 𝑓(·) for approximating the true mapping of historical observed data to the
future data 𝑇:
• Given a ST graph 𝐺𝑆𝑇 = (𝑉, ℰ, 𝐴, 𝑋).
• 𝑉: set of 𝑁 nodes, ℰ is set of edges, 𝐴 is binary adjacency matrix, and 𝑋 is node
features matrix.

7. 7
METHODOLOGY
Spatial-Temporal Aware Graph Construction
• Wasserstein Distance.

8. 8
METHODOLOGY
Spatial-Temporal Aware Graph Construction

9. 9
METHODOLOGY
Overall Architecture

10. 10
METHODOLOGY
Spatial-Temporal Attention Block
• Temporal attention:
• Spatial attention:

11. 11
METHODOLOGY
Spatial-Temporal Convolution Block
• Spatial graph convolution:
• Temporal gated convolution:

12. 12
EXPERIMENT AND RESULT
EXPERIMENT - BASELINES
• Measurement:
o Mean absolute error (MAE).
o Mean absolute percentage error (MAPE).
o Root mean square error (RMSE).
• Dataset: PEMS03, PEMS04, PEMS07 and PEMS08.
o Road traffic datasets from California.

13. 13
EXPERIMENT AND RESULT
EXPERIMENT – BASELINE
[1] Sutskever, I., Vinyals, O., & Le, Q. V. (2014). Sequence to sequence learning with neural networks. Advances in neural information processing systems, 27.
[2] Bai, S., Kolter, J. Z., & Koltun, V. (2018). An empirical evaluation of generic convolutional and recurrent networks for sequence modeling. arXiv preprint arXiv:1803.01271.
[3] Li, Y., Yu, R., Shahabi, C., & Liu, Y. (2017). Diffusion convolutional recurrent neural network: Data-driven traffic forecasting. arXiv preprint arXiv:1707.01926.
[4] Yu, B., Yin, H., & Zhu, Z. (2017). Spatio-temporal graph convolutional networks: A deep learning framework for traffic forecasting. arXiv preprint arXiv:1709.04875.
[5] Guo, S., Lin, Y., Feng, N., Song, C., & Wan, H. (2019, July). Attention based spatial-temporal graph convolutional networks for traffic flow forecasting. In Proceedings of the AAAI conference on artificial intelligence (Vol. 33, No. 01, pp. 922-929).
[6] Song, C., Lin, Y., Guo, S., & Wan, H. (2020, April). Spatial-temporal synchronous graph convolutional networks: A new framework for spatial-temporal network data forecasting. In Proceedings of the AAAI conference on artificial intelligence (Vol. 34, No. 01, pp. 914-921).
[7] Li, M., & Zhu, Z. (2021, May). Spatial-temporal fusion graph neural networks for traffic flow forecasting. In Proceedings of the AAAI conference on artificial intelligence (Vol. 35, No. 5, pp. 4189-4196).
[8] Fang, Z., Long, Q., Song, G., & Xie, K. (2021, August). Spatial-temporal graph ode networks for traffic flow forecasting. In Proceedings of the 27th ACM SIGKDD conference on knowledge discovery & data mining (pp. 364-373).
[9] Chen, Y., Segovia, I., & Gel, Y. R. (2021, July). Z-GCNETs: Time zigzags at graph convolutional networks for time series forecasting. In International Conference on Machine Learning (pp. 1684-1694). PMLR.
[10] Bai, L., Yao, L., Li, C., Wang, X., & Wang, C. (2020). Adaptive graph convolutional recurrent network for traffic forecasting. Advances in neural information processing systems, 33, 17804-17815.
o FC-LSTM [1].
o TCN [2]: 1D dilated convolution network-based temporal convolution network.
o DCRNN [3]: integrated graph convolution into a gated recurrent unit.
o STGCN [4]: integrated graph convolution into a 1D convolution unit.
o ASTGCN [5]: a spatial-temporal attention mechanism.
o STSGCN [6]: local spatial-temporal subgraph modules.
o STFGNN [7]: used a spatial temporal fusion graph to complement the spatial correlation.
o STGODE [8]: applied continuous graph neural network to traffic prediction in multivariate time series
forecasting
o Z-GCNETs [9]: zigzag persistence into time aware graph convolutional network for time series
prediction.
o AGCRN [10]: exploited learnable embedding of nodes in graph convolution.

14. 14
EXPERIMENT AND RESULT
RESULT – Overall Performance

15. 15
EXPERIMENT AND RESULT
RESULT – Overall Performance

16. 16
EXPERIMENT AND RESULT
RESULT – Overall Performance

17. 17
CONCLUSION
• Presented a novel deep learning framework DSTAGNN for traffic flow prediction.
o utilized spatial-temporal aware distance (STAD) derived from historic traffic data without relying on
a predefined static adjacency matrix.
o graph convolution operated on the Spatial-Temporal Aware Graph (STAG) generated from STAD can
reduce the dependency on prior information of the road network.
o combination of spatial-temporal attention module and multi-receptive field gated convolution,
DSTAGNN further boosts the awareness of dynamic spatial-temporal dependency in time series
data.