ML Anthology

Spatial-Temporal PDE Networks for Traffic Flow Forecasting

Tianshu Bao, Hua Wei, Junyi Ji, Daniel B. Work, Taylor T. Johnson
ECML-PKDD 2024 pp. 166-182
doi:10.1007/978-3-031-70381-2_11 /ecmlpkdd/2024/bao2024ecmlpkdd-spatialtemporal/

Abstract

Spatial-temporal forecasting is crucial in various domains, including traffic flow prediction for Intelligent Transportation Systems (ITS). Despite the challenges posed by complex spatial-temporal dependencies in traffic networks, Partial Differential Equations (PDEs) have proven effective for capturing traffic dynamics. However, recent trends favor data-driven approaches like Graph Neural Networks (GNNs) for traffic forecasting, often overlooking the principles described by PDEs. In this paper, we propose a Graph Partial Differential Equation Network (GPDE) that integrates PDE principles with GNNs to enhance traffic flow forecasting. Our approach leverages dynamic graph structures based on PDE flux functions, incorporating residual connections and learnable rates for improved model performance. Extensive experiments on real-world traffic datasets demonstrate the superiority of GPDE over existing methods in both short-term and long-term traffic speed prediction tasks.

PDF ECML-PKDD Semantic Scholar

Cite

Text

Bao et al. "Spatial-Temporal PDE Networks for Traffic Flow Forecasting." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2024. doi:10.1007/978-3-031-70381-2_11

Markdown

[Bao et al. "Spatial-Temporal PDE Networks for Traffic Flow Forecasting." European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases, 2024.](https://mlanthology.org/ecmlpkdd/2024/bao2024ecmlpkdd-spatialtemporal/) doi:10.1007/978-3-031-70381-2_11

BibTeX

@inproceedings{bao2024ecmlpkdd-spatialtemporal,
  title     = {{Spatial-Temporal PDE Networks for Traffic Flow Forecasting}},
  author    = {Bao, Tianshu and Wei, Hua and Ji, Junyi and Work, Daniel B. and Johnson, Taylor T.},
  booktitle = {European Conference on Machine Learning and Principles and Practice of Knowledge Discovery in Databases},
  year      = {2024},
  pages     = {166-182},
  doi       = {10.1007/978-3-031-70381-2_11},
  url       = {https://mlanthology.org/ecmlpkdd/2024/bao2024ecmlpkdd-spatialtemporal/}
}