DST-GTN: Dynamic Spatio-Temporal Graph Transformer Network for Traffic Forecasting

TL;DR

This paper tackles the challenge of traffic forecasting under time-varying spatial relationships by introducing Dynamic Spatio-Temporal (Dyn-ST) features and a DST-GTN model. DST-GTN first extracts temporal dependencies with a Temporal Transformer, then leverages a Dyn-ST Embedding and a Dynamic Spatio-Temporal Module (DSTM) composed of the Dynamic Spatio-Temporal Graph Generator (DSTGG) and Node Frequency Learning STGCN (NFL-STGCN) to uncover intricate ST dynamics, and finally projects forecasts via an output MLP. The approach yields state-of-the-art results on five public real-world datasets, with improved robustness and efficiency, demonstrating the practical value for urban traffic management. By explicitly modeling time-varying spatial patterns and balancing local/global information through adaptive filters, DST-GTN provides a principled framework for accurate and stable traffic forecasting in dynamic networks.

Abstract

Accurate traffic forecasting is essential for effective urban planning and congestion management. Deep learning (DL) approaches have gained colossal success in traffic forecasting but still face challenges in capturing the intricacies of traffic dynamics. In this paper, we identify and address this challenges by emphasizing that spatial features are inherently dynamic and change over time. A novel in-depth feature representation, called Dynamic Spatio-Temporal (Dyn-ST) features, is introduced, which encapsulates spatial characteristics across varying times. Moreover, a Dynamic Spatio-Temporal Graph Transformer Network (DST-GTN) is proposed by capturing Dyn-ST features and other dynamic adjacency relations between intersections. The DST-GTN can model dynamic ST relationships between nodes accurately and refine the representation of global and local ST characteristics by adopting adaptive weights in low-pass and all-pass filters, enabling the extraction of Dyn-ST features from traffic time-series data. Through numerical experiments on public datasets, the DST-GTN achieves state-of-the-art performance for a range of traffic forecasting tasks and demonstrates enhanced stability.

Figures (4)

• Figure 1: An example of a traffic flow system at three different times.
• Figure 2: The detailed architecture of DST-GTN. (a) represents the overall architecture of DST-GTN. (b) describes the Temporal Transformer Module, which captures temporal dependencies in the embedded data. (c) illustrates the Dynamic Spatio-Temporal Graph Generator (DSTGG) component of the Dynamic Spatio-Temporal Module (DSTM), which utilize Dyn-ST embedding to generate a global ST graph. (d) is the NFL-STGCN component of the DSTM, which adaptively learns the local and global information demand for each ST node to adjust the ST graph. The optimized two ST graph is finally employed in the ST graph convolution process.
• Figure 3: Robustness of different models on PEMS04 and PEMS08 datasets
• Figure 4: Traffic forecasting visualization.

Theorems & Definitions (3)

• Definition 1: Traffic Network
• Definition 2: Traffic Time-Series Data
• Definition 3: Multi-step Traffic Forecasting