FasterSTS: A Faster Spatio-Temporal Synchronous Graph Convolutional Networks for Traffic flow Forecasting

TL;DR

FasterSTS tackles the computational bottlenecks of spatio-temporal graph neural networks for traffic forecasting by introducing a fast graph computation paradigm and a spatio-temporal synchronous convolution kernel. It decomposes graph operations into node information aggregation and node representation projection, achieving $O(nN)$ and linear-like scalability, while modeling both static and dynamic temporal correlations through adaptive, per-dimension graphs and a reinforcement-learning-inspired temporal embedding. The approach integrates these components in stacked STSGCLs with residual connections and a fusion layer, and optimizes with MAE loss to handle data irregularities. Empirical results on four PeMS datasets show FasterSTS achieving state-of-the-art accuracy with significantly reduced computation time and memory, enabling more practical real-time traffic forecasting without relying on fused spatio-temporal graphs or heavy attention mechanisms.

Abstract

Accurate traffic flow prediction heavily relies on the spatio-temporal correlation of traffic flow data. Most current studies separately capture correlations in spatial and temporal dimensions, making it difficult to capture complex spatio-temporal heterogeneity, and often at the expense of increasing model complexity to improve prediction accuracy. Although there have been groundbreaking attempts in the field of spatio-temporal synchronous modeling, significant limitations remain in terms of performance and complexity control.This study proposes a quicker and more effective spatio-temporal synchronous traffic flow forecast model to address these issues.

Figures (3)

• Figure 1: The process of graph computation with different methods
• Figure 2: The architecture of FasterSTS
• Figure 3: The computation process of the spatio-temporal synchronous graph convolution kernel