Meta Dynamic Graph for Traffic Flow Prediction
Yiqing Zou, Hanning Yuan, Qianyu Yang, Ziqiang Yuan, Shuliang Wang, Sijie Ruan
TL;DR
Traffic flow prediction demands capturing intricate spatio-temporal dependencies. The paper introduces MetaDG, a GCRU-based framework that dynamically generates per-time-step node embeddings, meta-parameters, and graph structures to unify dynamics and heterogeneities. It employs Spatio-Temporal Correlation Enhancement (STCE) to produce robust node representations and Dynamic Graph Qualification (DGQ) to refine edge weights, yielding a dynamic adjacency tilde A_t and meta-parameters θ_t for each time step. A MetaDG Convolutional Recurrent Unit uses these components to perform time-step graph convolutions, enabling end-to-end learning of dynamic propagation patterns. Across four real-world datasets, MetaDG consistently outperforms state-of-the-art baselines, especially for long-horizon predictions, demonstrating practical impact for accurate and robust traffic forecasting.
Abstract
Traffic flow prediction is a typical spatio-temporal prediction problem and has a wide range of applications. The core challenge lies in modeling the underlying complex spatio-temporal dependencies. Various methods have been proposed, and recent studies show that the modeling of dynamics is useful to meet the core challenge. While handling spatial dependencies and temporal dependencies using separate base model structures may hinder the modeling of spatio-temporal correlations, the modeling of dynamics can bridge this gap. Incorporating spatio-temporal heterogeneity also advances the main goal, since it can extend the parameter space and allow more flexibility. Despite these advances, two limitations persist: 1) the modeling of dynamics is often limited to the dynamics of spatial topology (e.g., adjacency matrix changes), which, however, can be extended to a broader scope; 2) the modeling of heterogeneity is often separated for spatial and temporal dimensions, but this gap can also be bridged by the modeling of dynamics. To address the above limitations, we propose a novel framework for traffic prediction, called Meta Dynamic Graph (MetaDG). MetaDG leverages dynamic graph structures of node representations to explicitly model spatio-temporal dynamics. This generates both dynamic adjacency matrices and meta-parameters, extending dynamic modeling beyond topology while unifying the capture of spatio-temporal heterogeneity into a single dimension. Extensive experiments on four real-world datasets validate the effectiveness of MetaDG.