Answers are grounded in this paper and cite their sources.
Evaluates traffic forecasting using TEDDGN in urban planning, suggesting improved prediction performance.
Accurate traffic flow prediction is of great significance for urban planning and traffic management. Among existing methods, researchers have shown remarkable progress by utilizing spatiotemporal networks. Unfortunately, most of these methods ignore the heterogeneity of multi-scale features in traffic data, that is, the difference between trend components and event components, and modeling them together easily causes feature interference and affects prediction performance. In addition, without relying on prior knowledge, previous methods have also struggled to effectively model the dynamically changing spatial dependencies among road networks. To address the above issues, we introduce the trend-event decoupled dynamic graph network (TEDDGN), a novel model for traffic forecasting. TEDDGN first uses a decoupling module to separate traffic sequences into trend and event components, and then employs a multi-scale temporal learner to extract temporal patterns from each component. In the spatial dimension, TEDDGN designs the spatial feature extraction module that integrates data-driven dynamic graph generation methods to learn dynamic spatial dependencies, while introducing adaptive graph structures to supplement potential static spatial associations, thereby characterizing spatial relationships in the traffic network more comprehensively. Extensive experiments on three real-world traffic datasets show that TEDDGN outperforms state-of-the-art baselines across multiple evaluation metrics.
No takes yet. Share an insight, caveat, or question.
Zhang et al. (2026) studied this question.
Synapse has enriched 10 related papers. Consider them for comparative context: