Scholarlink

<p>Traffic flow prediction is a fundamental task in spatial-temporal forecasting; however, it remains highly challenging due to the intricate interdependencies between spatial and temporal dynamics. Existing graph convolutional network (GNN)-based methods face two key limitations: (1) the assumption of static graph structures limits their ability to model dynamic spatial-temporal heterogeneity; and (2) the reliance on discrete time-slice processing hinders the capture of continuous traffic dynamics. To address these challenges, we propose Adaptive Spatial-temporal Graph ODE Networks (ASTGODE), which leverage Neural ODEs to reformulate discrete spatial-temporal convolutions into a continuous modeling paradigm. ASTGODE further incorporates multi-modal graph convolutions to capture heterogeneous spatial dependencies, and a dynamic spatial-temporal adaptation module to address feature interaction heterogeneity. Extensive experiments on real-world traffic datasets demonstrate consistent improvements over existing baselines, validating its enhanced capability in spatial-temporal representation learning.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Adaptive spatial-temporal graph ODE networks for traffic flow forecasting

  • 期刊论文

  • 发表日期：

    2026年3月9日

查看全文
  • Shixiang Han,
  • Xu Wang,
  • Yi Jin,
  • Songhe Feng,
  • Congyan Lang,
  • Yidong Li

摘要

Traffic flow prediction is a fundamental task in spatial-temporal forecasting; however, it remains highly challenging due to the intricate interdependencies between spatial and temporal dynamics. Existing graph convolutional network (GNN)-based methods face two key limitations: (1) the assumption of static graph structures limits their ability to model dynamic spatial-temporal heterogeneity; and (2) the reliance on discrete time-slice processing hinders the capture of continuous traffic dynamics. To address these challenges, we propose Adaptive Spatial-temporal Graph ODE Networks (ASTGODE), which leverage Neural ODEs to reformulate discrete spatial-temporal convolutions into a continuous modeling paradigm. ASTGODE further incorporates multi-modal graph convolutions to capture heterogeneous spatial dependencies, and a dynamic spatial-temporal adaptation module to address feature interaction heterogeneity. Extensive experiments on real-world traffic datasets demonstrate consistent improvements over existing baselines, validating its enhanced capability in spatial-temporal representation learning.