Scholarlink

<p>Traffic flow forecasting is a key task in Intelligent Transportation Systems (ITSs), playing an important role in congestion mitigation, traffic management, and real-time traffic scheduling. With the rapid growth of urban traffic data, modern traffic prediction systems increasingly require high-performance computing (HPC), real-time inference capability, and efficient parallel processing. However, existing methods often model traffic networks as static graph structures, making it difficult to capture the complex and dynamic spatiotemporal dependencies of traffic flow. To address these limitations, this paper proposes a novel traffic forecasting framework, FusionGraphSAGE with Neural Networks (FGSNN). The proposed model introduces a Dynamic Multi-graph Fusion Mechanism (DMFM), which combines a predefined static adjacency matrix based on node distances with a dynamic adaptive graph to better capture evolving traffic relationships. In addition, a FusionGraphSAGE module is designed to effectively learn spatial dependencies among traffic nodes. By integrating a Spatiotemporal Synchronization Coupling Layer (STCLayer), temporal attention mechanisms, and GraphSAGE, the model achieves collaborative spatiotemporal feature learning and refined traffic flow prediction. Furthermore, FGSNN is designed with strong computational efficiency and scalability, enabling efficient training and real-time forecasting on large-scale traffic datasets, making it suitable for HPC environments and real-time ITS applications. Experimental results on two real-world datasets demonstrate that FGSNN consistently outperforms existing baseline models in both prediction accuracy and robustness.</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

FGSNN: a FusionGraphSAGE with neural networks for traffic flow prediction

  • 期刊论文

  • 发表日期：

    2026年7月6日

查看全文
  • Yanan Chen,
  • Yongmei Ma,
  • Shuchao Wang,
  • Yunbiao Wu,
  • Yong Zhou,
  • Zhixin Chen,
  • Weicai Peng

摘要

Traffic flow forecasting is a key task in Intelligent Transportation Systems (ITSs), playing an important role in congestion mitigation, traffic management, and real-time traffic scheduling. With the rapid growth of urban traffic data, modern traffic prediction systems increasingly require high-performance computing (HPC), real-time inference capability, and efficient parallel processing. However, existing methods often model traffic networks as static graph structures, making it difficult to capture the complex and dynamic spatiotemporal dependencies of traffic flow. To address these limitations, this paper proposes a novel traffic forecasting framework, FusionGraphSAGE with Neural Networks (FGSNN). The proposed model introduces a Dynamic Multi-graph Fusion Mechanism (DMFM), which combines a predefined static adjacency matrix based on node distances with a dynamic adaptive graph to better capture evolving traffic relationships. In addition, a FusionGraphSAGE module is designed to effectively learn spatial dependencies among traffic nodes. By integrating a Spatiotemporal Synchronization Coupling Layer (STCLayer), temporal attention mechanisms, and GraphSAGE, the model achieves collaborative spatiotemporal feature learning and refined traffic flow prediction. Furthermore, FGSNN is designed with strong computational efficiency and scalability, enabling efficient training and real-time forecasting on large-scale traffic datasets, making it suitable for HPC environments and real-time ITS applications. Experimental results on two real-world datasets demonstrate that FGSNN consistently outperforms existing baseline models in both prediction accuracy and robustness.