Scholarlink

<p>Traffic prediction, as a key technology in intelligent transportation systems, presents significant challenges due to numerous influencing factors, including the complexity of traffic conditions and the rapid, dynamic temporal changes they exhibit. Although several traffic prediction models have achieved promising results, there remains potential for improvement in spatial–temporal dynamic modeling and the integration of local and global information. This paper introduces a Spatial–Temporal Multi-scale Auto-Correlation Network (STMACN) for traffic prediction. STMACN includes a spatial–temporal embedding layer, a spatial attention module, a temporal auto-correlation module, and a multi-scale attention fusion module. The spatial–temporal embedding layer integrates graph node information and temporal cycles into the input data, thereby enhancing the prior knowledge. The spatial attention module adaptively mines spatial correlations between sensor nodes in the road network. The temporal auto-correlation module models long-term temporal dependencies and captures periodic features using the auto-correlation mechanism. The multi-scale attention fusion module further enhances the temporal auto-correlation module’s ability to capture periodic features at different time scales and improves the perception of local changes. Experimental results on four real-world traffic datasets show that our proposed STMACN outperforms state-of-the-art models. The source code and data are available at <a href="https://github.com/NathanZ137/STMACN">https://github.com/NathanZ137/STMACN</a>.</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

STMACN: spatial–temporal multi-scale auto-correlation network for traffic forecasting

  • 期刊论文

  • 发表日期：

    2025年9月8日

查看全文
  • Yuling Hong,
  • Jiaqi Zhang

摘要

Traffic prediction, as a key technology in intelligent transportation systems, presents significant challenges due to numerous influencing factors, including the complexity of traffic conditions and the rapid, dynamic temporal changes they exhibit. Although several traffic prediction models have achieved promising results, there remains potential for improvement in spatial–temporal dynamic modeling and the integration of local and global information. This paper introduces a Spatial–Temporal Multi-scale Auto-Correlation Network (STMACN) for traffic prediction. STMACN includes a spatial–temporal embedding layer, a spatial attention module, a temporal auto-correlation module, and a multi-scale attention fusion module. The spatial–temporal embedding layer integrates graph node information and temporal cycles into the input data, thereby enhancing the prior knowledge. The spatial attention module adaptively mines spatial correlations between sensor nodes in the road network. The temporal auto-correlation module models long-term temporal dependencies and captures periodic features using the auto-correlation mechanism. The multi-scale attention fusion module further enhances the temporal auto-correlation module’s ability to capture periodic features at different time scales and improves the perception of local changes. Experimental results on four real-world traffic datasets show that our proposed STMACN outperforms state-of-the-art models. The source code and data are available at https://github.com/NathanZ137/STMACN.