Scholarlink

<p>Accurate and reliable traffic flow prediction is fundamental to enabling proactive traffic management and intelligent expressway operations. However, the strong nonlinearities and evolving spatiotemporal dependencies in highway traffic data make it challenging for traditional methods and many existing deep learning models to capture dynamic spatial correlations and heterogeneous temporal patterns. To address these challenges, this study proposes a traffic flow prediction model named SCorr-Times-STGCN, which integrates dynamic correlation learning with multi-scale temporal modeling. The proposed framework consists of two key components: a Dynamic Spatial Correlation Block (SCorr-Block) and a Multi-Scale Temporal Modeling Block (Times-Block). The SCorr-Block adaptively constructs an adjacency matrix based on the mutual information between node historical features, thereby capturing dynamic spatial dependencies. The Times-Block first performs frequency-domain analysis via FFT to identify dominant periodic components, then applies softmax-weighted aggregation and multi-scale convolutional operations to simultaneously capture short-term fluctuations and long-term periodic patterns in traffic flow. Finally, a fully connected layer outputs the predicted values. Experimental validation was conducted using real-world expressway traffic datasets from California, USA. The results demonstrate that SCorr-Times-STGCN effectively enhances both short-term and long-term forecasting accuracy, offering strong potential for deployment in intelligent transportation systems and real-time traffic management applications.</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

Highway Traffic Flow Prediction Model Integrating Dynamic Correlations and Multi-Scale Temporal Modeling

  • 期刊论文

  • 发表日期：

    2026年5月7日

查看全文
  • Yanguo Huang,
  • Zijie Zhang,
  • Chao Liu,
  • Zhenyu Zhong,
  • Weifeng Liu

摘要

Accurate and reliable traffic flow prediction is fundamental to enabling proactive traffic management and intelligent expressway operations. However, the strong nonlinearities and evolving spatiotemporal dependencies in highway traffic data make it challenging for traditional methods and many existing deep learning models to capture dynamic spatial correlations and heterogeneous temporal patterns. To address these challenges, this study proposes a traffic flow prediction model named SCorr-Times-STGCN, which integrates dynamic correlation learning with multi-scale temporal modeling. The proposed framework consists of two key components: a Dynamic Spatial Correlation Block (SCorr-Block) and a Multi-Scale Temporal Modeling Block (Times-Block). The SCorr-Block adaptively constructs an adjacency matrix based on the mutual information between node historical features, thereby capturing dynamic spatial dependencies. The Times-Block first performs frequency-domain analysis via FFT to identify dominant periodic components, then applies softmax-weighted aggregation and multi-scale convolutional operations to simultaneously capture short-term fluctuations and long-term periodic patterns in traffic flow. Finally, a fully connected layer outputs the predicted values. Experimental validation was conducted using real-world expressway traffic datasets from California, USA. The results demonstrate that SCorr-Times-STGCN effectively enhances both short-term and long-term forecasting accuracy, offering strong potential for deployment in intelligent transportation systems and real-time traffic management applications.