Scholarlink

<p>Traffic flow prediction faces significant challenges in effectively capturing dynamic traffic patterns and their time-varying features. Although existing deep learning models have achieved some results, accurate traffic prediction still remains a complex task. Existing methods are often limited by their focus on discrete traffic data, ignoring key information such as continuous trend changes, which can hinder the prediction accuracy. To address these issues, this study proposes a trend-aware attention Spatio-temporal Graph Convolutional Network (TA-STGCN) for traffic flow prediction. It designs the trend-aware attention mechanism to effectively capture local traffic flow variations and develops a dual-channel attention module to comprehensively capture the temporal correlation of traffic flow. In addition, trend information is introduced to optimize the learning of dynamic spatio-temporal graphs, while a multi-graph convolutional network is used to encode both the dynamic spatial information and trend data of traffic nodes, thus modeling complex spatio-temporal dependencies. Evaluation on four real-world datasets confirms that our proposed TA-STGCN achieves superior prediction accuracy compared to current baseline methods.</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

TA-STGCN: trend-aware attention spatio-temporal graph convolutional network for traffic flow prediction

  • 期刊论文

  • 发表日期：

    2026年2月5日

查看全文
  • Guihui Chen,
  • Lihui Chen,
  • Zhongbing Li,
  • Yuli Wei,
  • Zhili He,
  • Fengrui Yang

摘要

Traffic flow prediction faces significant challenges in effectively capturing dynamic traffic patterns and their time-varying features. Although existing deep learning models have achieved some results, accurate traffic prediction still remains a complex task. Existing methods are often limited by their focus on discrete traffic data, ignoring key information such as continuous trend changes, which can hinder the prediction accuracy. To address these issues, this study proposes a trend-aware attention Spatio-temporal Graph Convolutional Network (TA-STGCN) for traffic flow prediction. It designs the trend-aware attention mechanism to effectively capture local traffic flow variations and develops a dual-channel attention module to comprehensively capture the temporal correlation of traffic flow. In addition, trend information is introduced to optimize the learning of dynamic spatio-temporal graphs, while a multi-graph convolutional network is used to encode both the dynamic spatial information and trend data of traffic nodes, thus modeling complex spatio-temporal dependencies. Evaluation on four real-world datasets confirms that our proposed TA-STGCN achieves superior prediction accuracy compared to current baseline methods.