Scholarlink

<p>In recent years, Graph Convolutional Networks (GCN) have gradually gained attention in the field of traffic prediction. However, existing methods have certain limitations in capturing dynamic spatial correlations and flexibly modeling multi-granularity temporal features. To address these issues, this paper proposes a traffic flow prediction model based on the fusion of dynamic graph convolution and gating mechanisms (DGC-MGGF) to tackle the complex spatio-temporal dependency modeling problems in urban traffic flow prediction. The model captures the dynamic spatial correlations of traffic flow in real time through a dynamic graph convolution mechanism. It also effectively models the multi-granularity temporal features of traffic data by utilizing a dilated convolutional network (D2G-TCN) combined with gating mechanisms. Moreover, the model enhances the representation of complex road network relationships in the spatial dimension by fusing the first-order and second-order neighborhood information of nodes through an enhanced graph convolution module. Experiments on real-world traffic datasets (PeMS03, PeMS04, PeMS07, PeMS08) demonstrate that, compared with mainstream traffic flow prediction models, the MAE is reduced by 0.34−21.02% and the RMSE is reduced by 0.37−18.99%. The experimental results validate the effectiveness of the proposed model.</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

DGC-MGGF: Dynamic Graph Convolution and Multi-granular Gated Fusion for Traffic Flow Forecasting

  • 期刊论文

  • 发表日期：

    2026年5月27日

查看全文
  • Jun li Liu,
  • Dongyan Li,
  • Can Cui,
  • Chuang Wang,
  • Penghe Huang,
  • Yuanyuan Li,
  • Wenjie Zhou

摘要

In recent years, Graph Convolutional Networks (GCN) have gradually gained attention in the field of traffic prediction. However, existing methods have certain limitations in capturing dynamic spatial correlations and flexibly modeling multi-granularity temporal features. To address these issues, this paper proposes a traffic flow prediction model based on the fusion of dynamic graph convolution and gating mechanisms (DGC-MGGF) to tackle the complex spatio-temporal dependency modeling problems in urban traffic flow prediction. The model captures the dynamic spatial correlations of traffic flow in real time through a dynamic graph convolution mechanism. It also effectively models the multi-granularity temporal features of traffic data by utilizing a dilated convolutional network (D2G-TCN) combined with gating mechanisms. Moreover, the model enhances the representation of complex road network relationships in the spatial dimension by fusing the first-order and second-order neighborhood information of nodes through an enhanced graph convolution module. Experiments on real-world traffic datasets (PeMS03, PeMS04, PeMS07, PeMS08) demonstrate that, compared with mainstream traffic flow prediction models, the MAE is reduced by 0.34−21.02% and the RMSE is reduced by 0.37−18.99%. The experimental results validate the effectiveness of the proposed model.