Scholarlink

<p>The current traffic flow prediction methods suffer from data sparsity and insufficient handling of nonlinear relationships in practical prediction scenarios. Therefore, this study proposes a prediction model that integrates attention mechanisms, temporal graph convolutions, and gated recurrent units. Experimental results show that before 150 iterations, the loss function value of the research model stabilizes around 0.0025 after reaching 150 iterations. The minimum loss value of the prediction model combining time convolutional networks and bidirectional gated recurrent units on the test set is 0.0037. The overall stability of the loss function curve of the prediction model combining whale optimization algorithm and gated recurrent units is not as good as that of the research algorithm; the minimum loss value of the S model is 0.0034 at 162 iterations. Sensitivity analysis results indicate that adjusting the weights of spatiotemporal features can significantly reduce the fluctuation range of prediction errors. The research findings provide minute-level prediction support for scenarios such as dynamic traffic signal control and emergency route planning during sudden events, while also offering a technical framework with both theoretical rigor and engineering feasibility for real-time decision-making and resource optimization in smart city transportation systems.</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

Traffic Flow Prediction Based on Time-Domain Graph Convolution and GRU

  • 期刊论文

  • 发表日期：

    2025年6月11日

查看全文
  • Yi Pan,
  • Yan Hu,
  • Zhu Zhao,
  • Qing Wang

摘要

The current traffic flow prediction methods suffer from data sparsity and insufficient handling of nonlinear relationships in practical prediction scenarios. Therefore, this study proposes a prediction model that integrates attention mechanisms, temporal graph convolutions, and gated recurrent units. Experimental results show that before 150 iterations, the loss function value of the research model stabilizes around 0.0025 after reaching 150 iterations. The minimum loss value of the prediction model combining time convolutional networks and bidirectional gated recurrent units on the test set is 0.0037. The overall stability of the loss function curve of the prediction model combining whale optimization algorithm and gated recurrent units is not as good as that of the research algorithm; the minimum loss value of the S model is 0.0034 at 162 iterations. Sensitivity analysis results indicate that adjusting the weights of spatiotemporal features can significantly reduce the fluctuation range of prediction errors. The research findings provide minute-level prediction support for scenarios such as dynamic traffic signal control and emergency route planning during sudden events, while also offering a technical framework with both theoretical rigor and engineering feasibility for real-time decision-making and resource optimization in smart city transportation systems.