Scholarlink

<p>To address the issue of noise in the spatial-temporal matrix of short-term traffic flow prediction models and the inability of these models to adaptively focus on the input features of traffic flow data—resulting in reduced prediction accuracy—this paper proposes a spatial-temporal fusion traffic flow prediction model. The model integrates the Mayfly Optimization Algorithm (MOA), Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Attention Mechanism, Convolutional Neural Network (CNN), and Gated Linear Unit (GLU). First, MOA is employed to optimize the parameters of CEEMDAN, which is used to decompose the original traffic flow data into multiple Intrinsic Mode Functions (IMFs) and a Residual component (Res). Then, the IMF components are combined to construct an IMF-based spatial-temporal matrix, which is input into the model to extract traffic flow features. Finally, the prediction results for each IMF component are reconstructed to achieve short-term traffic flow prediction. Experimental results demonstrate that, compared to baseline models such as ED-LSTM and CNN-LSTM, the proposed model reduces the Root Mean Square Error (RMSE) by 85.12% and 85.29%, respectively.</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

Spatial-temporal fusion short-term traffic flow prediction model based on mode decomposition algorithm

  • 期刊论文

  • 发表日期：

    2025年9月22日

查看全文
  • Xijun Zhang,
  • Ziyao Xia,
  • Hong Zhang,
  • Xianli Zhang

摘要

To address the issue of noise in the spatial-temporal matrix of short-term traffic flow prediction models and the inability of these models to adaptively focus on the input features of traffic flow data—resulting in reduced prediction accuracy—this paper proposes a spatial-temporal fusion traffic flow prediction model. The model integrates the Mayfly Optimization Algorithm (MOA), Complete Ensemble Empirical Mode Decomposition with Adaptive Noise (CEEMDAN), Attention Mechanism, Convolutional Neural Network (CNN), and Gated Linear Unit (GLU). First, MOA is employed to optimize the parameters of CEEMDAN, which is used to decompose the original traffic flow data into multiple Intrinsic Mode Functions (IMFs) and a Residual component (Res). Then, the IMF components are combined to construct an IMF-based spatial-temporal matrix, which is input into the model to extract traffic flow features. Finally, the prediction results for each IMF component are reconstructed to achieve short-term traffic flow prediction. Experimental results demonstrate that, compared to baseline models such as ED-LSTM and CNN-LSTM, the proposed model reduces the Root Mean Square Error (RMSE) by 85.12% and 85.29%, respectively.