Scholarlink

In order to accurately predict urban traffic flow, a prediction algorithm integrating Autoregressive Integrated Moving Average (ARIMA) and Convolutional Neural Networks (CNN) is proposed to deeply capture the temporal dynamic features, explore the spatial correlation and change rule of road sections, and improve the robustness and accuracy of real-time traffic flow prediction of urban traffic road network. First, the ARIMA model was used to capture linear relations, trends, and seasonal features in traffic flow data. Next, the CNN model learned high-level feature representations, especially spatial and temporal correlations, from the raw data. Then, the outputs of ARIMA and CNN were used as features and input to the random forest model. By combining several decision trees, random forest, an integrated learning technique, can reduce the chance of overfitting and increase prediction stability. The prediction accuracy of the integrated method can be obtained through experimental validation. Compared with ARIMA and CNN algorithms, it is found that there is a large improvement in the accuracy of the integrated algorithm. The algorithm has excellent results in the time-varying urban traffic flow prediction, which is suitable for the problem of seasonal trends, and also can achieve satisfactory results for the problem of short-term fluctuation of data. After in-depth research and experimental validation, the proposed prediction algorithm integrating Autoregressive Integrated Moving Average (ARIMA) and Convolutional Neural Networks (CNN) has demonstrated significant advantages in urban traffic flow prediction. This integrated algorithm is capable of capturing linear relationships, trends, and seasonal characteristics in traffic flow data, while also learning high-level feature representations, especially spatial and temporal correlations, from the data. Compared to standalone ARIMA or CNN algorithms, the integrated algorithm has achieved significant improvements in prediction accuracy, particularly when dealing with data featuring seasonal trends and short-term fluctuations.

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

Urban Traffic Flow Prediction Based on the Integration of ARIMA and CNN

  • 会议论文

  • 发表日期：

    2025年

查看全文
  • Yulong Zhang,
  • Xin Shu,
  • Yuxin Ma

摘要

In order to accurately predict urban traffic flow, a prediction algorithm integrating Autoregressive Integrated Moving Average (ARIMA) and Convolutional Neural Networks (CNN) is proposed to deeply capture the temporal dynamic features, explore the spatial correlation and change rule of road sections, and improve the robustness and accuracy of real-time traffic flow prediction of urban traffic road network. First, the ARIMA model was used to capture linear relations, trends, and seasonal features in traffic flow data. Next, the CNN model learned high-level feature representations, especially spatial and temporal correlations, from the raw data. Then, the outputs of ARIMA and CNN were used as features and input to the random forest model. By combining several decision trees, random forest, an integrated learning technique, can reduce the chance of overfitting and increase prediction stability. The prediction accuracy of the integrated method can be obtained through experimental validation. Compared with ARIMA and CNN algorithms, it is found that there is a large improvement in the accuracy of the integrated algorithm. The algorithm has excellent results in the time-varying urban traffic flow prediction, which is suitable for the problem of seasonal trends, and also can achieve satisfactory results for the problem of short-term fluctuation of data. After in-depth research and experimental validation, the proposed prediction algorithm integrating Autoregressive Integrated Moving Average (ARIMA) and Convolutional Neural Networks (CNN) has demonstrated significant advantages in urban traffic flow prediction. This integrated algorithm is capable of capturing linear relationships, trends, and seasonal characteristics in traffic flow data, while also learning high-level feature representations, especially spatial and temporal correlations, from the data. Compared to standalone ARIMA or CNN algorithms, the integrated algorithm has achieved significant improvements in prediction accuracy, particularly when dealing with data featuring seasonal trends and short-term fluctuations.