Scholarlink

<p>Accurate forecasting of container throughput volume (CTV) is essential for inland waterway transportation (IWT), especially for container-on-barge (IWT) operations that reduce congestion and improve freight sustainability. The Rhine River serves as a major IWT corridor in Northwest Europe, connecting inland ports in Belgium, Germany, the Netherlands, and France. This study develops a hybrid model to forecast quarterly CTV across the four Rhine-connected European Union (IWT) countries supporting IWT planning and operations. Quarterly CTV data in Twenty-foot Equivalent Units were collected from Eurostat, covering both Rhine and non-Rhine inland port networks. A hybrid ensemble forecasting model was constructed by integrating Extreme Gradient Boosting (XGBoost), Gated Recurrent Unit, and Long Short-Term Memory models with a Support Vector Regression meta-learner. Random forest was used for feature selection, and Optuna was applied for hyperparameter tuning. The models were trained on historical data and generated forecasts from the third quarter of 2021 through early 2024, followed by an extended evaluation covering mid-2024 to early 2025. Computations were executed using tensor processing unit-enabled high-performance computing resources to handle training complexity. The hybrid model outperformed all individual learners, achieving mean absolute percentage error values of 2.41%, 1.60%, 2.02%, and 2.36% for Belgium, Germany, the Netherlands, and France, respectively. These results show the model’s ability to learn nonlinear patterns that traditional statistical approaches, including autoregressive integrated moving average and exponential smoothing, often fail to capture. The proposed approach remains dependent on historical CTV data and selected external features, which may affect its robustness under changing economic or operational conditions. This study proposes a hybrid ensemble method that improves multi-country CTV forecasting accuracy, strengthening planning and operational decision-making in the Rhine-connected IWT network.</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

Forecasting container throughput in Northwest Europe inland waterways: a data-driven approach

  • 期刊论文

  • 发表日期：

    2025年12月4日

查看全文
  • Md Mehedi Hasan,
  • Senjuti Rahman,
  • Fan Bu

摘要

Accurate forecasting of container throughput volume (CTV) is essential for inland waterway transportation (IWT), especially for container-on-barge (IWT) operations that reduce congestion and improve freight sustainability. The Rhine River serves as a major IWT corridor in Northwest Europe, connecting inland ports in Belgium, Germany, the Netherlands, and France. This study develops a hybrid model to forecast quarterly CTV across the four Rhine-connected European Union (IWT) countries supporting IWT planning and operations. Quarterly CTV data in Twenty-foot Equivalent Units were collected from Eurostat, covering both Rhine and non-Rhine inland port networks. A hybrid ensemble forecasting model was constructed by integrating Extreme Gradient Boosting (XGBoost), Gated Recurrent Unit, and Long Short-Term Memory models with a Support Vector Regression meta-learner. Random forest was used for feature selection, and Optuna was applied for hyperparameter tuning. The models were trained on historical data and generated forecasts from the third quarter of 2021 through early 2024, followed by an extended evaluation covering mid-2024 to early 2025. Computations were executed using tensor processing unit-enabled high-performance computing resources to handle training complexity. The hybrid model outperformed all individual learners, achieving mean absolute percentage error values of 2.41%, 1.60%, 2.02%, and 2.36% for Belgium, Germany, the Netherlands, and France, respectively. These results show the model’s ability to learn nonlinear patterns that traditional statistical approaches, including autoregressive integrated moving average and exponential smoothing, often fail to capture. The proposed approach remains dependent on historical CTV data and selected external features, which may affect its robustness under changing economic or operational conditions. This study proposes a hybrid ensemble method that improves multi-country CTV forecasting accuracy, strengthening planning and operational decision-making in the Rhine-connected IWT network.