Scholarlink

<p>Global trade is mainly conducted via maritime transportation; however, exhaust emissions from vessel diesel engines significantly contribute to air pollution and global warming. Existing emission inventory method, such as bottom–up or top–down approaches proposed by the International Maritime Organization and European Environment Agency, are limited in reflecting dynamic characteristics of actual engine operation. This study proposes a hybrid prediction model that integrates a time-series forecasting transformer and XGBoost using real-time engine operational data. Key influencing variables are selected through least absolute shrinkage and selection operator regression. The proposed model enhances emission prediction accuracy across various operating conditions. Compared to conventional methods, root mean square error and mean absolute error were reduced by 34% and 40% for CO<sub>2</sub>, 45 and 47% for CO, and 40 and 48% for NO<sub>X</sub>. The model maintained stable performance for highly variable pollutants, with approximately 85% of CO<sub>2</sub> predictions within a ± 5% error range. Some limitations remain due to the variability of CO and nonlinear effect from controllable pitch propeller. This study provides a scalable, real-time framework for predicting vessel emissions, enabling improved air pollution monitoring and supporting emission reduction strategies in the maritime sector.</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

Real-time emission prediction for ship engine using stacked time-series learning: a transformer-XGBoost hybrid framework

  • 期刊论文

  • 出版模式：

    开放获取

  • 发表日期：

    2025年11月18日

查看全文
  • Seunghun Lim,
  • Jungmo Oh

摘要

Global trade is mainly conducted via maritime transportation; however, exhaust emissions from vessel diesel engines significantly contribute to air pollution and global warming. Existing emission inventory method, such as bottom–up or top–down approaches proposed by the International Maritime Organization and European Environment Agency, are limited in reflecting dynamic characteristics of actual engine operation. This study proposes a hybrid prediction model that integrates a time-series forecasting transformer and XGBoost using real-time engine operational data. Key influencing variables are selected through least absolute shrinkage and selection operator regression. The proposed model enhances emission prediction accuracy across various operating conditions. Compared to conventional methods, root mean square error and mean absolute error were reduced by 34% and 40% for CO2, 45 and 47% for CO, and 40 and 48% for NOX. The model maintained stable performance for highly variable pollutants, with approximately 85% of CO2 predictions within a ± 5% error range. Some limitations remain due to the variability of CO and nonlinear effect from controllable pitch propeller. This study provides a scalable, real-time framework for predicting vessel emissions, enabling improved air pollution monitoring and supporting emission reduction strategies in the maritime sector.