Scholarlink

Amid global efforts to improve energy efficiency and reduce emissions in the shipping industry, forecasting ship energy consumption has emerged as a key direction for green and intelligent shipping. Existing research has largely focused on point forecasts, which struggle to capture the uncertainties inherent in real-world environments. This paper develops a prediction-interval framework based on the LightGBM (LGBM) model and systematically compares several methods, including Quantile Regression (QR), Inductive Conformal Prediction (ICP), Jackknife+, Bootstrap, and Kernel Density Estimation (KDE). The experimental results indicate that although QR and ICP can achieve nominal coverage, they tend to produce overly wide intervals or exhibit stability issues. Bootstrap and KDE are intuitive and straightforward, but their robustness degrades under complex operating conditions. In contrast, Jackknife+ achieves a better balance between coverage and interval width, yielding compact and stable prediction intervals. Its comprehensive index (AIS) is 16.60, approximately 20.5% lower than QR and 5.1% lower than ICP, and it also outperforms Bootstrap and KDE. Overall, Jackknife+ significantly enhances the reliability and practical utility of energy-consumption forecasts, providing more valuable guidance for ship energy-efficiency management and operational decision making.

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

Ship Energy Consumption Interval Prediction Using LightGBM and Jackknife+

  • 会议论文

  • 发表日期：

    2026年

查看全文
  • Yuanhe Wu,
  • Jiale Li,
  • Zhihui Hu,
  • Zhiheng Lin,
  • Shiqi Li,
  • Yunfei Guo

摘要

Amid global efforts to improve energy efficiency and reduce emissions in the shipping industry, forecasting ship energy consumption has emerged as a key direction for green and intelligent shipping. Existing research has largely focused on point forecasts, which struggle to capture the uncertainties inherent in real-world environments. This paper develops a prediction-interval framework based on the LightGBM (LGBM) model and systematically compares several methods, including Quantile Regression (QR), Inductive Conformal Prediction (ICP), Jackknife+, Bootstrap, and Kernel Density Estimation (KDE). The experimental results indicate that although QR and ICP can achieve nominal coverage, they tend to produce overly wide intervals or exhibit stability issues. Bootstrap and KDE are intuitive and straightforward, but their robustness degrades under complex operating conditions. In contrast, Jackknife+ achieves a better balance between coverage and interval width, yielding compact and stable prediction intervals. Its comprehensive index (AIS) is 16.60, approximately 20.5% lower than QR and 5.1% lower than ICP, and it also outperforms Bootstrap and KDE. Overall, Jackknife+ significantly enhances the reliability and practical utility of energy-consumption forecasts, providing more valuable guidance for ship energy-efficiency management and operational decision making.