TITLE:
Predicting Ship Propeller Speed with Multi-Source Data Fusion and Physics-Informed LightGBM: A Novel Correction Framework
AUTHORS:
Min Chen, Yingchao Gou, Feiyang Ren
KEYWORDS:
Ship RPM Prediction, Physics-Informed LightGBM, Multi-Source Data Fusion
JOURNAL NAME:
Journal of Data Analysis and Information Processing,
Vol.13 No.4,
September
24,
2025
ABSTRACT: Accurate prediction of main-engine rotational speed (RPM) is pivotal for energy-efficient ship operation and compliance with emerging carbon-intensity regulations. Existing approaches either rely on computationally intensive physics-based models or data-driven methods that neglect hydrodynamic constraints and suffer from label noise in mandatory reporting data. We propose a physics-informed LightGBM framework that fuses high-resolution AIS trajectories, meteorological re-analyses and EU MRV logs through a temporally anchored, multi-source alignment protocol. A dual LightGBM ensemble (L1/L2) predicts RPM under laden and ballast conditions. Validation on a Panamax tanker (366 days) yields −1.52 rpm (−3%) error; ballast accuracy surpasses laden by 1.7%.