Research on Ship Main Engine Fuel Consumption Model With Data Integration and Noise Cleaning

Abstract

Shipping is a crucial component of international trade, but the climate impact of greenhouse gas emissions from the shipping industry has received widespread attention. Reliable fuel consumption prediction models are essential for optimizing speed and routes. This study proposes and validates a methodology applied to a VLCC that integrates meteorological data with operational data from the vessel while effectively cleaning noise data. Operational data of the vessel is collected via sensors and extensively analyzed. The collected operational data is processed to obtain pre-cleaning data by using the conventional data pre-cleaning method, followed by further data processing using the proposed method to obtain integrated noise-cleaned data. Data-driven fuel consumption models are developed using ANA, SVR, LR, LGBM, RF, GBDT, and XGBoost. Hyperparameter tuning is conducted using random search and k-fold cross-validation, and detailed fuel consumption prediction and analysis are performed. The findings demonstrate strong generalization capabilities of the models, as indicated by small standard deviations across validation sets. The data integration and noise cleaning method significantly enhanced data quality, yielding notable performance improvements: average reductions in MAE and MAPE for predicted SOG were 0.06 and 0.536%, respectively, with an R2 increase of 0.044. For predicted FC, the average reductions in MAE and MAPE were 8.967 and 3.521%, respectively, while R2 improved by 0.114. The SOTA models including RF, GBDT, and XGBoost demonstrated commendable performance. It is concluded that subject to further testing and validation the method holds the potential to develop into a decision-support tool for shipping companies engaged in the operation of ocean-going vessels.

Keywords

• Data integration
• noise cleaning
• data driven
• fuel consumption model