Abstract
As international climate policies become more stringent, accurate prediction and optimisation of fuel oil consumption (FOC) are now crucial for analysis of a ship’s navigation status, energy conservation, and reductions in greenhouse gas emissions. This study presents two approaches to FOC prediction (using real-time and time-series methods) and a framework for FOC optimisation through analysis of operational data and sailing speed adjustments for a container ship. XGBoost, an ensemble learning model, and Meta-BiLSTM, a deep learning model based on stacking theory, perform exceptionally well in FOC prediction, achieving mean squared errors of 0.04% and 0.07%, respectively. The ship’s route is optimally clustered based on meteorological data, ensuring continuity of the route within each cluster. An FOC prediction model is integrated with the proposed improved grey wolf optimiser (IGWO) algorithm to reduce FOC by adjusting the optimal sailing speed for each cluster along the route. For the ship studied here, an FOC reduction of 4.54% is achieved, equivalent to 33.14 tons. The speed optimisation method employed in this research appears to be more practical under operational conditions than alternative methods.