Ship fuel oil consumption prediction at sea and in port considering sustainable maritime industry: A comparative study of machine learning and deep learning approaches

Abstract

Reducing ship fuel oil consumption (FOC) is the key issue for the sustainable maritime industry. Ship FOC is produced not only at sea but also in port. From both marine and port city environmental perspectives, it is imperative to consider ship FOC both at sea and in port. For specific ships in uncertain navigation environments and different ports, ship FOC is in different patterns. Therefore, in this study, 18 machine learning (ML) or deep learning (DL) models with eight hyperparameter tuning methods were conducted and compared to select an optimal prediction model for each ship at sea and in port respectively. To validate the efficiency of selected models, experiments on 10 different ships with four types: three container ships, three bulk ships, three oil tanker ships, and one multi-purpose ship were conducted. The results show that irrespective of whether it is in port or at sea, optimal predictive models and hyperparameter combinations vary across different ships. In addition, the prediction accuracy of ship FOC at sea can be improved by 4.49%–16.46% after considering sufficient meteorological features. The research could provide practical management insights for relevant shipping companies and ports to achieve a sustainable maritime industry.