Route and speed optimisation of a general cargo ship using extreme gradient boosting and enhanced Deep Q-Network approaches

• Data from on-board sensors of a general cargo vessel and Copernicus service. • XGBoost-based model developed for providing fuel usage. • Enhanced Deep Q-Network applied for route and speed optimisation. • Multi-objective optimisation considering fuel usage, voyage time and safety. • Over 21% reduction in fuel usage and emission achieved by the proposed approach. To align with the IMO GHG strategy published in 2023 for reducing CO 2 emissions per transport work, optimising shipping operations is crucial. While applying alternative fuels is the key strategy, their high cost highlights the need to improve operational efficiency in shipping. Route optimisation is one of the key operational measures to reduce fuel oil consumption (FOC) and hence associated emissions. This study presents a novel reinforcement learning-based methodology for route optimization of a cargo ship retrofitted with a Gate Rudder (GR) system, simultaneously targeting FOC reduction, time cost and navigational safety. A foundational aspect of the research is the development of a FOC prediction model using Extreme Gradient Boosting to accurately forecast fuel consumption. The predicted FOC values are then adopted into the environment of an enhanced Deep Q-Network to simultaneously optimise ship speed and route. The results demonstrate that, with the proposed approach, fuel consumption can be reduced by up to 27.81 % compared to the original route operated at service speed prior to the installation of the GR system. Of this reduction, 7.79 % is attributable to the GR system itself, while the remaining 20.02 % results from the proposed route optimization method, considering fuel consumption, voyage time, and safety. This reduction results in a decrease of up to 10,379 kg in CO 2 emissions, which further highlights the environmental benefits of the proposed optimisation approach, as well as the GR system.