Overview
Date:
Starts 7 May 2026, 09:00Ends 7 May 2026, 13:00Location:
Lecture Hall VDL, Campus Johanneberg, Chalmers Tvärgata 4COpponent:
Qiang Meng, Professor, Director of the Centre for Transport Research, Department of Civil and Environmental Engineering, National University of Singapore, Singapore.Thesis
Read thesis (Opens in new tab)
Short-sea shipping (SSS), essential for European transport logistics, is increasingly challenged by strict environmental regulations such as the EU Emissions Trading System, fuel price volatility, and the need to maintain tight schedules on short voyages. This thesis reframes SSS operations as a data-driven voyage-optimisation problem, developing an integrated framework that combines operational ship data, metocean data, machine learning (ML) models, and advanced optimisation algorithms to minimise voyage fuel consumption and emissions while meeting estimated time of arrival targets.
The framework was developed through the two interconnected fields of modelling and optimisation. (1) Modelling established surrogate models for ship performance: multiple ML regression algorithms were benchmarked for fuel consumption prediction, with XGBoost identified as the most stable and reliable. Independently, Gaussian process regression was employed to estimate added resistance in head waves for model-scale ships. When integrated into a grey-box neural network fuel model, it reduced prediction errors by a factor of 3 relative to semi-empirical methods. (2) Optimisation involved introducing a methodology to optimise the total fuel consumption of a voyage, validated across two case-study ships.
For a double-ended ferry, a complete decision-support system using Bayesian optimisation (BO) was implemented to determine the optimal power profile across an entire voyage. This achieved simulated fuel savings of up to 43%, with an 18% reduction confirmed during full-scale sea trials. For longer SSS voyages, the framework was extended to a chemical tanker case study. A metocean-aware segmentation algorithm, MS-PELT, was developed to divide routes into operationally meaningful legs, outperforming state-of-the-art methods whilst enabling near-real-time application. Voyage optimisation was subsequently performed using parallel coupled dynamic programming (PCDP), achieving up to 14.8% fuel savings. A final refinement step combining PCDP and BO achieved a potential fuel saving of 9.3% relative to measured voyage fuel consumption.
The framework was developed through the two interconnected fields of modelling and optimisation. (1) Modelling established surrogate models for ship performance: multiple ML regression algorithms were benchmarked for fuel consumption prediction, with XGBoost identified as the most stable and reliable. Independently, Gaussian process regression was employed to estimate added resistance in head waves for model-scale ships. When integrated into a grey-box neural network fuel model, it reduced prediction errors by a factor of 3 relative to semi-empirical methods. (2) Optimisation involved introducing a methodology to optimise the total fuel consumption of a voyage, validated across two case-study ships.
For a double-ended ferry, a complete decision-support system using Bayesian optimisation (BO) was implemented to determine the optimal power profile across an entire voyage. This achieved simulated fuel savings of up to 43%, with an 18% reduction confirmed during full-scale sea trials. For longer SSS voyages, the framework was extended to a chemical tanker case study. A metocean-aware segmentation algorithm, MS-PELT, was developed to divide routes into operationally meaningful legs, outperforming state-of-the-art methods whilst enabling near-real-time application. Voyage optimisation was subsequently performed using parallel coupled dynamic programming (PCDP), achieving up to 14.8% fuel savings. A final refinement step combining PCDP and BO achieved a potential fuel saving of 9.3% relative to measured voyage fuel consumption.
Daniel Vergara
- Doctoral Student, Marine Technology, Mechanical Engineering