Comparative study of ship fuel consumption prediction models based on multi-source operational and environmental data under dynamic ocean conditions

Artikel i vetenskaplig tidskrift, 2026

Accurate prediction of ship fuel consumption is essential for improving maritime energy efficiency under variable hydrometeorological conditions. This study compares semi-empirical, physics-based, and artificial intelligence (AI)-based models using full-scale operational and environmental data from a Kamsarmax bulk carrier. The novelty of the work lies in the unified assessment of these three modelling methods under realistic and dynamic ocean conditions. The semi-empirical model estimates fuel use through resistance, propulsion efficiency, and engine fuel consumption relationships, while the physics-based model incorporates hydrodynamic, manoeuvring, and environmental effects. The AI model applies an attention-enhanced bidirectional long short-term memory network to learn nonlinear relationships from operational data. The comparative results show that the AI-based model provides the most accurate and stable predictions across unseen voyages, followed by the physics-guided model, whereas the semi-empirical model shows larger deviations under complex environmental conditions. These findings highlight the potential of AI-based prediction for fuel efficiency optimization and operational decision support, while physics-based modelling remains valuable for interpretation and diagnostic analysis.