Abstract
This paper evaluates hydrodynamic, data-driven, and hybrid approaches for modeling ship performance across varying operating conditions. Using full-scale measurements from a 200 m deep sea bulk carrier, shaft power is estimated with an empirical resistance-based method, machine learning models, and physics-informed machine learning that integrate calm water resistance from the Hollenbach method. All predictions methods are assessed under different speeds, power levels, and weather conditions. Results show that hybrid models outperform purely empirical and data driven approaches. In particular, the physics-informed neural network achieves the highest accuracy and robustness, performance across regimes and demonstrating resilience in sea states.