Prediction of the required propulsive power of a ship through guided neural networks

Abstract

The increasing environmental regulations from international organizations such as the International Maritime Organization’s Energy Efficiency Existing Ship Index (EEXI) and Carbon Intensity Indicator (CII) and the European Union’s Emissions Trading System (ETS) force shipping companies to boost their fleet operational efficiency. The main objective of these initiatives relies on precise propulsive power forecasting for vessels operating across different conditions. The presented Physics-Informed Neural Network (PINN) framework combines Partial Differential Equation (PDE) constraints with boundary conditions and physics-based relationships along with real measurement data. The network achieves better prediction accuracy of ship propulsion power through physical law incorporation during learning which reduces the need for labeled data in real-world scenarios. The proposed method maintains compliance with naval architectural basics by incorporating weather conditions and vessel draft considerations. Actual ship operational data tests demonstrate that the PINN model generates better performance than standard data-driven models and demonstrates its ability to unite physics-based classical analysis with data-oriented methods. The combined approach provides improved reliability and adaptability in maritime complex environments which leads to more efficient fleet operations that benefit the environment. The framework shows promise for future application to different vessel types which would enhance sustainable shipping methods.