Physics-informed machine learning models for ship speed prediction

This paper proposes a novel physics-informed machine learning method to build grey-box model (GBM) predicting ship speed for ocean crossing ships. In this method, the expected ship speed in calm water is first modeled by the physics-informed neural networks (PINNs) based on speed-power model tests....

Full description

Bibliographic Details
Published in:Expert Systems with Applications
Main Authors: Lang, Xiao, Wu, Da, Mao, Wengang
Language:unknown
Published: 1481
Subjects:
Transport Systems and Logistics
Astronomy
Astrophysics and Cosmology
Marine Engineering
Computer Science
Ship speed over ground
Physics-informed neural networks
Machine learning
XGBoost
Full-scale measurements
ETA
Grey-box model
Eta
Pacific
North Atlantic
Online Access:https://doi.org/10.1016/j.eswa.2023.121877
https://research.chalmers.se/en/publication/537824
Description
Summary:This paper proposes a novel physics-informed machine learning method to build grey-box model (GBM) predicting ship speed for ocean crossing ships. In this method, the expected ship speed in calm water is first modeled by the physics-informed neural networks (PINNs) based on speed-power model tests. Then the eXtreme Gradient Boosting (XGBoost) machine learning algorithm is integrated to estimate ship speed reduction under actual weather conditions. The proposed GBM has been compared against the traditional black-box model (BBM) using performance monitoring data from two ships. The results show that when the amount of data is sufficient for modeling, the GBM can increase the accuracy of speed prediction by about 30%. When data volume is limited, the GBM can also significantly improve the prediction results. Finally, the GBM is validated by checking its implementation for the ETA predictions of cross-Pacific or North Atlantic voyages. The highest cumulative error of sailing time estimated by the GBM is 5 h among all the study cases.

Similar Items