Data-driven modelling and prediction of ship manoeuvring motions via Dynamic Bayesian Networks

Çelik, C, Li, H ORCID: 0000-0002-4293-4763, Zou, L, Bashir, M and Yang, Z ORCID: 0000-0003-1385-493X (2026) Data-driven modelling and prediction of ship manoeuvring motions via Dynamic Bayesian Networks. Ocean Engineering, 369. ISSN 0029-8018

Preview

Text Data driven modelling and prediction of ship manoeuvring motions via Dynamic Bayesian Networks.pdf - Published Version Available under License Creative Commons Attribution. Download (3MB) | Preview

Abstract

Accurate modelling and prediction of ship manoeuvring motions are essential for navigation safety and control system development. Conventional physics-based manoeuvring models typically depend on predefined hydrodynamic formulations and extensive captive or free-running tests, which constrain their adaptability to varying operating conditions. This study develops a data-driven Gaussian Dynamic Bayesian Network (DBN) framework with time-lagged dependency learning for modelling ship manoeuvring dynamics and recursively predicting their evolution under new manoeuvring scenarios. The DBN explicitly characterises the temporal and conditional relationships between surge, sway, yaw motions, and rudder actions, allowing both dynamic modelling and interpretable assessment of underlying motion couplings. Free-running model tests of the KCS benchmark ship in deep and calm water are used for model training and testing. The DBN demonstrates strong predictive capability for both zig-zag and turning manoeuvres, with its performance evaluated against Long Short-Term Memory (LSTM) and Gaussian Process Regression (GPR) models. The results show competitive prediction performance across the investigated motion variables, with the relative accuracy varying among the three models, while the DBN requires substantially lower training time than the others. In addition to its predictive performance, the DBN provides a transparent representation of time-lagged dynamic dependencies, facilitating engineering interpretation of manoeuvring behaviour. Overall, the results demonstrate the potential of the proposed framework for recursive ship manoeuvring prediction.

Item Type:	Article
Uncontrolled Keywords:	Data-driven modelling; System identification; Dynamic Bayesian Networks; Ship manoeuvring; Probabilistic prediction; 4015 Maritime Engineering; 40 Engineering; Machine Learning and Artificial Intelligence; 0405 Oceanography; 0905 Civil Engineering; 0911 Maritime Engineering; Civil Engineering; 4005 Civil engineering; 4012 Fluid mechanics and thermal engineering; 4015 Maritime engineering
Subjects:	Q Science > QB Astronomy
Divisions:	Engineering and Built Environment
Publisher:	Elsevier
Date of acceptance:	13 October 2026
Date of first compliant Open Access:	5 November 2026
Date Deposited:	05 Nov 2026 11:04
Last Modified:	05 Nov 2026 11:04
DOI or ID number:	10.1016/j.oceaneng.2026.128869
URI:	https://researchonline.ljmu.ac.uk/id/eprint/29824

View Item