Incorporation of a System Identification–Based Ship Manoeuvring Model into Real-Time Collision Risk Analysis of Autonomous Ships

Celik, C ORCID: 0000-0002-2335-0455 (2026) Incorporation of a System Identification–Based Ship Manoeuvring Model into Real-Time Collision Risk Analysis of Autonomous Ships. Doctoral thesis, Liverpool John Moores University.

Text 2026celikphd.pdf - Published Version Access Restricted until 28 July 2027. Available under License Creative Commons Attribution Non-commercial. Download (3MB)

Abstract

Maritime transportation plays a vital role in global trade, supporting the continuous movement of cargoes across increasingly dense and operationally complex sea routes. As traffic intensity and the level of vessel autonomy continue to grow, ensuring navigational safety and reliable collision-risk awareness has become a critical challenge for both researchers and industry stakeholders. Addressing this need requires new modelling approaches that can capture high uncertainty in operations, evolving encounter dynamics, and real-time ship manoeuvring behaviour beyond the existing decision-support frameworks in the literature.
This thesis develops a probabilistic and data-driven decision-making framework to enhance collision-risk awareness and support autonomy-related anti-collision operations in ship navigation by linking Bayesian risk modelling with system identification-based manoeuvring prediction. The main novelties are that collision risk is assessed as a time-evolving and uncertain process, and that meaningful risk evaluation incorporates motion predictions that remain dynamically reasonable rather than relying solely on straight-line, constant-speed kinematic extrapolation.
The study first establishes a macro-level safety foundation by applying Bayesian Network (BN) modelling to global maritime collision accident data. This analysis identifies key Risk Influential Factors (RIFs) and their probabilistic dependencies, providing an interpretable causal perspective on collision occurrence at a worldwide scale. Building on this foundation, the thesis then addresses micro-scale ship-to-ship encounters by proposing a Dynamic Bayesian Network (DBN)-based collision risk prediction framework that operates in a temporal space using AIS-derived encounter information. The model incorporates both conventional geometric indicators and additional dynamic descriptors of encounter evolution, enabling real-time updating of collision risk levels and offering an interpretable representation of how risk develops as traffic situations change.
To support physically consistent prediction of vessel behaviour, an improved DBN methodology is developed for ship manoeuvring modelling within a system identification setting. Using free-running experimental data of the KRISO Container Ship (KCS) benchmark ship, the proposed non-parametric model learns probabilistic temporal dependencies among surge, sway, yaw motion states and rudder actions. The results demonstrate strong predictive capability across standard manoeuvres, achieving accuracy comparable to a representative deep learning baseline while requiring substantially lower computational effort and providing clearer interpretability of dynamic couplings.
Finally, the thesis demonstrates the utilisation of predictive manoeuvring information in two complementary applications: (i) enhancing collision risk evaluation through manoeuvring-consistent trajectory hypotheses under target ship behavioural uncertainty, and (ii) supporting autonomous path-following via a guidance-control design that produces realistic rudder actions and smooth trajectory tracking. Overall, the thesis contributes a coherent probabilistic pathway from global accident understanding to real-time encounter risk prediction and data-driven manoeuvring modelling, providing a foundation for future research toward more robust, uncertainty-aware autonomous maritime navigation systems. It has also generated 4 journal papers and 5 conference papers, including 7 published and 2 currently under review.

Item Type:	Thesis (Doctoral)
Uncontrolled Keywords:	Maritime safety; Maritime collision accidents; Maritime transportation; Collision risk; Dynamic Bayesian Networks; AIS data; Navigational safety; Ship manoeuvring; Nonparametric modeling; System identification; Gaussian Dynamic Bayesian Networks; Collision risk assessment; Motion uncertainty; Motion intention; Path following; Line-of-sight algorithm; Real-time collision risk; Maritime Autonomous Surface Ships; International Maritime Organization; COLREGs; Bayesian Networks; Proportional-integral-derivative algorithm
Subjects:	T Technology > T Technology (General) V Naval Science > V Naval Science (General) T Technology > T Technology (General) > T58.5 Information Technology
Divisions:	Engineering
Date of acceptance:	20 July 2026
Date Deposited:	29 Jul 2026 08:45
Last Modified:	29 Jul 2026 11:34
DOI or ID number:	10.24377/LJMU.t.00029025
Supervisors:	Yang, Z, Li, H, Bashir, M and Zou, L
URI:	https://researchonline.ljmu.ac.uk/id/eprint/29025

View Item