Robust vessel maneuvering modelling using set-membership identification

• A set-membership method is proposed for the identification of a vessel’s 3-DOF maneuvering model parameters. • The identified parameters include the hydrodynamic derivatives and actuation-related parameters. • The method relies on bounded descriptions of uncertainties (e.g., sensor noise, disturbances, etc.) to return a feasible parameter set and a nominal parameter estimate. • The optimisation problem is transformed from a nonlinear non-convex problem to a set of linear quadratic problems. • Validation using data from experimental trials with a full-scale vessel demonstrates improved computational time and state prediction accuracy. System identification of full-scale surface vessels must address significant uncertainties arising from model mismatch, sensor noise, and environmental disturbances. To provide safety, robustness, and constraint satisfaction guarantees, especially for autonomous navigation applications, it is essential to quantify the bounds of parametric model uncertainty. This paper proposes a set-membership identification method for estimating key parameters of a nonlinear vessel maneuvering model, including inertia and added-mass terms, other hydrodynamic derivatives in the Coriolis-centripetal, damping matrices, and actuation-related parameters. The method provides a bounded-error characterisation of uncertainties, offering a reliable framework for modelling the effects of measurement noise, wind, and waves. It involves computing a data-driven parameter set (DDPS) using input-output measurements and model assumptions, which is further used to compute a feasible parameter set (FPS). The parameter estimates are then obtained by iteratively solving a quadratic program over the FPS polytope. Validation of the method using experimental data from a full-scale catamaran demonstrates improved accuracy of up to 26.5% as compared to existing approaches, significantly faster computational times, and its capability to provide bounded parameter estimates.