State-dependent Nomoto modeling for high-speed autonomous surface vessels
Ocean Engineering, vol.366, 2026 (SCI-Expanded, Scopus)
- Publication Type: Article / Article
- Volume: 366
- Publication Date: 2026
- Doi Number: 10.1016/j.oceaneng.2026.127396
- Journal Name: Ocean Engineering
- Journal Indexes: Science Citation Index Expanded (SCI-EXPANDED), Scopus, Applied Science & Technology Source, Compendex, Environment Index, Geobase, ICONDA Bibliographic, INSPEC, The International Construction Database (ICONDA), Academic Search Ultimate (EBSCO), Engineering Source (EBSCO)
- Keywords: Autonomous surface vehicles (ASV), Nomoto model, Nonlinear optimization, Software-in-the-loop (siL), System identification
- Yıldız Technical University Affiliated: Yes
Abstract
Accurate mathematical modeling of Autonomous Surface Vessels (ASVs) is a core part of the predictive control and Software-in-the-Loop (SiL) deployment. The Nomoto models, frequently utilized for such applications, provide a simplified mathematical framework to describe a ship’s maneuvering behavior; however, they fail to capture high-speed transient nonlinearities. This limitation arises because the Nomoto parameters are treated as constants in the original formulation, whereas they are inherently functions of the rudder angle, ship speed, and acceleration. This paper presents a novel hybrid system identification framework that extracts “state-dependent Nomoto parameters” (K and T not as constants but functions) and coupled roll/pitch dynamics via a foundational forecasting model and physics-guided optimization. Raw telemetry data are constrained by sensor noise and filter-induced phase lag to extract the Nomoto parameters. A time-series foundation model (TimesFM) is utilized for zero-shot, phase-preserving denoising to address this. Adaptive windowed Ordinary Least Squares (OLS) generate localized parameter estimates, which are then mapped into a quasilinear structure via L2[jls-end-space/]-regularized Ridge regression, linking steering dynamics directly to the vessel’s kinematic state. Finally, a global L-BFGS-B optimization minimizes the discrete-time integration error across the trajectory. The identified model is validated within a SiL architecture.