Physics-informed identification of marine vehicle dynamics using hydrodynamic dictionary library-inspired adaptive regression

Dynamic modeling of unmanned marine vehicles (UMVs) plays an integral role in implementing the autopilot and intelligence capabilities of these vehicles. However, there is a trade-off between accuracy and interpretability, which highlights the need for a modeling approach that balances both factors. To address this challenge, this paper presents a novel, open-source, physically motivated data-driven method named hydrodynamic dictionary library-inspired adaptive regression (HLAR). This approach combines hydrodynamic terms from classical models to form a priori dictionary and utilizes the sequential thresholded least-squares (STLS) algorithm and Bayesian optimization to identify model counterparts and adaptively optimize the hyperparameters. The proposed method was validated using KVLCC2 ship experimental data, NPS-AUV simulation data and Star Ocean II AUV experimental data, and it can achieve both high accuracy and good interpretability with low complexity. Experimental results show that the proposed method is 100 times faster than the Gaussian process and achieves the same level of accuracy. •A novel, open source physically motivated data-driven method is proposed for unmanned marine vehicle system identification.•The scheme utilizes the STLS algorithm and Bayesian optimization to automatically discover model counterparts and optimize hyperparameters.•The proposed method has been validated using ship experimental data, AUV simulation data and AUV experimental data.•The presented approach shows better generalization ability and computational efficiency than regular Gaussian Process.