Abstract
Maneuvering models of a ship are quite complex and require accurate knowledge of the various hydrodynamic derivatives and coefficients to model the maneuvering trajectories undertaken by a ship accurately. Discerning these coefficients through practical tests in specialized facilities is expensive and time-consuming. Data-driven identification of the ship maneuvering coefficients using free-running model data is a possible alternative. In this study, the maneuvering motions of the KCS hull are simulated using the MMG model, which are then used as a starting point to identify an Abkowitz model for the vessel. The coefficients of the Abkowitz model are predicted without presuming any knowledge of the MMG model used to generate the data. Three different approaches — Least squares, LASSO optimization, and Support Vector Machines — are used to identify the hydrodynamic coefficients of the Abkowitz model. The identified coefficients are used to generate test trajectories not seen in training data and are compared with the MMG model to verify that the dynamics are accurately captured. The common issue of parameter drift occurring due to the problem being ill-posed is discussed, and sparsity-based solutions are suggested to overcome them.