Nonparametric Identification Modeling of Ship Manoeuvering Motion Based on Gaussian Mixture Model

The dynamic simulation of the ship dynamics model using non-parametric identification can achieve high-precision simulation of the ship simulator and adaptive control of the intelligent ship.In order to reduce the error in the area with large changes and improve the recognition efficiency, this paper improves the traditional Gaussian mixture model (GMM) and proposes a non-parametric identification model based on the improved GMM to simulate and predict the motion state of the ship. The GMM is trained with multidimensional data such as hydrodynamics and rudder angle, the traditional initial model data is improved in sections, the expectation maximization (EM) algorithm is used for parameter optimization, and the improved model parameters are obtained by calculating the modified Gaussian mixture coefficients. The hydrodynamics of the ship are calculated by Gaussian mixture regression using the trained model and the predicted rudder angle, and the maneuvering motion state is solved by the Runge-Kutta method. The feasibility and effectiveness of this method are validated through experiments on the motion state of different rudder angles using turning test and zigzag test.