Non-parametric identification modeling and prediction of intelligent ship maneuvering motion based on real ship test at sea

Abstract

The rapid development of intelligent ships has put forward higher requirements for the accuracy and practicability of ship maneuvering motion modeling and prediction. This paper takes the first intelligent research and training dual-purpose ship “Xin Hong Zhuan” of Dalian Maritime University as the research object. A multi-output support vector regression black box modeling method based on Bayesian Optimization(BO) and mixed kernel function is proposed for ship maneuvering motion prediction. This method uses a mixed kernel function multi-output support vector regression (MK-MO-$\nu$-SVR) model combining radial basis function and polynomial kernel function, and uses Bayesian Optimization algorithm to optimize the model hyperparameters. Taking the real ship test data of the intelligent ship “Xin Hong Zhuan” as the dataset, the superior performance of the proposed model in the short-term and long-term prediction of ship zigzag maneuver and turning circle maneuver is verified by comparing different kernel function combinations. The results show that the MK-MO-$\nu$-SVR model proposed in this paper performs well in prediction accuracy and generalization ability, and can effectively predict ship maneuvering motion, which provides a powerful technical means for navigation safety and decision support of intelligent ships.