Adaptive hybrid-kernel function based Gaussian process regression for nonparametric modeling of ship maneuvering motion

Abstract

A novel adaptive hybrid-kernel function based Gaussian process regression (AHKGPR) is proposed for nonparametric modeling of ship maneuvering motion. With the aid of Gaussian process regression (GPR) and nonlinear kernel function, the nonparametric ship dynamic model is identified based on the training data collected during free-running tests. The hybrid-kernel function which combines the good properties of radial basis function (RBF) and polynomial function with adaptive gain coefficients is designed to improve the generalization ability and the prediction accuracy of GPR model. The hyperparameters in the kernel function and the gain coefficients are tuned and optimized by genetic algorithm (GA). To verify the effectiveness of AHKGPR, a comparative study between AHKGPR and the widely used RBF kernel function based GPR (RBFGPR) is carried out. Two cases are considered, one for the Mariner class vessel with the dataset generated from simulation and the other for the container ship KCS with the real measured dataset of free-running model tests provided by SIMMAN 2020 Workshop. The results show that AHKGPR has higher prediction accuracy and better generalization ability.