Kernel-based support vector regression for nonparametric modeling of ship maneuvering motion

A nonparametric identification method based on nu('nu')-support vector regression (nu-SVR) is proposed to establish robust models of ship maneuvering motion in an easy-to-operate way. Assisted by the kernel trick, the nonlinear model learns implicitly in high-dimensional feature space without a priori model structure. The nu-SVE controls the sparsity automatically, resulting in high efficiency. To improve the practicality, a parameter tuning scheme combining the hold-out validation and the simulation of dynamic processes is designed to avoid over-fitting. Taking the KVLCC2 ship as the study object, the experimental data from the SIMMAN database are used to evaluate the method. The selection and pre-processing of training data are discussed. The identified model shows good generalization performance in the prediction of multiple maneuvers not involved in the training set, verifying the effectiveness of the method.