Time series prediction of ship maneuvering motion based on dynamic mode decomposition

Abstract

In order to reveal the dynamic characteristics and achieve rapid time series prediction of ship maneuvering motion, a reduced-order dynamic mode decomposition (DMD) algorithm is applied to reconstruct and predict the zig-zag and turning circle maneuvering motions. A case study is conducted for a KVLCC2 tanker using its free-running model test data. First, all DMD modes are extracted from the test data, and the dominant DMD modes are selected according to their contributions to the dynamical systems of ship maneuvering motions. Then, the dynamical systems of ship maneuvering motions are reconstructed and predicted by the reduced-order and full-order DMD algorithms. A comparison between reduced-order, full-order DMD algorithms and Gaussian process regression (GPR) is conducted. The dynamic characteristics of the dynamical systems are revealed according to the growth rates and frequencies of the dominant DMD modes. The effects of the truncation rank and input samples are analyzed by a parametric study, which indicates that the truncation rank and input samples are crucial to the prediction accuracy. Besides, the computational time of the different algorithms is compared and analyzed.