Simulative evaluation of applying optimized support vector machines to identify the simplified ship dynamic model

Abstract

Model-based simulation is a sufficient and cost-effective approach for studying and analyzing the maneuverability of ships involving the maneuvers prediction. One important foundation for ensuring its implementation is the determination of the ship dynamic model with relatively low complexity and high accuracy. This study aims at contributing to this research point from two aspects: one is the simplification of dynamic model of ships through 6 degrees of freedom (DOF) nonlinear and complex ship dynamic model based on reasonable assumptions; the other one is to identify parameters of the simplified model using support vector machines (SVM) and maneuvering data. The artificial bee colony algorithm (ABC) is used to remedy the problem of SVM about selecting optimal parameter values. For the numerical simulation study, a container ship with well-proven dynamic model is applied to generate clean and polluted maneuvering data. Comparison with the first order linear Nomoto model indicates that the simplified steering model can capture more complicated motions and shows better performance. The optimized SVM by ABC is a convenient and effective alternative for identification of ship dynamic models. It is noticeable that the higher level of measurement noise is, the worse influence on identification results is. But in some degree, the filter approach can mitigate the negative influence and in turn improve the identification results.