A Constrained Least Squares Algorithm for Maneuvering Dynamics Identification of Marine Crafts

Aiming at establishing an accurate maneuvering model of marine crafts, a novel constrained least square algorithm (CLS) is proposed in this paper for parameter identification. Firstly, an improved second-order Nomoto model is established and discretized into a matrix form. Secondly, a dedicated CLS algorithm is derived, whose loss function with priori knowledge of partial model parameters can suppress the over-parametrization phenomenon. Finally, experimental results indicate that the CLS algorithm has better identification accuracy and generalization ability than the traditional least squares algorithm and the support vector machine algorithm. Specifically, the identification error of turning index
$K$
is only 1.46% and various heading angle generalization errors are all within 5° for different maneuvering trials.