Identification-based 3 DOF model of unmanned surface vehicle using support vector machines enhanced by cuckoo search algorithm

The combination of least square support vector machine (LS-SVM) and cuckoo search (CS) algorithm was first proposed to identify the dynamic models of unmanned surface vehicle (USV). The 3-DOF of Abkowitz model was selected to describe the USV's dynamics. The zigzag test was carried out in the Qinghuai river. The input data and output data obtained by the experiment were selected and filtered to identify the USV's dynamics. The back propagation neural network (BPNN) is a popular method to identify the ship dynamics and was adopted, in this paper, to compare the LSSVM. In addition, the frequently optimization algorithm including particle swarm optimization (PSO) and cross validation (CV) were also selected to enhance the LSSVM which compare to the CS-LSSVM. The results showed that the CS-LSSVM had a better predictive capability than the BPNN, PSO-LSSVM and CV-LSSVM in predicting the surge velocity and sway velocity and the values were close to the experimental data. The related mean square errors of CS-LSSVM was the lowest in these methods and has the fastest convergence speed. It can be tested that CS-LSSVM would be a potential method to online parameter identification for USV in the future.