Application of an On-line Trainning Predictor/Controller to Dynamic Positioning of Floating Structures

An on-line training functional-link neural network predictor/controller for dynamic positioning of water surface structures is described in this paper. To develop a neural network for time-evolving systems, the deterministic on-line training model in a traditional parameter identification theory and the functional-link network are combined. The system’s previous input and output are used to be additional enhancements to the functional-link network. The on-line training neural network predictor acquires the knowledge about the system using a small number of samples of the latest system status measured on board of the structure. The trained functional-link neural network is used with an optimal controller to control the output of the system. The accuracy and robustness of the on-line training predictor are demonstrated through the numerical simulations of two ship maneuvers. The on-line training neural network predictor/controller is applied to the dynamic positioning (station-keeping) of a ship in a uniform current with and without external environmental disturbances. The results of the numerical simulations are very satisfactory.