Neural Learning Control for Discrete-Time Autonomous Marine Surface Vessels

Abstract

This paper studies neural learning control problem for a kind of 3 degree-of-freedom fully-actuated autonomous marine surface vessels (MSVs) with unmodeled and unknown nonlinear dynamics in discrete-time domain. Based on the discrete-time dynamic model of the autonomous MSV, an adaptive neural network (NN) controller is first proposed to make the MSV follow the given recurrent trajectory. Then combining a new stability result of linear time-varying system in discrete-time domain with 2-steps delay and the deterministic learning theory, it is proved that the estimated NN weights converge to their optimal values, exponentially. By analyzing the convergence characteristic of the estimated NN weights, the convergent constant weights can be synthetically stored as knowledge, which can accurately identify the unknown nonlinear dynamics of the MSV. Subsequently, a neural learning controller is proposed to accomplish similar control tasks without re-adapting to the unknown nonlinear dynamics. Finally, simulation results for a 3 degree-of-freedom fully actuated MSV are presented to show the validity of the neural learning control scheme.