Identification of vessel kinetics based on neural networks via concurrent learning

Abstract

This paper is concerned with system identification for autonomous surface vehicles subject to unknown kinetics. The considered unknown kinetics stems from model uncertainties, unmodeled dynamics and external disturbances caused by wind, waves and ocean currents. The identification method is developed based on neural networks owing to its universal approximation property. In the adaptive weight law design, a concurrent learning method is involved to utilize the instantaneous data and the recorded data for adaptation. By using the proposed identification approach, the output weights will approach and stay bounded within a small neighborhood of ideal weights without a persistence of excitation condition. Finally, by resorting to the Lyapunov theory, the performance of the proposed kinetics identification method is analyzed.