Online Neural-net Control System for Ship Motion Stabilization

Abstract

An online neural-net control system, in which learning is performed in a loop totally independent from the control loop, is proposed for the problem of ship motion control, including both roll and yaw stabilization at the same time. Based on the experimental data, disturbance model caused by sea wave, including roll moment and yaw moment, is presented. With the disturbance model as input, a recurrent neural network is proposed to approaching the forward model of the real ship, and the real time recurrent learning algorithm is described to train the forward model. Then neural-net controller is presented to reduce to the roll and yaw synthetically. This paper proposes the adaptation process of control system and applies it to the HD702 ship. The approaching accuracy of forward model network and the control effect of the whole system are investigated.