Simulation of artificial neural networks for ship steering control

Interest in the potential of artificial neural networks for automatic control has grown in recent years. This is due to the fact that these networks appear to provide a form of nonlinear control, with fixed controller parameters, which has the predictability of a conventional linear controller but the performance of an adaptive system in the face of large changes of operating point or of plant parameters. This paper describes simulation studies involving two approaches to the use of neural networks for ship steering control. In one of these the network was trained to behave like a specific conventional controller while in the second approach the network was trained to learn the inverse dynamics of the plant. One of the important features of ship steering control problems is that the dynamics of the ship change significantly with forward speed. This phenomenon has been used to investigate the ability of neural networks to control a system having time-varying dynamics. Simulation results show that both of the approaches considered can yield control systems which have a satisfactory level of performance for a wide range of conditions. However, an appropriate choice must be made for the network configuration and for the traing data to allow given performance specifications to be met.