Ship Steering Control by Neural Networks Trained using Feedback Linearization Control Laws

Abstract

One problem of ship steering systems is that the dynamics of the vessel are dependent upon the forward speed. Since artificial neural networks can provide a nonlinear controller which performs well for a wide range of plant dynamics such networks are of potential interest for ship steering applications. This paper describes simulation studies in which a feedforward network is trained to behave like a feedback linearization controller. Results suggest that the approach can yield a control system having a satisfactory level of performance for a range of operating conditions. The choice of network configuration and training data sets are, however, of considerable importance.