Recurrent Neural Networks for Ship Modeling and Control

摘要

This paper aims at analyzing the implementation of these techniques for theidentification of a maneuvrability model, and then for developing an adaptive autopilot scheme. The different recurrent neural architectures which are used, and the principles of the associated training methods, detailed and illustrated with a set of realistic simulations of ship maneuvers (presently the Charles de Gaulle aircraft carrier). A comparison with classical methods permits a better understanding of its interest and complementarity, and likewise to unmystify such an approach.