| Abstract | In this paper, a nonlinear dynamic neural network model is proposed for the identification of ship manoeuvring coefficients. A generalized Widrow-Hoff learning rule in feed-forward, back propagation networks give reasonable answers when presented with inputs not previously seen. With a proper optimization technique, the network is an excellent choice for nonlinear system identification, using reliable and inexpensive computing hardware. It is shown that these networks can be used to identify ship manoeuvres in every practical application, including an open-loop manner. An efficient, quick off-line training process allows the use of neural networks in predicting the ship's future position, speed and heading with input from rudder angle, engine rpm and the ship kinematics. Dealing with input noise from ship sensors is implicit in the proposed method and is very effective, in fact some level of noise is necessary for the training purposes. In the paper the combination of the MMG standard manoeuvring model with the neural network prediction simulates the real-time realization of ship manoeuvring. |
|---|
