A Method Of Prediction Ship Maneuverability Based On Neural Network

With the development of international trade, the world's marine is expanded rapidly, which makes the amount of ships in the sea increased evidently, single ship tonnage enlarged sharply, the ship speed improved constantly; which creates the sea transportation density extended, the voyage more complicated. According to the statistics of data in recent years, the casualty ships because of ship maneuverability are about more than 200 every year, so the study on the ship maneuverability is attached more and more importance.When ships are at their actual voyages the working conditions (such as loading, depth of drafu ship speed) and the navigating environments(such as the sea route, depth of water, wind, wave, current)are variable to a large extent. The whole during is a procedure of model varying with time nonlinear and large disturbance. The introduction of the intelligent controlling method can contribute to the improvement of automation degree for ship avoiding collision and maneuvering systems.In recent years, with the second rise of neural network study, the neural network was applied in many different fields, such as mode distinguish, system identification, image processing and auto control, of which the application in the ship maneuverability was paid more and more attention. The reason is that the neural network has its outstanding advantages: (1) It can approach any complicated nonlinear function relations; (2) It can adapt and learn the dynamic characteristics of non-defined system; (3) It has good robustness because all the information is stored in the coefficients between the network cells; (4) Its computing speed is high because of the parallel process of the information distribution.In this paper , the method of neural network was used to ship maneuverability. The ship maneuvering hydrodynamics, turning circle and zigzag curve were studied and predicted . In the aspect of ship maneuveringhydrodynamics, taking bulk carriers and oil tankers for an example, the prediction was done; in the turning circle, different ships under different rudder angles were studied and predicted; in the zigzag curve, different ships under constant rudder angle were studied to predict another two ships. It is verified that the prediction accuracy of ship maneuvering hydrodynamics, turning motion and zigzag curve are acceptable in practice, if the learning samples, network's structure and learning algorithm are reasonable.