Neural network estimators: application to ship position estimation

Ship motion prediction is an important practical problem in design of many ocean systems i.e. fire controls, air craft landing and take off. As such the problem has received considerable attention in the past and various attempts for designing estimators via the statistical methodology were made. In this paper the design of the ship position estimator via neural networks is considered. The ship position estimation is viewed as an adaptive estimation problem for partially unknown systems. New powerful neural estimators based on dynamic recurrent neural networks are applied to the solution of the ship position problem. The new proposed neural estimators, and the state of the art statistical filters, are comparatively evaluated via extensive simulations. The results show that the neural algorithms have excellent performance, achieving significant computational savings due to their massively parallel structure