Neural Networks for System Identification of Coupled Ship Dynamics

Abstract System identification of coupled ship dynamics is problematic with standard least squares methods due to the non-linear, multivariable nature of the system. Neural Networks have therefore been applied to this problem, as they are particularly suitable for approximating non-linear, multivariable functions. In this paper, results of identification with Neural Networks are given for a ship motion simulation based on a standard mathematical model, and for real data collected from a 1/50th scale model of the system. The method is seen to be successful at various operating points, and ideas for extension of the work are discussed.