Online prediction of ship roll motion based on a coarse and fine tuning fixed grid wavelet network

Abstract

A new coarse and fine tuning fixed grid wavelet network is proposed for online predicting ship roll motion in regular waves. This wavelet network is composed of discrete wavelet basis functions, whose structure and parameters are adjusted online based on a sliding data window. Coarse tuning refers to changing the structure of the wavelet network, and fine tuning refers to only changing the coefficients of the wavelet network. In every sliding data window, the Givens rotation method is first used to finely tune the wavelet network, then the orthogonal least squares algorithm and the error reduction ratio criterion will be used to coarsely tune the wavelet network if the fine tuning network does not satisfy the relevant conditions for model validation. This process guarantees that the method can establish not only the model of the weakly nonlinear motion but also the model of the strongly nonlinear motion. In this way, a concise prediction model of ship roll motion is obtained. This network exploits the attractive features of wavelet and the fitting capability of conventional neural network. The prediction results show that the modeling method is feasible, and it provides an effective tool for online predicting ship roll motion.