Ship motion prediction system development based on neural network

Abstract

Ship motion prediction is critical for the safety of vessels operating offshore environment, especially ship-to-ship mooring. Owing to the dynamic and nonlinear characteristics of ship motions, it is very difficult to accurately predict the ship motions using analytical methods. Nonlinear autoregressive exogeneous inputs (NARX) network, which is a kind of dynamic artificial neural network (ANN), however, enables the prediction of multi-step ship motions instantly by using recursive input-output mapping. However, ANN could not only generate unreasonalbe prediction results in process of time. In this paper, hybrid ship motion prediction system (HSMPS), denoted as ANN-based ship motion prediction system in conjunction with a ship response amplitude operators (RAOs) based linear ship motion estimation system, is developed. HSMPS was tested using numerical simulation data and experimental data and, it shows that it offers about 30% lower error rate than ANN-only ship motion prediction system based on RMSE.