Recurrent-type Neural Networks for Real-time Short-term Prediction of Ship Motions in High Sea State

TL;DR

This paper evaluates recurrent neural networks, including LSTM and GRU, for real-time short-term prediction of ship motions in high sea states, demonstrating promising results in a CFD simulation context.

Contribution

It compares the performance of RNN, LSTM, and GRU models for ship motion nowcasting in high sea states using CFD data, highlighting their effectiveness.

Findings

All three models provided promising and comparable prediction accuracy.

The models achieved about 20 seconds ahead prediction in high sea state conditions.

Recurrent neural networks are suitable for real-time ship motion prediction in challenging sea states.

Abstract

The prediction capability of recurrent-type neural networks is investigated for real-time short-term prediction (nowcasting) of ship motions in high sea state. Specifically, the performance of recurrent neural networks, long-short term memory, and gated recurrent units models are assessed and compared using a data set coming from computational fluid dynamics simulations of a self-propelled destroyer-type vessel in stern-quartering sea state 7. Time series of incident wave, ship motions, rudder angle, as well as immersion probes, are used as variables for a nowcasting problem. The objective is to obtain about 20 s ahead prediction. Overall, the three methods provide promising and comparable results.