Reconstruction and prediction of global whipping responses on a large cruise ship based on LSTM neural networks

Abstract

This study presents a model based on Long Short-Term Memory (LSTM) neural networks for the reconstruction and prediction of global whipping responses using ship motion data. The model is fine-tuned and trained based on a dataset of heave, pitch, surge motion data, vertical acceleration data, and vertical bending moments (VBM) from a large cruise ship model experiment by utilizing 5-fold cross-validation. The established model is tested with a pre-split test dataset and used to carry out reconstruction and multi-step prediction of the VBM amidship in following and head seas. The reconstructed results demonstrate a high degree of fit with the measured VBM amidship under both regular and irregular wave cases, indicating the model's ability to capture whipping responses. The findings of multi-step prediction indicate that the prediction accuracy and the model's ability to capture whipping responses decrease as the horizon increases, and the model is more accurate in predicting whipping responses when they are stronger.