Extremely short-term ship motion prediction using real-sea motion datasets based on the VMD-LSTM-FDR network

Ship motion in real sea is inherently random and uncertain, significantly impacting operational safety. The study proposes an extremely short-term ship motion prediction model integrating Variational Mode Decomposition (VMD), LSTM networks, and a Frequency Domain Rectification (FDR) algorithm. The motion time series is decomposed into intrinsic mode functions (IMFs) using VMD. Then, these IMFs are used as the input of a multi-input multi-output LSTM network for extremely short-term prediction. In this process, the FDR algorithm is introduced to rectify amplitude predictions. Moreover, the optimal advance prediction time is discussed. The prediction model applied to sea-trial motion data achieves 80%-90% accuracy in extremely short-term motion prediction. To verify prediction performance of the prediction model in different types of ships, it is applied to ships of different tonnages: an offshore supply ship (20900t), a crane ship (51000t), and a fishing vessel (50t) operating in nearshore or open-sea area. The extremely short-term motion prediction achieved an average accuracy exceeding 80%. It is found that the prediction accuracy rate in nearshore area is generally higher than that in open-sea area. Increasing the data sampling rate is an effective approach to enhance the accuracy of extremely short-term ship motion prediction, while higher sampling rate results in larger training efforts. Prediction performance is affected by the length of training datasets, with training dataset length at ∼10 2 timesteps balancing accuracy and computational cost. The model has good generalizability for the same ship under different conditions, while its performance drops sharply when directly applied without pretraining to different ships. • A VMD-LSTM-FDR network model for extreme short-term motion prediction is proposed. • Frequency Domain Rectification method effectively corrects amplitude prediction errors. • 80–90% prediction accuracy is achieved with high generalizability across different vessels. • The prediction accuracy in nearshore area is generally higher than that in open-sea area. • Increasing sensor sampling rates improves prediction accuracy with higher computational costs.