AUTHOR=Wang Huifeng , Yin Jianchuan , Wang Nini , Wang Lijun 

TITLE=A multi-dimensional data-driven ship roll prediction model based on VMD-PCA and IDBO-TCN-BiGRU-Attention

JOURNAL=Frontiers in Marine Science

VOLUME=Volume 12 - 2025

YEAR=2025

URL=https://www.frontiersin.org/journals/marine-science/articles/10.3389/fmars.2025.1547933

DOI=10.3389/fmars.2025.1547933

ISSN=2296-7745

ABSTRACT=IntroductionThe motion of a ship at sea is complex. This motion is affected by environmental factors such as wind, waves, and currents. These factors cause the ship’s movement to be nonlinear, dynamic, and uncertain. Such complex motion can impact the ship’s performance and pose a safety risk. This has become an urgent problem in maritime safety. This study aimed to improve the prediction of a ship’s roll motion with high accuracy. As such, the study proposes a combined prediction model. This model integrates data decomposition, dimensionality reduction, deep learning, and optimization techniques.MethodsThe model uses the variational mode decomposition (VMD) method to break down the ship’s roll motion data into components at different scales. This improves the smoothness of the data. Principal component analysis (PCA) is applied to reduce the dimensionality of the decomposed components. This step helps remove noise and redundant features that could affect the prediction results. The core of the model combines temporal convolutional networks (TCNs) and bidirectional gated recurrent units (BiGRUs). These deep learning techniques enable the model to extract both spatial features and temporal dependencies from the data. An attention mechanism is added to focus on the most important features,improving the prediction accuracy of the model. Finally,the improved dung beetle optimization (IDBO) algorithm is used to optimize the hyper-parameters of the model. This step further enhances the model performance.ResultsSimulation experiments were conducted using full-scale data from the Yukun ship. The results show that the proposed prediction model has a root mean square error reduction of about 78.25% and an increase of about 65.63% reliability compared with TCN.DiscussionThe  model outperforms traditional methods in terms of accuracy and stability. This demonstrates its potential for improving the prediction of ship motion an attitude.