An adaptive real-time ship roll motion prediction scheme based on two-stage multi-resolution decomposition

Real-time prediction of ship roll motion is crucial for enhancing marine safety and efficiency. To address the complex characteristics of ship roll dynamics, including nonlinearity, time-varying dynamics, and uncertainty induced by environmental disturbances and sailing conditions, an adaptive real-time ship roll neural prediction scheme is proposed based on a two-stage decomposition framework integrating empirical mode decomposition (EMD) and discrete wavelet transformation (DWT). The multi-resolution decomposition capabilities of EMD and DWT are combined with variable neural networks to achieve robust prediction performance. The decomposition order and the prediction model input order are adaptively determined based on EMD and Lipschitz quotients methods, respectively. The adaptability of the neural prediction scheme is enhanced with the network dimension, hidden units’ locations, and connecting parameters being real-time adjusted in a sequential learning mode. The two-stage EMD-DWT transformation and the parallel neural prediction strategies ensure the accuracy and stability of the prediction, and the sequential learning strategy of sliding data window enables fast processing speed and adaptability to time-varying dynamics. The feasibility and effectiveness of the proposed ship roll prediction scheme are validated through simulations based on the measured data of the real ship trial.