Ship Heave Motion Prediction Using an Integrated EMD and Sequential Learning Approach

Precise estimation of ship’s heave movement is essential for the efficiency and safety of marine actions. To extract the underlying message in ship heave motion time series thus improving heave motion prediction accuracy, an online sequential prediction mechanism for ship heave movement estimation is presented within a empirical mode decomposition framework. The time series of ship heave movement is decomposed into sub-series with homogeneous nature, and the obtained components are discerned and estimated by variable neural networks whose parameters are optimized by Bayesian optimization method. To achieve a more adaptive and rapid prediction strategy, a real-time updated sliding data window is utilized as system observer. Simulation of ship heave movement prediction is conducted using the measured data of the ship Yukun and the results demonstrated that the presented neural prediction strategy is effective in processing speed and prediction precision.