Measurement and Prediction Scheme for Ship Heave Motion Based on UKF and PSR-LSTM

The floating cranes used to install large wind turbines are at risk of instability and collision because of the dynamic and changing sea conditions. Ship-motion measurement and prediction of heave is applied for wave compensation, but a high prediction accuracy is difficult because of measurement drift and phase advance. To address this issue, a ship heave motion measurement and prediction scheme based on unscented Kalman filter, phase space reconstruction and long short-term memory (LSTM) is proposed. First, the Stewart platform is used to simulate ship motions, and data are collected by inertial sensors and are processed by unscented Kalman filter. Second, the mutual information is applied to calculate the delay of the ship-motion time series, and the Cao method is used to obtain the embedding dimension, then phase space reconstruction is carried out. Third, the reconstructed data are trained by the multidimensional LSTM to predict extremely short-term ship motions. Simulation results showed that the proposed method provides better measurement and better prediction accuracy, especially under more complex sea conditions.