Real-time Prediction of Parametric Roll Motion via Power-activation Feed-forward Neural Network with Model Experiment Data

Abstract

The Power-activation Feed-forward Neural Network (PFN) is used to achieve real-time prediction of the ship’s parametric roll motion. The theoretical rationality of real-time prediction based on the ship’s rolling motion time series data is verified. Sequence-to-Sequence models are proposed and used to compare the PFN model, Long Short-Term Memory model, and Convolutional Neural Network. Three different groups of model experiment data are used for comparison. Results show that PFN has advantages in real-time prediction of parametric roll motion due to its time-varying weight adjustment methods, with a more effective mapping mode, higher accuracy, and shorter computing time.