Objective Parametric roll threatens both cargo and personnel safety. The International Maritime Organization (IMO) has established the second-generation intact stability criteria and operational guidelines for ships to prevent it. Forecasting of parametric rolling in irregular waves is crucial for real-time ship maneuvering.

Methods The C11 container ship model experienced parametric rolling in sea state 7 conditions. The experimental data were obtained through model basin tests conducted at the China Ship Scientific Research Center. A time-history data set of parametric rolling was created. A data-driven model using LSTM neural network was developed. Nonlinear features were identified through inputting historical periodic data, and the neural network was employed to optimize temporal relationships.

Results The hyper-parameters of the forecasting model were trained with the objective of maximizing the accuracy of rolling motion predictions. The training and validation sets show that the LSTM model woks for forecasting parametric rolling in irregular waves. The minimum mean absolute error for rolling motion forecasting was 0.128, and the minimum forecasting error for the maximum rolling angle was 0.62% on the test set.

Conclusion The model forecasts parametric roll with high accuracy, providing a technical approach for real-time prediction of ship parametric rolling and stability warning.