Project Description
Real-time prediction of the incident wave train or vessel motions on a wave-by-wave basis is of great interest to the offshore industry, facilitating safe and efficient ship to ship transfers, offshore lifting, active control of energy extraction devices, walk-to-work vessel services, etc., but has remained challenging to implement. While linear wave theories can provide initial estimates of the water level or vessel motions, the real sea is non-linear and, therefore, the inclusion of higher order harmonics are generally required for accurate and reliable predictions. Given the complexity of the underlying hydrodynamics, the inclusion of machine learning may potentially offer an efficient means of obtaining the predictions.
This project investigated the capacity of different machine learning models. The project is the first step in providing real-time wave forecasts that can facilitate safe and efficient ship to ship transfers, offshore lifting operations, active control of energy extraction devices, and walk-to-work vessel services.
A machine learning model (LSTM) was identified and run successfully on Matlab. Code was also developed to decompose the linear and second harmonics of the waves. Initial results using tank testing data show good linear wave-by-wave predictions, and parametric studies are conducted to identify the most important parameters that could improve the accuracy in prediction.