Assessment of trimodal wave spectral parameters using machine learning methods and vessel response statistics to enhance safety of marine operations

Safe execution of a Marine operation (MO) depends on accurate values of operational limits and forecasting quality of environmental parameters, especially wave spectral parameters (significant wave height, peak period, and peak direction). Since offshore sites are generally characterized by the presence of more than one wave system e.g. wind seas and swells, it becomes challenging to assess accurate multimodal spectral parameters. Apart from forecasts, buoy or vessel motions can be used to estimate wave spectral parameters. This paper deals with an assessment of trimodal wave spectral parameters using three machine learning (ML) methods, i.e. Random forest, Extra trees, and Convolutional neural networks, for an offshore site in the Norwegian Sea. The methods use synthetic statistics of vessel responses as features and spectral parameters computed from a spectrum partitioning method, for training the ML algorithms. It is found that the Extra trees method has the best performance and is very accurate for predicting spectral parameters of three wave systems. Findings from this paper can be used to further develop efficient on-board safety systems and help superintendents make more informed decisions.