Machine Learning Based Prediction of Wave-Induced Vessel Response

 Abstract

 “Waiting on weather” is a costly restraint on offshore vessel operability. Vessel operating windows are determined based on the relationships between the weather and vessel movement, and uncertainties in these predictions may result in vessel operations being ceased prematurely. To improve the efficiency of offshore operations, existing assumptions and calculations based on conventional response amplitude operators (RAOs) should be challenged and improved. A machine learning approach is presented as a means of enriching these conventional RAOs with data.

The machine learning model uses sea state forecasts to predict vessel response spectra. The model is cleverly formulated to use any existing RAO as a fallback solution in the absence
of sufficient data. When applied to a comprehensive real-world scenario, the model predominantly outperforms the “best” available existing RAO. The results can be used not only to improve wave-vessel response predictions, but also to improve our understanding of existing RAOs and their shortcomings. Ultimately, the work can contribute to reducing overconservatism in weather-based restrictions on offshore vessel operability.

Authors

Ali Cetin
Tech Lead Data Lab at 4Subsea

Vegard Rørvik Solum
Data Scientist at 4Subsea

Christina M. Evans
Subsea7