Back to search

IKTPLUSS-IKT og digital innovasjon

Next Minutes prediction system for ocean waves and vessel motions based on physics-informed neural networks

Alternative title: Neste minutters prediksjon av havbølger og fartøysbevegelser basert på fysikkbaserte nevrale nettverk.

Awarded: NOK 12.0 mill.

Ocean wave dynamics are the principal challenge in all operations at sea. Marine operations such as dynamic positioning (DP), offshore crane operations, subsea and wind farm installation and service, and sea transportation are complex operations that are highly sensitive to ocean wave dynamics. The main objective of this project is to use physics-informed neural networks to further develop and improve next minutes predictions of ocean waves, wave forces and vessel motions to increase accuracy, reliability and ahead prediction time for higher and more complex seas, using the standard onboard X-band navigation radars on the vessel. We will develop the world’s first next minutes wave and vessel motion prediction system, only using existing navigation radars, thoroughly tested and ready to be commercialized into new products and services. By developing auto calibration only software installation will be needed. To check and further develop the next minutes predictions, the incoming waves will be measured by the MIROS wave radar already installed in the bow of the DOF vessels Skandi Africa and Skandi Vega. The corresponding vessel motions will be measured by the onboard Kongsberg MRU (Motion Response unit) system. NORCE has in cooperation with MIT Ocean Research already developed the RIMARC (Radar IMAge ReConstruction and wave prediction) system capable of forecasting deterministic (i.e. phase-resolved) wave and vessel motions in real-time 4-8 minutes in advance from standard onboard X-band navigation radars for DP (Dynamic positioning) standstill operations. The even greater challenge in this project is to predict next minutes waves and vessel motion also for free sailing operations. The system can in the future be further developed to rapidly learn by physics based machine learning how to operate a vessel in an optimum way in high sea states, both regarding safety and performance. This is a necessary step towards development of autonomous vessels and operations.

Ocean wave dynamics is the main challenge for all ocean operations. Ocean operations nowadays are based on subjective assessments due to lack of sufficient wave pattern data. Marine operations such as dynamic positioning (DP), offshore crane operations, subsea and wind farm installation and service, and sea transportation are complex operations which are highly sensitive to ocean wave dynamics. At the same time artificial intelligence (AI) methodologies have shown the power to transform large volumes of sensor data from this complex environment into detailed real-time information and actionable knowledge. NORCE has in cooperation with MIT Ocean Research newly developed the RIMARC (Radar IMAge ReConstruction and wave prediction) system capable of forecasting deterministic (i.e. phase-resolved) wave and vessel motions in real-time 4-8 minutes in advance from standard onboard X-band navigation radars for DP under standstill operations. However, the existing linear models cannot predict high wave steepness accurately, and the advanced nonlinear higher-order models require too long computation time to be used onboard operating vessels. Then, in this project we will develop new physics-informed neural networks (PINN) based prediction methods that incorporate physical equations to perform next minutes prediction of ocean wave pattern, wave forces and vessel motions in real-time. The most critical R&D challenges is to be able to develop a system to be of practical use. Then the prediction systems must be accurate, robust and reliable. An even greater challenge is to predict next minutes wave and vessel motion also for free sailing operations. Because we already have developed the RIMARC software for next minutes prediction of waves, that already shows good correlation with full-scale testing in DP (standstill) mode, a prediction model with very high accuracy and reliability is expected to be developed by using new PINN methods.

Funding scheme:

IKTPLUSS-IKT og digital innovasjon