finalRAVE-ML

Motivation: computationally efficient generation of improved sea state forecasts to support operational decision-making

Reliable sea state forecasts are essential for the safe and efficient planning of offshore activities such as maintenance, logistics, and future decommissioning of offshore wind farms. Although operational numerical wave models provide valuable forecasts, their accuracy is limited by uncertainties in atmospheric forcing, parameterized physical processes, and spatial resolution.
Building on the results of the OpenRAVE project, finalRAVE-ML extends an existing data assimilation approach by integrating quality-controlled measurements from offshore wind farms into data-driven prediction models. The objective is to generate computationally efficient, spatially distributed nowcasts and short-range forecasts that complement conventional numerical models and provide improved decision support for offshore operations.

Data, modeling, and model calibration

The project combines wave, current and wind measurements with numerical wave model data within a scalable data infrastructure. Based on these datasets, machine learning models-including Random Forest regression and hybrid Graph Neural Network (GNN)-LSTM and CNN-LSTM architectures-are developed to represent both spatial and temporal sea state dynamics.
A two-stage modeling strategy combines an offshore model driven by assimilated observations with a higher-resolution coastal model to improve the representation of nearshore processes while maintaining computational efficiency.
Model performance is validated using measurements of vertical vessel accelerations recorded on Crew Transfer Vessels (CTVs) and Service Operation Vessels (SOVs) during real offshore operations, providing a direct assessment of operational relevance.

Project information

• Project duration: 01 June 2025 – 31 May 2028
• Partners: LuFI, BSH, DNV, Fraunhofer IWES