Session: 06-17-01 AI Technology for Ocean Engineering

Submission Number: 154981

A Data-Driven Approach to Offshore Wind Forecasting in the Celtic Sea 

Accurate weather forecasting is crucial in supporting various industry sectors, including offshore wind which is central to global net zero energy generation ambitions. In this space, like many others, we are entering a new age of opportunity where machine learning models, trained on historical data and observations, can be deployed in place of directly solving the physics-based equations that govern atmospheric and ocean dynamics. Such models exploit the vast historical datasets to learn patterns which may not always be well represented explicitly by physical equations. These machine learning, observation-based, weather forecasting methods offer the potential to both increase accuracy and efficiency of weather forecasting compared to traditional Numerical Weather Prediction (NWP).

In this study, we adapt and apply a previously developed machine learning framework for low-cost operational forecasting (MaLCOM) to offshore wind forecasting in the Celtic Sea. Our approach uses an attention-based long short-term memory (LSTM) recurrent neural network to learn temporal patterns from a network of observations. This is combined with a random forest-based spatial nowcasting model, trained on ERA5 reanalysis data, to provide a complete spatiotemporal prediction for the region. We integrate winds derived from wave spectra measured by wave buoys, demonstrating the framework’s value even with imperfect input data.

We validate this new capability for short-range prediction of 10 m wind speeds in the Celtic Sea, using independent observations from floating lidar units deployed by Celtic Sea Power in 2022. Results confirm the framework’s flexibility and suitability for regional wind prediction, showing reasonably good forecasts despite imperfect input data.

This work extends our previous machine learning-based predictions of regional ocean wave and current conditions to operational wind forecasting in the Celtic Sea. The novel application of this technology demonstrates the potential for a fundamentally different approach to generating and using metocean data, particularly wind forecasts, in the offshore renewable energy sector. Furthermore, these lightweight, data-driven predictions can be run on-demand using standard computers, offering new opportunities for improving real-time decision-making to support offshore planning and workability.

Presenting Author: Ajit Pillai University of Exeter

Presenting Author Biography: Ajit is a Senior Lecturer in Autonomous Systems and Robotics with a research focus on the development and deployment of optimization algorithms to aid in the design of offshore renewable energy. He holds a Research Fellowship from the Royal Academy of Engineering to develop new techniques to integrate numerical physics-based models with targeted, dynamic measurement campaigns to reduce offshore uncertainty and develop a new framework for spatial data.

He has led and contributed to several EPSRC and European projects applying optimization and machine learning to improve the design and operation of offshore renewable energy devices, as well as developing new hydrodynamic approaches for modelling offshore systems. Prior to joining the University of Exeter, Ajit obtained an EngD in offshore renewable energy through the Industrial Doctoral Centre for Offshore Renewable Energy (IDCORE). His EngD research, completed in partnership with EDF Energy R&D UK Centre, led to the development of a methodology and tool for the optimization of offshore wind farm layouts considering the sites and constraints relevant for future gigawatt scale wind farms in European waters. Ajit also holds an MSc in Sustainable Energy Systems from The University of Edinburgh and a BSc in Mechanical Engineering from Columbia University.