Posters

Abstract

Major concerns arise about the feasibility and scalability of current operations and maintenance strategies for floating wind. The maintenance of floating offshore wind turbines faces challenges that are different from fixed offshore wind and could lead to significantly higher costs1. In this context, precise and probabilistic weather forecasts are required to ensure the security of the operations and to optimize their planning and execution. Numerical Weather Prediction (NWP) models show limitations in ultra-short-term forecasting and uncertainty estimation due to their high computational cost. Deep learning on the other hand can be used to complement NWP on specific tasks due to design feasibility and relatively low computational costs. Combining NWP forecasts of wind around the site of interest, and a collection of in-situ measurements, the deep learning architecture can help assimilate recent measurements to correct the forecast in the short-term. Using a likelihood criterion as loss function, it can be trained to predict both the mean value and the uncertainty of a variable. It can furthermore be trained for the joint forecasting of met-ocean variables of interest of offshore wind operations. Such multivariate probabilistic forecasts can then be used for the planification and execution of maintenance operations. This work presents the application of deep learning methods for the joint probabilistic forecast of 2D wind at an un-seen offshore location, mimicking an offshore maintenance operation setup. Convolutional models are compared to Recurrent models and statistical baselines using both deterministic and probabilistic criterion. These methods are then scored on an operational sequence of a Major Component Replacement for floating offshore wind, at a site representative of floating wind conditions in the gulf of Lion in the French Mediterranean Sea. The goal is to assess the value of the different forecasts - deterministic and probabilistic - according to a metric closer to the real cost function of maintenance operations. The presented work is part of the collaborative research program FLOWTOM (FLoating Offshore Wind Turbines Operations and Maintenance) led by France Energies Marines.