Wave forecasting for offshore wind farms using machine learning

Abstract:

Wave forecasting is an important parameter for performing maintenance in offshore wind farms. In order to minimize costs linked with wasted trips to perform maintenance or wasted opportunities to repair offshore wind farms an accurate wave forecast is necessary. However, numerical models operate in larger grid cells, covering a large area and providing a unified forecast for the whole wind farm, which does not correspond to reality. In this work data from multiple wave radars throughout a wind farm were used together with a machine learning methodology to deliver a high spatial resolution forecast for stop and go operations in offshore wind farms. Using this type of forecasting model, it is possible to take advantage of regions sheltered from rougher sea state, due to interactions between structures and waves, under specific condition. Wave radars are place in different locations correspond to different level of sheltering from neighboring wind turbines, bathymetry and proximity to the coast. The forecasting capabilities using different amount of wave radar data and different time windows are discussed for different architectures of artificial neural networks. Finally, the number of available real and forecasted maintenance windows for different types of vehicles used for maintenance and for the different locations throughout the wind farm are calculated in order to define the accuracy of the model, in the scope of spatially distributed offshore wind farm accessibility.