Posters

Abstract

Turbulence intensity (TI) is a key variable for wind farm development. Throughout the various stages of wind farm development, accurate, site-wide TI information is required for site selection, wind turbine suitability and estimation of aerodynamic losses. However, obtaining reliable TI measurements in the field requires the installation and maintenance of tall masts over prolonged periods of time. The costs associated with such measurement campaigns, make high quality TI measurements sparse in space and time. Especially offshore (where the nearest mast-based TI measurement might easily be hundreds of kilometers from the development zone) and in complex terrain (where TI is strongly heterogenous), the density of reliable TI measurements is generally too low for an accurate, site-wide assessment. On the contrary, advances in GPU computing allow large-eddy simulation (LES) to provide site-wide TI at high spatial resolution (<100 m) for a fraction of the cost of a single measurement setup. But the uncertainty of LES predictions of TI is higher than measurements. Therefore, we propose to combine the high-density spatial information of LES with low uncertainty, but point-based, measurements using machine learning (ML) techniques to obtain measurement-corrected, site-wide TI predictions. The accuracy of the combined LES-ML methodology is assessed by means of a leave-one-mast-out cross validation for 30+ offshore and onshore sites of various complexity. Additionally, the uncertainty of the methodology for different site types (offshore, onshore, complex) is assessed and benchmarked against pure LES and a direct extrapolation of nearby measurements. Both mean TI and 90th percentile TI by wind speed bin are considered, as well as the sectorial and temporal (diurnal/seasonal) variations.