Posters

Abstract

Representative turbulence intensity (TI) and its derived quantity, effective turbulence intensity, are metrics used in fatigue structural integrity assessment of wind turbines (WT). The representative TI at a WT location is defined as the sum of the mean turbulence and 1.28 times the standard deviation (SDV) of turbulence. Both the mean and SDV are obtained by transferring climatology measurements from the location of a mast to the WT location using a CFD model. When multiple climatologies are available, the standard practice in most wind energy software packages is to assign only one climatology to each WT position, typically the closest or most representative one. However, if measurements differ significantly between masts, the representative TI will not be evenly distributed across the wind farm. Instead, it will exhibit sharp transitions between WTs associated with different climatologies, even when those WTs are located relatively close to one another. The approach proposed in this study defines the representative TI at each WT position as a weighted linear combination of the transferred TI from each climatology at each 10-minute measurement, with the weighting factor being the inverse distance between the WT and each mast. While the mean TI of the weighted timeseries is equal to the weighted mean of all transferred climatologies, the SDV of the weighted timeseries underestimates the true SDV. This study proposes a method to correct the weighted SDV. The approach is based on the Central Limit Theorem and the Delta method. It assumes that all transferred climatologies can be regarded as statistical samples taken from the same population. Ideally, they should have the same SDV, since they all describe the same true population parameter and representative TI at the WT location. Under this assumption, a relationship is derived between the weighted SDV and the covariances of the transferred climatologies, which simplifies to a new definition of the SDV that better represents the true TI conditions at the WT location. The method was tested in five cases: one idealized case with flat terrain and four real-world cases with increasing terrain complexity. The idealized case was evaluated using combinations of climatologies with varying degrees of cross-correlation. The real-world cases were taken from real sites and feature measurement data from several climatologies. In each case, several WT locations were examined. The uncorrected SDV was consistently underestimated; however, after applying the correction method, both the SDV and representative TI were closer to the linear weighted average of all transferred climatologies. Furthermore, for the real-world cases, a round-robin test was conducted by removing one climatology at a time and predicting the SDV and representative TI at its location with two different methods, one using the closest climatology and another using our approach of weighting all remaining climatologies. The results were compared with measurements and a bias was calculated. In all cases, weighting the climatologies produces lower biases when using the weighted SDV than the one from the closest mast, indicating that our method improves the estimation of representative turbulence.