Uncertainty quantification for wind energy applications

Uncertainties are omni-present in wind energy applications, both in external conditions (such as wind and waves) as well as in the models that are used to predict key quantities such as costs, energy yield, and fatigue loads. This report summarizes and reviews the application of uncertainty quantification techniques to wind energy problems. In the wind industry, including uncertainties in predictions has classically been done by using Monte Carlo methods. Recently, more advanced methods have been considered (e.g. polynomial chaos expansion, stochastic collocation, and Gaussian process regression), which are based on smartly sampling the model (e.g. a complex aerodynamic blade model). These methods generally have a greater efficiency compared to Monte Carlo (depending on model properties) and additionally yield computationally cheap surrogate models. Furthermore, surrogate models purely based on data (e.g. via proper orthogonal decomposition) have received significant interest, especially for the representation of turbulent wind turbine wakes. Both model-driven and data-driven surrogate models play a crucial role in making control and optimization studies feasible. In the near future, we expect that recent trends in uncertainty quantification, namely Bayesian model calibration and optimization under uncertainty, will become increasingly popular in wind energy applications.