Predictive and stochastic reduced-order modeling of wind turbine wake dynamics

Abstract

Abstract. This article presents a reduced-order model of the highly turbulent wind turbine wake dynamics. The model is derived using a large eddy simulation (LES) database, which cover a range of different wind speeds. The model consists of several sub-models: (1) dimensionality reduction using proper orthogonal decomposition (POD) on the global database, (2) projection in modal coordinates to get time series of the dynamics, (3) interpolation over the parameter space that enables the prediction of unseen cases, and (4) stochastic time series generation to generalize the modal dynamics based on spectral analysis. The model is validated against an unseen LES case in terms of the modal time series properties as well as turbine performance and aero-elastic responses. The reduced-order model provides LES accuracy and comparable distributions of all channels. Furthermore, the model provides substantial insights about the underlying flow physics, how these change with respect to the thrust coefficient CT, and whether the model is constructed for single wake or deep array conditions. The predictive and stochastic capabilities of the reduced-order model can effectively be viewed as a generalization of a LES for statistically stationary flows, and the model framework can be applied to other flow cases than wake dynamics behind wind turbines.