Data-driven modelling of wind turbines

In this paper we present a novel approach that allows global modelling of the power control dynamics of a wind turbine based on measured data. The approach is based on the assumption that all the nonlinearities over the operating range arise from a static aerodynamic mapping, which is interconnected with linear, time-invariant dynamics. This so called Hammerstein structure is exploited to simplify the model identification procedure. The global model is suited to control design methods such as model predictive control or can be used to extract local linear models. The approach is demonstrated on a benchmark example, the 5MW NREL/Upwind reference turbine and is shown to work well. Tools from convex optimisation and the recently introduced nuclear norm techniques prove to be instrumental to the successful implementation of the algorithms.