System identification

Conventionally, wind turbine control design is performed based on physical modeling of the turbine's dynamics. However, due to the uncertainty in the physical parameters involved in such ``first-principles'' models, the control design is performed by keeping conservative stability margins, such as a gain margin of two and a phase margin of 45 degrees. A noticeable improvement of the performance of the wind turbine controllers is expected to be achievable when more accurate (i.e. less uncertain) models are available. One way to get increased model accuracy is by following an orthogonal approach to the physical modeling, namely by attempting to model the phenomena observed in reality by using measured data from the (operational) wind turbine. This is possible by using system identification that aims at fitting the parameters of a so-called ``black-box'' model to the measured data. This results in models that try to explain the data rather than the physical phenomenon.

Within the We@Sea project “System identification: an essential step for bringing advanced control algorithms into practice” a number of identification algorithms have been developed and optimized for the estimation of control-relevant wind turbine dynamics. In particular, attention has been paid on turbine model identification for the purpose of basic rotor speed control by collective pitch, drive train damping by generator torque, tower for-aft damping by collective pitch, and tower sideward motion damping by generator torque. These models can be used for either improving the existing control loops (e.g. by fine-tuning the underlying controller/filter parameters), or for achieving additional functionality by designing control ``add-ons'' for fatigue reduction. The focus is on closed-loop identification techniques that give the possibility of identifying open-loop models from input/output data collected in the presence of the basic turbine controller. An important condition for achieving consistent identification results is to additionally excite the turbine inputs (blade pitch angles and/or generator torque set-points), at least in the frequency region of interest. Special attention is paid on the design of excitation signals that achieve enough excitation for system identification without noticeable increase of the loads. Furthermore, different methods for validating the identified model have been developed and used at the different stages in the project.