Identification of a dynamic center line model and its implementation in FLORIDyn

Abstract

Wind energy plays an essential role in the transition to a sustainable future and wind turbines allow us to utilize it. Wind turbines are often installed in close proximity to make the best use of the given space and to save costs. This also has the disadvantage that the turbines influence each other due to wakes. This is where wind farm control strategies can help to increase the power generated by the farm again. Model based wake steering is a promising strategy which utilizes a surrogate model to determine the best set-points for the wind farm. FLORIDyn is one surrogate model which provides basic flow dynamics at a very low computational cost.

In this thesis we increase the fidelity of FLORIDyn by extracting features from high fidelity simulations and extending the model. A crucial element of the wake description is the center line, especially for wake steering. We present a pipeline which 1.) extracts the center line from given flow field data, 2.) converts the center line into FLORIDyn model inputs, 3.) identifies a state-space model to translate the low fidelity center line behaviour to one closer to the high fidelity, and eventually 4.) extends the model.

To achieve this, we test and compare DMDc and MOESP system identification methods in two simulation cases with a 10 deg and 20 deg yaw step. The results indicate that the DMDc implementation has a better fit than the MOESP models. To get an idea of the general behaviour, we also test the models in a 15 deg case they have not been trained for.

Previously, the FLORIS and FLORIDyn parameters had only been trained for steady state conditions. This study can be seen as a proof-of-work that shows that dynamic extensions for FLORIDyn are possible, and how to perform this for the center line. The DMDc system identification showed to be a promising tool in identifying high fidelity dynamics from SOWFA simulations, due to its ability of identifying large scale physical systems. Research on the implementation of the FLORIDyn model in a model based wake steering framework is required to make a trade-off between the need for a more dynamical model, and following additional computational cost.

The current application of the framework is limited to the use of two dimensional SOWFA flow field snapshots at hub height, a turbulence free environment, the use of a single wind velocity and single yaw step data in the system identification. Further research and more extensive training data is required to broaden the framework to more real-life scenarios. Furthermore, the computational efficiency of the identified DMDc model can still be improved by decreasing the amount of center line states, making it better applicable in a model based framework.