Project description
Within the joint project LibaRetion, a holistic, model-based control concept for modern wind turbines (WTs) is being developed and validated in practice. Using a lidar-assisted wind field prediction, the future behavior of the turbine is to be pro-actively optimized: based on a model of the control plant simulated at runtime, the optimal actuating variables are computed and applied, so that the turbine can already respond to an approaching wind field before it reaches the rotor.
The sub-project presented here develops the data-driven core of this wind field prediction. Using methods from machine learning, an algorithm is developed that fuses a large number of measurement signals to enable a robust and adaptive estimate of the wind field with a look-ahead time of a few seconds. This is based on measurements from a lidar system positioned in front of the rotor, which are processed into a prediction of the wind field approaching the turbine. In addition, turbine signals are used to reconstruct, after the fact, which wind field actually acted on the turbine. By comparing the prediction with the actual effect, the algorithm continuously learns and improves its estimation quality during ongoing operation.
The following research questions, among others, are addressed in developing the data-driven wind field prediction:
- How can lidar measurements and turbine signals be fused such that a robust wind field estimate remains possible even with measurement gaps, noise, or limited measurement geometry?
- How can the quality or reliability of a wind field prediction be continuously assessed, so that it can be fed into a higher-level control system as a quality metric?
- How can a self-learning algorithm be continuously improved at runtime by comparing the prediction with the wind field that actually acted on the turbine, without compromising real-time capability?
- How can the wind field prediction be adaptively adjusted to different atmospheric conditions?
- How can the developed prediction be designed so that it can be integrated into a model predictive control scheme and deliver genuine added value there compared to classical feedforward approaches?
The developed data-driven wind field prediction will be tested on two wind turbines over the course of the project and integrated, together with the project partners, into a model predictive control system. This will allow the effectiveness of the overall concept from the lidar measurement through the wind field prediction to the optimized actuation of the turbine to be validated under real operating conditions.
Preceding project
Funding
