On Practical tuning of Model Uncertainty in Wind Turbine Model Predictive Control

Abstract

Model predictive control (MPC) has in previous works been applied on wind turbines with promising results. These results apply linear MPC, i.e., linear models linearized at different operational points depending on the wind speed. The linearized models are derived from a nonlinear first principles model of a wind turbine. In this paper, we investigate the impact of this approach on the performance of a wind turbine. In particular, we focus on the most non-linear operational ranges of a wind turbine. The MPC controller is designed for, tested, and evaluated at an industrial high fidelity wind turbine simulator. We show how the linearization results in modeling errors, leading to major performance issues, which in the worst cases cause safety shutdowns of the wind turbine due to, e.g., too high rotor speeds. We propose an approach in which this problem is handled by adjusting relevant model parameters in the linearized model to fit the actual physical wind turbine behavior. We evaluate the MPC with the different model parameters, and show that, e.g., over-speed events are avoided, and a good performance of the wind turbine control is obtained.