A Study on Horizon Length for Preview-enabled Model Predictive Control of Wind Turbines* This work was supported in part by Envision Energy, the Hanse-Wissenschaftskolleg in Delmenhorst, Germany, and a Palmer Endowed Chair at the University of Colorado Boulder.

While a growing body of research into model predictive control (MPC) for wind turbines shows that MPC can outperform baseline controllers, literature comparing various MPC formulations is scarce. In this paper, we compare MPC based on numerical linear time-invariant (LTI) and linear parameter-varying (LPV) models with differing prediction horizons. Our MPC formulation includes constraints on the turbine control inputs and explicitly handles preview disturbance and scheduling parameter information provided by lidar. Unsurprisingly, when simulated on a nonlinear model of a wind turbine, the LPV-based controller generally performs better than its LTI counterpart. Further, longer prediction horizons lead to improved performance in the LTI case. However, we find that for the LPV-based MPC, there is no clear improvement in performance with horizon length, with short horizons performing similarly (and in some metrics better) than long horizons. We discuss potential takeaways from this surprising result and its implications for the use of lidar-enabled MPC for wind turbines.