Model Predictive Controllers for Reduction of Mechanical Fatigue in Wind Farms

Abstract

We consider the problem of dispatching wind farm (WF) power demand to individual wind turbines (WTs) with the goal of minimizing mechanical stresses. We assume that wind is strong enough to let each WTs produce the required power and propose different closed-loop model predictive control (MPC) dispatching algorithms. Similar to the existing approaches based on MPC, our methods do not require to replace WT hardware components, but only software changes in the supervisory control and data acquisition (SCADA) system or integration with the middleware system of the WF. However, differently from other MPC schemes, we augment the model of a WT with an auto regressive moving average (ARMA) predictor of the wind turbulence, which captures the wind dynamics over the MPC control horizon. This allows us to develop both stochastic and deterministic MPC algorithms. In order to compare different MPC schemes and demonstrate improvements with respect to classic open-loop schedulers, we performed simulations using the SimWindFarm toolbox for MATLAB. We show that MPC controllers allow to achieve reduction of stresses even in the case of large installations, such as the 100-WTs Thanet offshore WF.