MPC of Variable Speed Wind Turbines

Abstract

With the expansion of wind energy systems, there is placed increased scrutinyon the control systems of these systems. Small improvements in the controllerscan mean much in terms of increased power production and reducedmaintenance. As such, the predominant simple control systems are little by littlebeing replaced by more advanced, model-based approaches.The model predictive controller (MPC) has since 2007 been suggested as ameans of wind turbine control. In this report, realistic simulations of a windturbine is used to compare the performance of the MPC against a traditionalcontroller. The MPC is shown to impose much tighter control on the relevantoutputs. However, this does barely result in an increased power output, asalmost all the gains are lost by the virtue of the aerodynamical properties ofrotor blades and the drive-train.To make predictions of future wind speeds is a possible way to improve theperformance of the model predictive controller. Purely autoregressive modelswere implemented to make very short-term wind speed predictions, and wereable to increase the accuracy from a persistent approach by approximately 20%.Whether this wind speed prediction does improve the model predictive controlleris debatable. The controller performs slightly better according to some criterias,and slightly worse in others. It is especially harrowed by a very hard utilizationof the inputs. This is caused by the fact that the predictions have a tendency toovershoot, and predict more turbulence that what really occurs.