Model‐based receding horizon control of wind farms for secondary frequency regulation

Abstract

In this study, we propose the use of model‐based receding horizon control to enable a wind farm to provide secondary frequency regulation for a power grid. The controller is built by first proposing a time‐varying one‐dimensional wake model, which is validated against large eddy simulations of a wind farm at startup. This wake model is then used as a plant model for a closed‐loop receding horizon controller that uses wind speed measurements at each turbine as feedback. The control method is tested in large eddy simulations with actuator disk wind turbine models representing an 84‐turbine wind farm that aims to track sample frequency regulation reference signals spanning 40 min time intervals. This type of control generally requires wind turbines to reduce their power set points or curtail wind power output (derate the power output) by the same amount as the maximum upward variation in power level required by the reference signal. However, our control approach provides good tracking performance in the test system considered with only a 4% derate for a regulation signal with an 8% maximum upward variation. This performance improvement has the potential to reduce the opportunity cost associated with lost revenue in the bulk power market that is typically associated with providing frequency regulation services. Copyright © 2017 John Wiley & Sons, Ltd.

Online model calibration for a simplified LES model in pursuit of real-time closed-loop wind farm control

“…Furthermore,5 time-ahead predictions with these models are limited to the time-invariant steady-state, limiting their use for APC. 1 There is a smaller yet significant number of dynamic surrogate wind farm models (e.g., Munters and Meyers, 2017;Boersma et al, 2017b;Shapiro et al, 2017a), which attempt to model the dominant temporal dynamics inside the farm. These models can be used for control on the seconds-scale, and furthermore allow time-ahead predictions, some even under changing atmospheric conditions.…”

“…These models can be used for control on the seconds-scale, and furthermore allow time-ahead predictions, some even under changing atmospheric conditions. Specifically, the dynamic surrogate model employed in Shapiro et al (2017a) is computationally feasible, but only models the 10 flow in one dimension, and furthermore allows no turbine yaw or changes in the wind direction, limiting its applicability.…”

“…designed a traditional Kalman filter (KF) for their low-fidelity "FLORIDyn" model, showing marginal improvements compared to optimization using a static model. Shapiro et al (2017a) present a one-dimensional 25 dynamic wake model used with receding horizon control for secondary frequency regulation, using an estimation algorithm following Doekemeijer et al (2016). Furthermore, Iungo et al (2015) used dynamic mode decomposition to obtain a reducedorder model of the wind farm dynamics, which was then combined with a traditional KF for state estimation.…”

Wind farms providing secondary frequency regulation: evaluating the performance of model-based receding horizon control

“…Our recent work (Shapiro et al, 2017a) sought to overcome these challenges by developing a time-varying extension of the classic Jensen wake model (Katić et al, 1986) that accounts for the dynamics of wake advection through the farm. This new wake model was incorporated into a predictive model-based receding horizon control framework to coordinate an array of wind turbines to provide secondary frequency regulation by modulating the thrust coefficients of individual turbines.…”

“…Feedback from measurements of the velocity at each turbine was used to correct modeling errors. This approach showed promising results when tested in a large-eddy simulation (LES) model of a wind farm in which turbines were represented using actuator disk models (Shapiro et al, 2017a). In these simulations, we used set-point reductions of only 50 % of the maximum regulation provided, but were able to track a sample regulation signal with the wind farm test system used.…”