Due to change in the wake characteristics during curtailment, the estimation of the available power of a wind farm under downregulation is a complex process, especially for high resolution data. The state of the art industry practice is to aggregate the possible (or available) power signals from the individual turbine SCADA, which does not take into account the reduction in the wake losses thus overestimates the production capacity. Here we present and validate the PossPOW Algorithm which corrects the "downregulated wake", and shown to perform significantly better in a series of curtailment experiments in the Horns RevI wind farm, compared to the current practice.


The PossPOW algorithm mainly consists of three parts  see the figure to the left
 Estimation of the local wind speed at the upstream locations (as they are not affected by the downregulated wake),
 Advection of the upstreamwindthrough downstreamvia a wake model (to replace the reduced wake with the nominal velocity deficit) and calculate the production capacity, and
 Test the estimated wind farm scale possible power and validate the algorithm.

The approach is designed to work with 1second SCADA data aiming to calculate the realtime wind farm power curve as well, which makes the algorithm suitable for operation monitoring and can be further implemented in smart control strategies.
The wind speed at the turbine locations are estimated using the pitch, rotational speed and active power from the turbines together with a generic approximation of the power coefficient, Cp, in terms of pitch angle and tip speed ratio. The method is applied and validated on HornsRevI, Thanet and Lillgrund offshore wind farms as well as the NREL 5MW simulations, under both normal and curtailed operations.
For the second part of the project, the existing wake models are reviewed extensively and implemented in several test cases on Sexbierum onshore and Lillgrund offshore wind farms, see the recent paper "Wind turbine wake models developed at the technical university of Denmark: A review" [1]. The available robust wake models are seen to provide comparable results to the advanced, highly detailed models. However, they are tuned for 10min averaged data to acquire long term, statistical information. Therefore, in order to model the wind speed through the wind farm for higher time resolutions, the Larsen wake model is recalibrated using 1Hz SCADA. To adjust the parameters in the Larsen wake model, a methodology to estimate the Turbulence Intensity, TI, across the wind farm based on the rotor effective wind speed is presented. The methodology is implemented in Lillgrund and Horns RevI offshore wind farms and compared with the met mast as well as the nacelle anemometer and standard deviation of the produced power. Secondwise estimated rotor effective wind speed and the tabulated TI at the upstream and downstream turbine locations are used for both the recalibration in Thanet and the validation in Horns RevI for single wake cases.


Figure to the left: The recalibrated realtime wake model is implemented in the wind farm scale, considering the multiple wakeinteraction as in Larsen et al. [2],the pragmatic meandering as inAinslie[3] together with the localised time delay.
In regard to the most conservativeDanish TSO, Energinet.dkrequirements on provision of the data for the compensation during mandatory downregulation, the percentage error was analysed in 5minutes intervals when evaluating the model performance in wind farm scale power estimation. The applied realtime power curve estimation approach is shown to be in a good agreement with the active power signal, also considering the 5% error band at the power plant level as stated in the same regulation. Then the model is further implemented in the downregulated state.

To test the performance of the PossPOW algorithm under curtailment, a series of dedicated experiments are conducted in Horns RevI in early spring 2015. In view of the fact the available power of a turbine in the wake of a curtailed turbine is not measurable, the idea is to take advantage of a rather simple layout such as Horns RevI, using the similarity of the flow along the neighbouring rows. Accordingly, two of the upstream turbines are curtailed for the westerly winds under specific inflow conditions where start and stop triggers are applied considering the incoming wind direction and inflow speed. Six downregulation events are executed in total, where the inflow conditions satisfy the prerequisite criteria. The results of the experiments show that the PossPOW algorithm is capable of reproducing the wind speed at the reference turbine, correcting the reduced wake deficit due to downregulation hence providing the actual possible power with a time resolution of a second. It is also noted that, the effect of downregulation continues towards approximately five downstream turbines along the row for upright incoming winds. The available power signals obtained from the PossPOW algorithm are compared to the current state of the art industry practice where the individual SCADA signals of the possible power are aggregated. The comparison is presented according to the Danish regulations regarding the quality of the wind farm scale available power signal, where the accuracy is limited to 5% of the actual produced power over 15minutes intervals. It is seen that the current practice consistently overestimates the production capacity where the PossPOW algorithm mostly stays within the error band. The mean error of the current practice is observed to be 35% among the six experiments performed, where the PossPOW algorithm is shown to be capable of reducing that error down to 10%. Furthermore, the error range of the current practice and the PossPOW algorithm indicate much narrower distribution for the developed methodology corresponding to lower level of uncertainties.
References
[1] Göçmen, Tuhfe, et al. "Wind turbine wake models developed at the technical university of Denmark: A review." Renewable and Sustainable Energy Reviews 60 (2016): 752769.
[2] Larsen, Torben J., et al. "Wake effects above rated wind speed. An overlooked contributor to high loads in wind farms." EWEA Annual Conference and Exhibition 2015. 2015.
[3] Ainslie, J. F. "Wake modelling and the prediction of turbulence properties."Proceedings of the Eighth British Wind energy Association Conference, Cambridge, Mar. 1986.