Improvements in Wind Power Forecasts through use of the WRF Wind Farm Parameterization evaluated with Meteorological and Turbine Power Data

Forecasting wind power production remains a great challenge for utilities and power operators, especially in large wind farms in which wake effects reduce the power production of waked turbines, or in geographic areas with numerous wind farms that may interact. The mesoscale numerical prediction of the Weather Research and Forecasting (WRF) model incorporates a Wind Farm Parameterization (WFP) that uses elevated drag and turbulent kinetic energy production to represent the effect of wind turbines on the atmosphere, but this parameterization has not yet been validated in an onshore wind farm. To test model accuracy in simulating wind power production, as well as possible improvements to the WRF WFP, we use observations of winds and wind turbine power production in a 300-MW wind farm. We test a number of simulation approaches:

• WRF with no wind farm parameterization,
• WRF with the current form of the WFP,
• WRF with a WFP modified to account for a weighted-average of wind speed across the turbine rotor (rotor-equivalent wind speed, REWS), and
• WRF with a WFP modified to reduce the amount of turbulence generated by the turbines, as suggested by previous large-eddy simulations.

For the model evaluation, we focus on a location with flat terrain and a time period with simple synoptic conditions but strong enough winds for the wind turbines to be active. We use the August 2013 time period of the Crop and Wind Energy eXperiement in 2013 (CWEX-13) in central Iowa; numerous summertime nocturnal low level jets (LLJs) occurred during this time period. The simulations are nested down from 9- to 1-km horizontal resolution with about 20 m resolution in the turbine rotor layer. During this period, WRF simulations of wind profiles throughout the boundary layer verify well against Windcube 200S scanning LiDAR observations of wind speed and direction profiles, as well as with Windcube v1 wind measurements in the lowest 200 m.

Initial results compare simulations of WRF with no WFP to WRF with the current WFP, evaluated by the total farm production. As shown in Figure 1, which shows 10-min resolution predictions of total farm production compared to observed production, the use of the WFP greatly reduces the large bias seen in the WRF simulations without any WFP, from +43 MW to -4 MW. The large scatter in both cases demonstrates that improvement is required. Overall, wind power forecasting in large wind farms or in areas with numerous wind farms should incorporate some form of a wind farm parameterization.

Figure 1: Scatterplots of power production, between observed power and WRF-calculated power with no WFP (left), and between observed power and simulated power using the current WFP (right).