Kris Shrestha James Belanger Judith Curry Jake Mittelman Phillippe Beaucage Jeff Freedman John Zack Medium Range Wind Power Forecasts for Texas
Increases of wind penetration into energy grids has created the need for forecast information beyond a few days Operational Wind Power Forecasting Extended range wind power forecasts for Texas Stabilize energy cost and supply Natural gas trading and sales Maintenance scheduling Maximizing grid integration
Datasets for Forecast Verifications Model: medium-range forecasts – ECMWF ENS (VarEPS) 3-hourly / day 1-6, 0.250° 6-hourly / day 7-10, 0.250° 6-hourly / day 11-15, 0.500° – ECMWF HRES (deterministic) 3 to 6-hourly / 10 days, 0.125° Observations – ERCOT wind power generation 15 mins, regional average – ECMWF operational analyses, 100 m winds 6-hourly, 0.125° – Wind Forecast Improvement Project (WFIP) SODAR 10 min, 7 stations in Texas
Spatial Resolutions & SODAR Locations WFIP SODAR locations
Power Conversion Wind speed is converted to output power by a standard curve and scaled to % of rated limit (1.5 MW).
Example Wind Speed Forecast Diurnal cycle: grid level
Diurnal cycle: regional average
SODAR vs ECMWF Diurnal Variability 6 hour interval
Evaluation of power forecasts over three spatial domains (out to 10 days) – Single grid cell (1/4°) – Individual regions – ERCOT average Power Forecast Verification
Single Cell (1/4°): Cleburne WFIP SODAR – 22 Jul to 13 Oct, 2012 analyses sodar
Power Forecast Verification: Regions in Texas Average of Multiple Cells – Weather Regions – 13 Mar to 13 Oct, 2012 – ECMWF ensemble mean vs. ECMWF analyses ERCOT avg
Power Forecast Verification: All of ERCOT ERCOT Regional Average – 13 Mar to 13 Oct, 2012 – ECMWF ens mean vs. ECMWF analyses, ERCOT power [ persistence Reduced corr. vs ERCOT data compared to analyses results from assumptions about power curve, and lack of power-weighted regional average Ensemble mean starts to outperform deterministic ~120 hours
Towards increasing prediction skill: statistical post processing deterministic ensemble mean climatology adjusted Challenge to statistical post-processing using reforecasts: Need gridded historical wind data set at Hub height ( m) with similar data quality as verification data set Mean BiasRoot Mean Square Error
Summary Regionally averaged wind power shows useful prediction skill at the medium range Climatological and real time hub height wind data is needed to optimize the statistical post processing Accurate simulation of regional power generation requires power weighted averaging and power curves