4/19/20111Short‐and Long‐Term Wind Farm Power PredictionAndrew KusiakIntelligent Systems LaboratoryIntelligent Systems Laboratory2139 Seamans CenterThe University of Iowa Iowa City, Iowa 52242 – 1527andrew‐[email protected] : 319‐335‐5934 Fax: 319‐335‐566http://www.icaen.uiowa.edu/~ankusiakOutline Forecasting from weather based and SCADA data Basic methods for wind farm power prediction using data from weather forecasting models Short‐term and long‐term prediction Conclusions4/19/2011 Intelligent Systems Laboratory 2NAM weather forecasting model• North American Mesoscale (NAM) model• A day‐ahead forecasting with maximum forecast length of 84 hours•The spacing between model’s grid points is40 kmThe spacing between model s grid points is 40 km• A new 84‐hour forecast is issued 4 times daily at 00, 06, 12, and, 18 GMT• The forecasted value is saved every 3 hours• Data source for long‐term wind farm power prediction4/19/2011 Intelligent Systems Laboratory 3RUC weather forecasting model• Rapid Update Cycle (RUC ) model• Short‐term forecasting model with maximum 12‐hour forecast length• The spacing between model’s grid points is 20 km• Hourly forecasts are issued at 00, 03, 06, 09, 12, 15, 18, and 21 GMT • Besides the 12‐hour forecast, a9‐hour forecast is issued every other hour• Data source for short‐term wind farm power prediction4/19/2011 Intelligent Systems Laboratory 44/19/20112General characterization of weather forecasting data• 16 model data points surrounding the wind farm site are extracted for wind power forecasting 4/19/2011 Intelligent Systems Laboratory 5The wind farm is located between the model data points: 6, 7, 10, and 11NAM data descriptionParameter Description UnitSpd_10m Wind speed 10 m above the surface m/sDir_10m Wind direction 10 m above the surface degSpd_XXmbAverage wind speed in the lowest XX mb of the atmosphere (XX is 30, 60 and 90 respectively)m/sDir XXmbAverage wind direction in the lower XX mb of the atmosphere (XX is 30, 60 and 90 ti l )deg4/19/2011 Intelligent Systems Laboratory 6_respectively)gAD_30mb Average air density in the lowest 30 mb of the atmosphere kg/m3PTdiff_30mb_sfcPotential temperature difference between the surface and 30 mb above the surface; Measure of atmospheric stability in lower spaceKSHTFL Sensible heat flux at the surface; Indicator of surface heating or coolingW/m2VEG Percentage of the surface that is covered by vegetation %Each of the 16 points has 12 parameters; 12 * 16 = 192!Millibar‐height conversion1000 mb ~ 360 feet (110 m)850 mb ~ 5000 feet (1500 m)700b~ 10 000 ft(3000 )4/19/2011 Intelligent Systems Laboratory 7700 mb~ 10,000 feet (3000 m)500 mb ~ 18,000 feet (5400 m)250 mb ~ 34,000 feet (10,200 m)RUC data descriptionParameter Description UnitSpd_10m Wind speed 10 m above the surface m/sDir_10m Wind direction 10 m above the surface degSpd_XXmbAverage wind speed in the lowest XX mb of the atmosphere (XX is 30, 60 and 90 respectively)m/sDir_XXmbAverage wind direction in the lower XX mb of the atmosphere (XX is 30, 60 and 90 respectively)degAD_30mb Average air density in the lowest 30 mb of the atmosphere kg/m3PTdiff_30mb_sfcPotential temperature difference between the surface and 30 mb above the surface; measure of atmospheric stability in lower spaceK4/19/2011 Intelligent Systems Laboratory 8Each of the 16 points has 10 parameters; 16 * 10 = 160!4/19/20113General characteristics of SCADA data• Supervisory Control and Data Acquisition (SCADA) systemcollects wind turbine data• The wind farm in this research contains 76 wind turbines• SCADA collects data for more than 120 parameters at each turbine (including wind speed, wind direction, power and so on) and many wind farm level data• Data is stored at 10‐minute intervals (10‐minute average data)• Source for data‐driven performance analysis of wind farm4/19/2011 Intelligent Systems Laboratory 9Data mining• Data mining is a tool for extracting knowledge and solving problems4/19/2011 Intelligent Systems Laboratory 10Parameter selection and transformation• The four closest model points 6, 7, 10, and 11 are selected as predictors for the wind farm power• To obtain an accurate prediction model with a data mining approach, the original high‐dimension data need b fdi ldi i dto be transformed into low‐dimension data vectors • The original 192‐demension NAM data has been reduced to a 8‐demension predictor for long‐term wind farm power prediction; The original 160‐demension NAM data has been reduced to a 6‐demension predictor for short‐term wind farm power prediction4/19/2011 Intelligent Systems Laboratory 11Measures of prediction accuracy• AE: Absolute error (%)• Definition: The absolute value of the difference between the predicted and actual power output, and it is expressed as percentage of the installed nameplate ratingˆ()()100%yt T yt TAENRP • MAE: Mean absolute error (%), average of the absolute error over particular data set• Std (%): The standard deviation of the AE• MAE and Std are widely used metrics in industry and research4/19/2011 Intelligent Systems Laboratory 121()NiAE iMAEN1(() )1NiAE i MAEStdN4/19/20114SCADA data predictionsHorizon MAE Standard Deviation10 Minute prediction2.213 2.50120 Minute prediction3.912 4.08330 Minute prediction5 1435 1494/19/2011 Intelligent Systems Laboratory 1330 Minute prediction5.1435.14940 Minute prediction6.062 5.91750 Minute prediction6.721 6.56760 Minute prediction7.384 6.98770 Minute prediction8.025 7.514SCADA data predictionsHorizon MAE Standard Deviation1h prediction5.850997 5.6545494/19/2011 Intelligent Systems Laboratory 14p2h prediction9.336708 8.9165293h prediction11.82863 11.234144h Prediction14.99185 13.20335Short‐term power predictions: RUC data• Output: hourly power, average power over an hour• Following the RUC forecasting horizon and steps, prediction can be done from 1 hour to 12 hours ahead (T + 1, T + 2,…, T + 12)4/19/2011 Intelligent Systems Laboratory 15T+3 power prediction4/19/2011 Intelligent Systems Laboratory 16MAE(%): 9.76 Std(%):8.694 4/19/20115T+6 Power prediction4/19/2011 Intelligent Systems Laboratory 17MAE(%): 10.94 Std(%):9.99 T+8 Power prediction4/19/2011 Intelligent Systems Laboratory 18MAE(%): 10.57 Std(%):9.91 T+10 Power prediction4/19/2011 Intelligent Systems Laboratory 19MAE(%): 11.06