Presentation is loading. Please wait.

Presentation is loading. Please wait.

Forecasting Uncertainty Related to Ramps of Wind Power Production

Similar presentations


Presentation on theme: "Forecasting Uncertainty Related to Ramps of Wind Power Production"— Presentation transcript:

1 Forecasting Uncertainty Related to Ramps of Wind Power Production
European Wind Energy Conference, Warsaw, April 2010 Arthur Bossavy, Robin Girard, Georges Kariniotakis Center for Energy and Processes MINES-ParisTech/ARMINES

2 Introduction Need to improve wind power forecasting with focus on extreme situations Various temporal/space scales Focus on uncertainty and weather predictability Distribution tail events To contribute to An increased and more secure wind integration to power grid Lower costs (i.e: reduced imbalances)

3 Centered prediction intervals of coverage:
Introduction A problem with usual wind power forecasts Centered prediction intervals of coverage: predictions observations

4 Objectives of the work 1. Improve the reliability
of usual confidence intervals w.r.t ramp events 2. Forecast confidence intervals to estimate the uncertainty of ramps timing

5 Outline A methodology for ramps detection
A probabilistic model using ramps information Forecast of ramps timing using ensembles

6 Detection of ramps FILTERING THRESHOLDING Threshold Ramp detected
intensity timing Threshold

7 Evolution of the ramp intensity through Denmark
Detection of ramps Evolution of the ramp intensity through Denmark

8 Outline A methodology for ramps detection
A probabilistic model using ramps information Forecast of ramps timing using ensembles

9 A probabilistic model using ramps information
Objective: Produce more reliable probabilistic forecasts by using information on forthcoming ramps

10 3-stage forecasting process using ramps information
Production of spot forecasts SCADA NWP Spot forecasting model Ramps detection forecasts spot Thresholding Filtering Ramps TIMING INTENSITY Probabilistic processing Probabilistic forecasting model NWP SCADA TIMING INTENSITY Ramps Information

11 Case-studies 1 wind farm in Ireland, 1 in Denmark
18 months of data (02/01-08/02 and 01/03-07/04) Hourly power measures Hourly wind speed/direction NWP forecasts (10m height). Probabilistic model based on the Quantile Regression Forests procedure

12 Evaluation measures Reliability: Sharpness:

13 Results Wind farm in Denmark Wind farm in Ireland
Forecasts underestimate quantiles Reliability improved for highest quantile forecasts Sharpness remains unchanged

14 Results Estimation of the uncertainty may be improved at ramps
Need of more tests: other quantile estimation methods

15 Outline A methodology for ramps detection
A probabilistic model using ramps information Forecast of ramps timing using ensembles

16 Forecast of ramps timing using ensembles
Objective: Aggregate ramps information provided by members of a wind power forecasts ensemble

17 Forecast of ramps timing using ensembles
Filtering members of a forecasts ensemble More than 35 over 51 members predicting this ramp h1 h3 h2

18 Forecast ramps timing using ensembles
Proposal for a probabilistic forecast of ramps timing Mean value for the ramp timing: Confidence intervals:

19 Case-studies and evaluation results
Case-studies: 3 wind farms in France Wind speed forecasts ensemble (51 members from the EPS system of ECMWF) Random Forest procedure Evaluation of forecast probabilities: Brier Score: Brier Skill Score w.r.t Climatology:

20 Visualization of confidence intervals
70% 57% 65% 39% 39 members predicting the increasing ramp 15 members predicting the decreasing ramp 43% 28%

21 Conclusions The probabilistic model using ramps information may be valuable when estimating the highest quantiles The approach based on ensembles provide confidence intervals to forecast ramp occurrence. Reliability w.r.t Climatology is improved Need of more experiments

22 Acknowledgments Project SAFEWIND: « Multi-scale data assimilation, advanced wind modelling and forecasting with emphasis to extreme weather situations for a safe large-scale wind power integration » Industrial partners of the project for providing data

23 Thank you for your attention!
29


Download ppt "Forecasting Uncertainty Related to Ramps of Wind Power Production"

Similar presentations


Ads by Google