Temporal forecast uncertainty of ramp events

1
Temporal forecast uncertainty of ramp events
Jeremy Parkes - Garrad Hassan Partners
Ltdjeremy.parkes_at_garradhassan.com
2
Contents

• Background and general forecasting method
• Ramp event definition
• Calculation of temporal uncertainty
• Presentation of uncertainty
• Conclusions

3
Forecasting Method
NWP Forecast
Historic SCADA
Live SCADA
NWP Forecast
Site geography
NWP Forecast
Suite of Models
Adaptive statistics
Climatology
Time Series
Model adaptation
Intelligent Model Combination
Site geography
Wind speed forecast
Live SCADA

• Optimised combination of NWP (Numerical Weather
  Predictions)
• Incorporation of mesoscale models
• Regular live feedback from the wind farm
• Learning Algorithms for
• Meteorology
• Power models

Power model
Model adaptation
Power forecast
4
Current Forecast Uncertainty
Hourly data 24 hours in advance
Existing methods dont sufficiently capture the
temporal uncertainty of ramp events
5
Ramp definition

• What is a ramp event?
• Statistical analysis of wind farm production and
  discussion with clients helped formulate the
  definition used in this study

Ramps
Ramp event definition a change in power 50
capacity over a duration of 4 hours
6
Ramp definition

• Ramp event definition a change in power 50
  capacity over a duration of 4 hours

7
Calculation of temporal uncertainty

• Probability distributions of ramp occurrence
  based on ramp event forecasts and relative timing
  of measured ramps.

st 4 hours
Ramp before forecast
Ramp after forecast
8
Calculation of temporal uncertainty

• Probability distributions are adjusted based on
• An ensemble of NWP (Numerical Weather
  Predictions)
• Forecast horizon
• Wind farm specifics

9
Presentation of temporal uncertainty
Probable timing of ramp
Predicted centre of ramp
10
Results - individual wind farm

• True forecast of a ramp event
• Actual ramp within 12 hours of forecast ramp
• One actual ramp per forecast ramp, closest taken

Forecast horizon (hrs) 3 24
Forecast accuracy () 67.2 59.1
11
Results - individual wind farm

• Forecast horizon T24 hours

Good capture of rapid change in production
12
Results - portfolio

• 6 times fewer ramps for a portfolio

13
Results - portfolio
Portfolio forecasts at T24 hour horizon
Good capture of ramp events
14
Further work

• Further analysis of portfolio ramp forecasting
• Improve the combination of ensemble NWP
• Further discussion with end users to guide what
  is most useful both in terms of the central
  forecast and the presentation of the uncertainty