Forecasting Wind Energy Variability using Statistical Downscaling Techniques

Forecasting Wind Energy Variability using Statistical Downscaling Techniques

Peter CoppinEuropean Wind Energy Conference March 2009

CSIRO. Forecasting Wind Energy Variability

Acknowledgements

Division of Marine and Atmospheric Research• Robert Davy (principal researcher)• Milton Woods• Chris Russell• Peter Coppin

Funded by the Australian Government – Department of Resources Energy and Tourism (DRET)

• Australia Wind Energy Forecasting System (supplied by ANEMOS)

In collaboration with• Australian Bureau of Meteorology• European Union Framework 7 “SAFEWIND” Project partners

CSIRO. Forecasting Wind Energy Variability

SE states 35,000MW max demand

Core wind generating area

CSIRO. Forecasting Wind Energy Variability

Typical winter / early spring conditions

CSIRO. Forecasting Wind Energy Variability

IR Satellite image – 1225 hr on Sept 11 2004 showing organised convection

CSIRO. Forecasting Wind Energy Variability

Example of problem – high variability in South Australia

High wind variability at moderate wind speeds

• large swings in aggregate wind power

• wind farm aggregation can amplify the absolute magnitude of power changes

• variation can exceed available spinning reserve response capability

0.0

0.2

0.4

0.6

0.8

1.0

Nor

mal

ised

pow

er

11-Sep-200400:30

11-Sep-200406:30

11-Sep-200412:30

11-Sep-200418:30

12-Sep-200400:30

0.0

0.2

0.4

0.6

0.8

1.0

Nor

mal

ised

pow

er11-Sep-2004

00:3011-Sep-2004

06:3011-Sep-2004

12:3011-Sep-2004

18:3012-Sep-2004

00:30

west1100MW

west2100MW

central1350MW

central2100MW

east1200MW

east2400MW

Cha

nge

in M

W

-150

-100

-50

0

State total generation

Generation in 6 regions

Regional Generation at 12:30

CSIRO. Forecasting Wind Energy Variability

Steps to forecasting wind variability

• Create a numerical index to describe severity

• Calculate index for observations

• Formally correlate observed variability with weather patterns derived from Numerical Weather Prediction (NWP) products via EOF analysis

• Produce prediction of variability based on combination of important patterns – using Random forest techniques – express in terms of wind speed and wind energy production ramp rate at suitable forecast time horizon

• Check predictability with observations (separate data segment)

CSIRO. Forecasting Wind Energy Variability

Modelling wind variability

Initial proof-of principle based on analysis fields (i.e. not forecast mode)

• NCEP Global Tropospheric Analyses ( ds083.2 )• six hourly meteorological fields at 1.0° resolution [1999 - ]

• Surface wind speed measurements: • 6 locations across South eastern Australia• Used for calculating variability index and validation

Forecast mode trial

• Bureau of Meteorology W-LAPS model• Six hourly fields at 1° and 0.1° grids (data set length?)• 12 hours ahead

CSIRO. Forecasting Wind Energy Variability

Index of wind variability

• from Woods, Davy & Coppin, 2007• Variability index is six hour running standard deviation of 2 hour

high-pass filtered (10min) raw data-5

05

10

15

20

m/s

11/09/2004 13/09/2004 15/09/2004

Original dataFluctuationsVariability index

CSIRO. Forecasting Wind Energy Variability

Available (and relevant) variables for EOF analysis

• NCEP Global Tropospheric Analyses ( ds083.2 )• six hourly meteorological fields at 1.0° resolution [1999 - ]

Geopotential height Air temperature Relative humidityU wind speedV wind speedVertical velocityAbsolute vorticityHeight of planetary boundary layerPrecipitable waterCloud waterLifting indexConvective inhibitionConvective available potential energyetc

CSIRO. Developing a prediction of wind variability

EOF example : Temperature-2

-10

12

Time

01 Jun2003

05 Jun2003

09 Jun2003

13 Jun2003

-2-1

01

2

Time

01 Jun2003

05 Jun2003

09 Jun2003

13 Jun2003

-2-1

01

2Time

01 Jun2003

05 Jun2003

09 Jun2003

13 Jun2003

CSIRO. Forecasting Wind Energy Variability

Quantitative modelling

Meteorological inputs(transformed using EOF)

Empirical model(random forest)

Sum of variability indexat 6 sites

Model period: 2003-4

CSIRO. Forecasting Wind Energy Variability

Winter – important patterns explaining variability

Height of planetary boundary layer – EOF1 Absolute vorticity – EOF3

Velocity component V – EOF2

CSIRO. Forecasting Wind Energy Variability

NCEP Results – winter / early spring

• Prediction of variability index (using analysis NWP products)

Important predictors:EOF number

Height of planetary boundary layer 1Absolute vorticity 3V wind speed 2U wind speed 1

R2=0.74

Aggregate 6 locations in SE Australia

CSIRO. Forecasting Wind Energy Variability

NCEP Results – winter / early spring

• Prediction of variability index (using analysis NWP products)

Important predictors:EOF number

Height of planetary boundary layer 1Absolute vorticity 3V wind speed 2U wind speed 1

R2=0.74

Single siteAggregate 6 locations in SE Australia

CSIRO. Forecasting Wind Energy Variability

Summer – important patterns explaining variability

Geopotential height – EOF2

Convective available potential energy – EOF1Cloud water – EOF1

CSIRO. Forecasting Wind Energy Variability

NCEP Results - summer

• Prediction of variability index (using analysis NWP products)

Important predictors:EOF number

Cloud water 1Geopotential height 2CAPE 1V wind speed 2

R2=0.66

Aggregate 6 locations in SE Australia

CSIRO. Forecasting Wind Energy Variability

NCEP Results - summer

• Prediction of variability index (using analysis NWP products)

Important predictors:EOF number

Cloud water 1Geopotential height 2CAPE 1V wind speed 2

R2=0.66

Single siteAggregate 6 locations in SE Australia

CSIRO. Developing a prediction of wind variability

Modelling of 6-hour mean ramp rate • aggregate power from 6 locations vs fitted variability index

Mean wind power ramp rate modelling

CSIRO. Forecasting Wind Energy Variability

Short-term wind power ramp rate modelling

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14

0.0

0.1

0.2

0.3

0.4

0.5

0.6

var index

max

imum

10m

in r

amp

rate

0.00 0.02 0.04 0.06 0.08 0.10 0.12 0.14

0.0

0.1

0.2

0.3

0.4

var index

ram

p ra

te

90%50%10%

Modelling of 10 min ramp rate

•aggregate power from 6 locations (normalised)

•relationship of 6 hourly maximum ramp rate (left) and instantaneous ramp rate (right) to variability index

CSIRO. Forecasting Wind Energy Variability

W-LAPS forecast mode trial

• EOF analysis uses both 1° (100km) and 0.1° (10km) fields

• Correlations performed on 12-hour ahead forecast products

• Single site results available

• Example EOF model fit:

Variable Grid domain EOF number

Vertical velocity (omega) Inner 1

Mean sea level pressure Inner 3

24 hour precipitation Inner 1

Latent heat flux Outer 2

Atmospheric boundary layer height

Outer 2

CSIRO. Forecasting Wind Energy Variability

W-LAPS Results – 12 hours ahead

• Forecast of variability index (using analysis NWP products)• Single site

Winter Summer

CSIRO. Forecasting Wind Energy Variability

W-LAPS Results – 12 hours ahead

• Forecast of variability index (using analysis NWP products)• Aggregate of several sites in one state

Winter Summer

CSIRO. Forecasting Wind Energy Variability

Further work

• Implement working system• ready for coding into ANEMOS system as a module

• Further validate forecast-mode models for aggregate variability• requires additional surface data

• Quantify value to system operators of variability information• cost benefit analysis

Contact UsPhone: 1300 363 400 or +61 3 9545 2176

Email: enquiries@csiro.au Web: www.csiro.au

Thank you

Centre for Australian Weather and Climate ResearchA partnership between CSIRO and the Bureau of MeteorologyRobert Davy

Phone: 02 6246 5604Email: robert.davy@csiro.auWeb: www.csiro.au/weru