Validation of Energy Predictions by

Comparison to Actual Performance

Clint Johnson, Andrew Tindal, Keir Harman, AnneMarie Graves

Garrad Hassan America

WINDPOWER 2008
Houston, TX
Overview

Comparison of actual production vs. GH predictions

Observations and explanations for deviations

Amendments to GH methods

Next steps for GH and the industry

Validation of GH Energy Predictions

Comparison of actual vs.
predicted production

Validation of GH Energy Predictions

Energy production validation database
Contains 41 wind farms in North America
Operational periods from 1 to 8 years
Contains 113 wind farm years of energy production
Raw metered substation production
Unit is wind farm year
Includes some public domain data from the Energy
Information Administration (EIA)
Availability data from 31 wind farms

Validation of GH Energy Predictions

Distribution of Annual Energy Production
113 North American wind farm years relative to GH Projected P50
30
Actual production

25
GH Predicted distribution
No of wind farm years

20
Wind farm years: 113

Average: 90%
15

10

0
50%

55%

60%

65%

70%

75%

80%

85%

95%

100%

105%

110%

115%

120%

125%

130%

135%

145%

150%
90%

140%
Actual annual production / GH Predicted P50

Validation of GH Energy Predictions

 Observations and Explanations

Validation of GH Energy Predictions

 Possible Sources of Under-performance
Analysis methodology
Wind Resource Prediction Error
Measurement bias Quality measurements are
Long-term adjustment still crucial!
Extrapolation to hub height
Wind flow modelling

Energy loss factor prediction error

Wake loss modelling
Availability
Turbine performance
Curtailment Largest contributing factors
Electrical

Windiness of operating period

Validation of GH Energy Predictions

Distribution of Annual Energy Production
70 Availability corrected wind farm years relative to GH Projected P50
25
Actual production

20 GH Predicted distribution
No of wind farm years

Wind farm years: 70

15
Average: 93%

10

0
50%

55%

60%

65%

70%

75%

80%

85%

95%

100%

105%

110%

115%

120%

125%

130%

135%

145%

150%
90%

140%
Actual annual production / GH Predicted P50

Validation of GH Energy Predictions

Focus on 2007

Adjust database for actual availability achieved by

wind farms in 2007

Adjust database for windiness of 2007

Use publicly available wind speed data from long-term
meteorological stations
Identify regions of similar wind patterns

Validation of GH Energy Predictions

2007 Indicative windiness across the US

0%

+2% +1%

0% +1%
-2%
+2%
-3%

-7%

Note wind speed NOT energy

Validation of GH Energy Predictions

 Focus on 2007
Comparison of actual production against GH Projected P50
after adjusting each wind farm production to average wind speed
and for availability

Windiness and
Windiness
All data availability
adjusted
(41 wind farms) adjusted
(41 wind farms)
(27 wind farms)
Average ratio
Actual/predicted 90% 92% 96%

Conclusion: Average ratio within 5 % of ideal result

Validation of GH Energy Predictions

 Critical Appraisal of Methods

GH have gone through a process to attempt to identify what elements of our

analysis may lead to bias

This has involved the very detailed analysis of the 10-minute SCADA data
from many North American wind farms

We have identified areas where we believe there is potential for bias and have
amended our processes in the light of these findings

Validation of GH Energy Predictions

 Further Investigation of Energy Loss Factors

Availability
What levels of availability are being achieved?
What availability levels do we expect to see in the future?

Turbine Performance
Are we interpreting manufacturer power curves correctly?
How does power performance vary in different wind regimes?
Are individuals turbines within a wind farm operating as they should?

Wake loss modelling

Are existing models sufficient for very large wind farm developments?

Validation of GH Energy Predictions

 Availability Data North America vs. Europe
Annual moving average: Europe Annual moving average: N America
Count of wind farms: Europe (2nd y-axis) Count of wind farms: N America (2nd y-axis)
100% 220

99% 200

180
98%
Standard GH Availability Ramp-up Assumption for North America
160
Wind Farm System Availabilty [%]

97%

140

Number of wind farms

96%
120
95%
Very little data! 100
94%
80

93%
60

92%
40

91% 20

90% 0
0 1 2 3 4 5 6

Years of wind farm operation

Validation of GH Energy Predictions

Are we interpreting manufacturers power
curves correctly?
IEC 6-1400 Pt 12 says:

Care shall be taken in locating the meteorological mast. It shall

not be too close to the wind turbine since the wind speed will be
influenced/changed/affected in front of the wind turbine

Is the presence of the turbine reducing the wind speed

measured during a power curve test?
Is there an industry-wide, systematic bias in energy production
assessments?

Validation of GH Energy Predictions

 Blockage of wind speeds in IEC test
IEC 6-1400-12-1 guidance
4D 2.5D 2D Install met mast between
2 and 4 D from turbine
2.5 D recommended
Distance compromise
between blockage effect
Delta U and correlation
D

(1-1.5%) U? At 2.5 D small blockage effect

still present
Not questioning validity of measurements
Questioning the interpretation

Validation of GH Energy Predictions

 Site Specific Power Curve Adjustments

Analysis of 10-minute SCADA data and general industry

knowledge indicate:

Reduced power performance in high turbulence

Reduced power performance at sites with steep slopes

GH is making site specific adjustments to power curves

Validation of GH Energy Predictions

Site Specific Power Curve Adjustments
Variation of performance with turbulence intensity

Knee of power
curve degrades
with increase in TI

Difference in AEP = 1%

Validation of GH Energy Predictions

 Are turbines performing as they should?
Power curve comparison of 2 turbines in the same wind farm

Ramp-up in turbine
performance will be assumed
Power

Normal Turbine

Underperforming Turbine

It takes time and effort to make

turbines perform as they should

Wind speed

Validation of GH Energy Predictions

How good are our wake models in large wind farms
with low ambient turbulence?

GH Eddy Viscosity Model has been

validated against actual production

Recent validation of large offshore

projects shows some under-prediction

Similar effect may be happening in large onshore

projects

Apply adjustment informed by offshore experience

Validation of GH Energy Predictions

 Summary and Conclusions

Raw results show over-prediction

2007 results within 5 % of ideal result when availability and

windiness are considered

Availability data show long ramp-up

Key question for medium term availability levels will O&M
money be spent to achieve and maintain high availability?
Why would the industry not do this?

Validation of GH Energy Predictions

 Summary and Conclusions
Critical appraisal of methods based on science and data.
From this five potential causes of bias identified and
adjustment made:

1. Availability
2. Power curve blockage effect adjustment
3. Steep slope / high turbulence adjustment
4. Poor power performance in initial years of operation
5. Large wind farm wake model adjustment

Net reduction in AEP of 2 % to 5 % depending on site

Validation of GH Energy Predictions

 Summary and Conclusions

From the above discussion and GH revised methods, under-

performance can be explained

Industry needs to continue to critically review actual

performance data from wind farms

Industry needs to review analytical methods based on science

and data GH has extensive R&D program.

Wind measurement campaigns and good data have a VITAL

role to play in good predictions

Validation of GH Energy Predictions

Thank you!
Validation of GH Energy Predictions