Statistical Characteristics of Winds at Xcel Wind...
Transcript of Statistical Characteristics of Winds at Xcel Wind...
Statistical Characteristics of Winds
at Xcel Wind Farms
Yuewei Liu, Yubao Liu, Will Y.Y. Cheng and Gregory Roux
Wind Energy Prediction - R & D Workshop
11-12 May 2010
© 2009, University Corporation for Atmospheric Research. All rights reserved.
1. Introduction
2. Classification of intra-farm wind variation patterns
3. Node Level Power Curve (NLPC)
4. Wind/power persistence
5. Summary
Data:Met-tower measurementsTurbine nacelle wind speeds© 2009, University Corporation for Atmospheric Research. All rights reserved.
XXXX Farm2 Met-towers (A & H)
and274 Wind Turbines
A
H
4 – 6 m/s6 – 8 m/s
2 – 4 m/s8 – 10 m/s0 – 2 m/s
12 – 14 m/s> 14 m/s
10 – 12 m/s
• A major farm• Good historical and
real-time data• Modeling testbed• Complex terrain
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Average Nacelle Wind Speed at XXXX Wind Farm
8.6 – 8.9 m/s8.9 – 9.2 m/s
8.3 – 8.6 m/s9.2 – 9.5 m/s9.5 – 9.8 m/s9.8 – 10.1 m/s
> 10.1 m/s
8.0 – 8.3 m/s
5.4 – 5.6 m/s5.6 – 5.8 m/s
5.2 – 5.4 m/s5.8 – 6.0 m/s6.0 – 6.2 m/s6.2 – 6.4 m/s
> 6.4 m/s
5.0 – 5.2 m/s
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Wind Patterns and Frequencies of Intra-farm Flows at XXXX Farm
Objectives: Understand ingredients of farm power production Obtain climatologies of farm-wide microscale flows
(turbines sit 400-700m apart) Use the knowledge for model output post-processing Provide guidance for farm data assimilation
(Power FCSTWeather Regime Weather Pattern)
Methodology and data: SOMs (Self-Organizing Maps): A neural network
scheme for pattern recognition and classification (automated data feature extraction). 10-min average nacelle and met-tower measurements
© 2009, University Corporation for Atmospheric Research. All rights reserved.
4 Patterns Based on Raw SpeedsWinter Season: 20081023-20090209
4 – 6 m/s6 – 8 m/s2 – 4 m/s
8 – 10 m/s10 – 12 m/s
12 – 14 m/s> 14 m/s
0 – 2 m/s
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Spatial Patterns (SOMs): 20081023-20090210
4 – 6 m/s6 – 8 m/s2 – 4 m/s
8 – 10 m/s10 – 12 m/s
12 – 14 m/s> 14 m/s
0 – 2 m/s
8 Patterns based on Raw Speeds
1 32
5
4
76 8
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Objective: Relative speed distribution
N _ speedi =speedi
speed
speed =1n
speedii=1,274∑
© 2009, University Corporation for Atmospheric Research. All rights reserved.
4 – 6 m/s6 – 8 m/s2 – 4 m/s
8 – 10 m/s10 – 12 m/s
12 – 14 m/s> 14 m/s
0 – 2 m/s
4 Patterns Based on Normalized SpeedWinter 20081023-20090209
© 2009, University Corporation for Atmospheric Research. All rights reserved.
4 – 6 m/s6 – 8 m/s2 – 4 m/s
8 – 10 m/s10 – 12 m/s
12 – 14 m/s> 14 m/s
0 – 2 m/s
8 Patterns Based on Normalized SpeedWinter 20081023-20090209
1 32
5
4
76 8
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Self Organizing Maps (SOMs) Analysis: Findings
1. SOMs is a very effective tool for intra-farm flow statistical property analysis.
2. Raw wind speed: Wind events dominate the patterns (useful for ramp event analysis)
3. Normalized speed: Spatial distribution dominates the patterns (useful for sub-grouping nacelle data, and can be used provide guidance for data assimilation)
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Motivations: Wind power prediction Node level power vs. individual turbine generation
Methodology & data: Node level power curve: Referred to nacelle average wind speed Referred to nacelle medium wind speed Piecewise Linear Regression (Node level power
prediction) XXXX Farm: 20081023-20081123 Apply the regression equations to winter 2009
(20090902-20091014)
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Node Level Power
09-02_02:00-10-14_04:45 (2009)
0
50
100
150
200
250
300
350
0 5 10 15 20 25
Mean-Wind (m/s)
Tota
l Pow
er(M
W)
09-02_02:00--10-14_04:45 (2009)
0
50
100
150
200
250
300
350
0 5 10 15 20 25
Medium-Wind (m/s)
Tota
l Pow
er (M
W)
Referred to Mean Referred to Medium
© 2009, University Corporation for Atmospheric Research. All rights reserved.
10-23_07:00-11-23_06:50 (2008)
0
50
100
150
200
250
300
350
0 5 10 15 20 25
Mean-Wind(m/s)
Tota
l Pow
er(M
W)
0
50
100
150
200
250
300
350
1 97 193 289 385 481 577 673 769 865 961 1057
1153
Modeled (red) and observed (blue) power during the validation period
p=25.x-87.5 (3.5<=x<6.5)p=30.x-120. (6.5<=x<8)p=38.33x-186.667 (<8=x<=11)p=31.5x-111.5 (11<x<13)p=305 (13<=x<=25)p=0 (x<3.5 or x>25)
Node level power (MW) curve referred to Nacelles averaged wind speed (m/s)09-02_02:00-10-14_04:45 (2009)
0
50
100
150
200
250
300
350
0 5 10 15 20 25
Mean-Wind (m/s)
Tota
l Pow
er(M
W)
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Findings: The statistical analysis showed a stable regression
between the farm-wide mean wind and the node level power output for XXXX farm.
The node level piecewise linear regression equations of winter 2008 fit winter 2009.
The robustness of NLPR suggests that reliable power forecasts can be achieved by accurately forecasting the farm-wide mean hub-height wind speed.
Future work: Investigate node level power curve referred to average
speed for other wind farms
© 2009, University Corporation for Atmospheric Research. All rights reserved.
• Persistence Properties:
Persistent forecast performs “surprisingly well” up to 4 hours (Giebel et al.) – A Rule of Thumb
Persistence property is of high interest and value.
How rapidly winds vary in a wind farm? Persistence is the reflection of weather/wind regime:• among different wind farms• different periods at the same farm This important information can be used to guide the usage of
wind farm data for NWP model data assimilation and/or for statistical predictions.
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Method:
Data: winter (Oct. 23, 2008 – Feb. 10, 2009)Persistence error: 10min, 0.5hr, 1hr, 2hr, 3hr
summer (July 14, 2009 – Aug. 31, 2009)Persistence errors:15min, 0.5hr, 1h, 2hr,3hr
y
c h a n g ep = pt − pt−1
c h a n g ep : p o w e rc h a npt : p o w e ra tt i m etpt−1 : p o w e ra tt i m et −1
© 2009, University Corporation for Atmospheric Research. All rights reserved.
10min
0.5hr
15min
0.5hr
Winter (10/23/2008 – 2/10/2009) Summer (7/14/2009 – 8/31/2009)
© 2009, University Corporation for Atmospheric Research. All rights reserved.
1hr
2hr
1hr
2hr
Winter (10/23/2008 – 2/10/2009) Summer (7/14/2009 – 8/31/2009)
© 2009, University Corporation for Atmospheric Research. All rights reserved.
UCAR Confidential and Proprietary. © 2008, University Corporation for Atmospheric Research. All rights reserved.
Winter 3hr Summer 3hr
Winter (10/23/2008 – 2/10/2009) Summer (7/14/2009 – 8/31/2009)
© 2009, University Corporation for Atmospheric Research. All rights reserved.
• Findings Overall: > 50% cases change by 50% within 1 – 2 hr
Summer: Unstable, changes much faster than in winter
Strong power events occur in less frequency in summer; the strong power typically last for very short (< 1h) life.
• Future Works Variation of persistence for major weather regimes,
diurnal variations and for major SOMs wind patterns.
Provide guidance for NWP wind farm data assimilation Provide guidance for enhanced (RTFDDA) nowcasting
through post-processing using the wind farm data.
© 2009, University Corporation for Atmospheric Research. All rights reserved.
Met-tower and wind turbine data from XXXX Farm are analyzed to study
• Microscale flow patterns and occurrence frequencies • Node Level Power Curve and its potential usage• Wind/Power persistence properties
© 2008, University Corporation for Atmospheric Research. All rights reserved.
On-going work includes:• How to make use of the information to guide data
assimilation and model output statistical post-processing• Conduct analysis on the dependency of these statistical
properties on seasonal and diurnal weather regimes.