The Good, the Bad and the Ugly Extreme Wind
Wiebke Langreder1
Jørgen Højstrup1
Lasse Svenningsen2
1 Suzlon Energy A/S, Denmark2 EMD A/S, Denmark
Contents (Part 3)
• The task: Mission Impossible?
• What we have done so far
• What is new
• Our results and recommendations
• Outlook
Terminology
• Extreme Wind = Maximum 10-minute average wind speed with recurrence period 50 years
• In IEC language: Vref
Positive Thinking?
Denmark 1999
Spain 2009
Japan
Inappropriate
The task: Mission Impossible?
• Predict maximum 10-minute average wind speed in 50 years.
• Normal situation: 1-5 years of data
• Extreme winds are not related to mean wind speed.
The task: Mission Impossible?
Objective:
Choose method to – Minimize uncertainty– Minimize bias
Contents
• The task: Mission Impossible?
• What we have done so far
• What is new
• Our results and recommendations
• Outlook
Establish MethodLong-time series are split in shorter sub-sets,each method is applied to each sub-set.
LT
Sub-set 1 → Vref
Sub-set 2 → Vref
Sub-set 3 → Vref
Sub-set 4 → Vref
Sub-set 5 → Vref
”True” Reference Value
Assumption
The “true” Vref is determined:
• using full data set
• extracting Annual Maxima (Periodical Maxima)
• Gumbel distribution
MethodNormalisation with this ”true” value
N subsets → N results per method → Standard deviation→ Bias
PM: LT → ”True” Vref
Sub-set 1 → Vref
Sub-set 2 → Vref
Sub-set 3 → Vref
Sub-set 4 → Vref
Sub-set 5 → Vref
Previous Methods
• EWTS European Wind Turbine Standard Vref depending on k factor
– 360 degree– sector with highest mean v
• PM Periodical Maximum
• POT Peak-over-thresholdGumbel
Contents
• The task: Mission Impossible?
• What we have done so far
• What is new
• Our results and recommendations
• Outlook
New development• Parameter describing Gumbel distribution
are determined graphically
New development
Possible reasons for non-linearity:
• Wrong way to extract extreme events?
• Wrong way to plot/fit?
• No convergence towards Gumbel?
Better extraction/plotting
IMIS - Improved method of independent storms
(Cook/Harris)
Different two-stage process to extract
Different way to fit regression
Improved convergence• Samples extracted from Weibull parent not
necessarily exponential
• Slow convergence towards Gumbel (exponential)
• Pre-conditioning
• Substitution of V with Vc
High end of Vc → exponential
Gumbel → exponential
Tatata: faster convergence
Pre-conditioning
Two methods:
V2 (dynamic pressure)
Vk (Weibull shape parameter (Cook/Harris))
Additional New Development
• Effect of measurement period:
Length of sub-sets: 1, 2, 3 and 5 years
Contents
• The task: Mission Impossible?
• What we have done so far
• What is new
• Our results and recommendations
• Outlook
Statistical relevance
15 sites (Europe, US, Asia, Roaring 40th)– 158 1 year periods– 77 2 year periods– 49 3 year periods– 22 5 year periods
PM PM k PM 2 POT POT k POT 2 IMIS IMIS k IMIS 21 year absolute 97% 91% 90% 98% 92% 91%
std dev 18% 15% 15% 25% 20% 20%
2 year absolute 106% 99% 99% 96% 91% 91% 100% 94% 94%std dev 28% 21% 21% 12% 11% 11% 18% 15% 14%
3 year absolute 108% 101% 101% 95% 91% 90% 100% 95% 95%
std dev 21% 16% 16% 10% 9% 9% 13% 11% 11%
5 year absolute 105% 100% 100% 92% 88% 88% 96% 91% 91%std dev 13% 12% 12% 9% 8% 8% 11% 10% 10%
ResultsPre-conditioning Different Methods
Period
Result - EWTS
• EWTS 360degr lowest results
• EWTS max similar numbers as PM-POT-IMIS
• based on distribution → less sensitive to actual period
• very difficult to identify ”correct” sector
EWTS 360
EWTS max
1 year absolute 89% 94%std dev 16% 16%
2 year absolute 87% 95%std dev 14% 16%
3 year absolute 86% 92%std dev 14% 17%
5 year absolute 81% 87%std dev 5% 9%
Recommendation (1/3)
• Use POT 2 (= dynamic pressure)
• lowest standard deviation and lowest standard error of the mean for 1 year periods
Disadvantage:
• Result very sensitive to highest measured wind speed in measurement period
Recommendation (2/3)
• Use EWTS max (sector with the highest average wind speeds)
Advantage:
• Independent of period
Disadvantage:
• Difficult to identify sector
Recommendation (3/3)
Combine the two methods• Engineering approach, taking the average of
EWTS (max) and POT 2POT2
POT 2 EWTS max1 year absolute 90% 93%
std dev 15% 12%
2 year absolute 91% 94%std dev 11% 10%
3 year absolute 90% 91%std dev 9% 11%
Outlook
• Check sensitivity to outlier
• If Vref depends on highest measured wind speed:Better results for a X year data set by using POT for each year seperately and then average?
•Try correlation with NCEP/NCAR to find out about level of highest measured wind speed in a sample
Acknowledgement
Thanks to
• www.winddata.com
• www.undeerc.org/wind
• www.bom.gov.au/inside/cgbaps