a time series comparison of pre- construction energy yield ...a time series comparison of...
TRANSCRIPT
A Time Series Comparison of Pre-
Construction Energy Yield Model
and Operational Data
Alex Clerc, Lee Cameron & Hannah Staab
December 9, 2014
EWEA Technology Workshop, Malmö
Introduction and Contents
• We are seeking to answer two questions:
– How accurate is the RES energy yield methodology?
– What can be done to improve it?
• Presentation contents:
– Data set description
– Validation Results
– Validation Methodology Details
– Lessons Learned
0
2
4
6
8
10
12
14
16
-12% -8% -4% 0% 4% 8% 12%Num
ber
of
Win
d F
arm
Years
Energy Prediction Error
Site speed up map using Ventos CFD
The validation database consists of:
• 31 wind farms
• 638 turbines
• 1.2 GW installed capacity (0.4% of entire world capacity)
Validation Database
Wind farm count by region and turbine manufacturer (31 total)
Senvion
Validation Results
• Overall observed error by turbine for 31 wind farms. Average validation
result: -2% after correcting for unforeseen losses using SCADA analysis.
0
50
100
150
200
250
-35% -30% -25% -20% -15% -10% -5% 0% 5% 10% 15% 20% 25%
Num
ber
of
Turb
ines
Energy Prediction Error
Before SCADA Corrections After SCADA Corrections
Over prediction Under prediction
Validation Results
• Subset of 18 wind farms representative of modern development.
• Average result: 0%
• Latest modelling methodology used (not the original yield prediction)
• Standard deviation lower than predicted by uncertainty model (note IAV
not applicable because windiness corrections have been made)
0
2
4
6
8
10
12
14
16
-12% -8% -4% 0% 4% 8% 12%
Num
ber
of
Win
d F
arm
Years
Energy Prediction Error
Over prediction Under prediction
Problem 1: Original pre-construction models may be outdated
in terms of layout, exposure, curtailment schemes, modelling
methodology
Solution: Create an “as-built yield model” using up to date
information and methodology
Example: Wind farm extended in 2007, original yield does not
consider wake effect of extension
Validation Methodology
Alta I (2003)
Alta II (2007)
Historic and Operational Windiness
Problem 2: Understanding of long-term windiness is based on
specific reference data period.
Solution: Time series comparison of modelled and measured
production, accounting for real windiness at the site. This
requires Time Series MCP (TSMCP).
Validation Methodology
Start of
Operations
(e.g. MERRA, available hourly)
Hourly predictions by turbine
10-minute turbine data
Monthly site meter data
Validation Methodology
• Wind farm climate and yield modelled on an hourly basis
• Model is run for every turbine, not just the WF as a black box
• Model identical to pre-construction yield model, but by turbine by hour
0
5
10
15
20
25
30
0
2
4
6
8
10
12
14
Sep
-07
Dec
-07
Mar
-08
Jun
-08
Sep
-08
Dec
-08
Mar
-09
Jun
-09
Sep
-09
Dec
-09
Mar
-10
Jun
-10
Sep
-10
Dec
-10
Mar
-11
Jun
-11
Sep
-11
Dec
-11
Mar
-12
Jun
-12
Sep
-12
Dec
-12
Mar
-13
Jun
-13
Sep
-13
Mo
nth
ly E
ne
rgy
Yie
ld [
GW
h]
Mo
nth
ly W
ind
Sp
ee
d [
m/s
]
Month
Reference Wind Speed TSMCP Site Wind Speed Net Yield Model Net Yield
Validation Methodology
0
2
4
6
8
10
12
14
Sep
-07
Dec
-07
Mar
-08
Jun
-08
Sep
-08
Dec
-08
Mar
-09
Jun
-09
Sep
-09
Dec
-09
Mar
-10
Jun
-10
Sep
-10
Dec
-10
Mar
-11
Jun
-11
Sep
-11
Dec
-11
Mar
-12
Jun
-12
Sep
-12
Dec
-12
Mar
-13
Jun
-13
Sep
-13
Mo
nth
ly E
ne
rgy
Yie
ld [
GW
h]
Month
Net Yield Model Net Yield
• Wind farm climate and yield modelled on an hourly basis
• Model is run for every turbine, not just the WF as a black box
• Model identical to pre-construction yield model, but by turbine by hour
Validation Methodology
0
2
4
6
8
10
12
14
Sep
-07
Dec
-07
Mar
-08
Jun
-08
Sep
-08
Dec
-08
Mar
-09
Jun
-09
Sep
-09
Dec
-09
Mar
-10
Jun
-10
Sep
-10
Dec
-10
Mar
-11
Jun
-11
Sep
-11
Dec
-11
Mar
-12
Jun
-12
Sep
-12
Dec
-12
Mar
-13
Jun
-13
Sep
-13
Mo
nth
ly E
ne
rgy
Yie
ld [
GW
h]
Month
Monthly Average Model Net Yield Model Net Yield
• Wind farm climate and yield modelled on an hourly basis
• Model is run for every turbine, not just the WF as a black box
• Model identical to pre-construction yield model, but by turbine by hour
Problem 3: Measured production can differ from the model not
just due to availability, but also due to other losses (icing,
curtailment, sub-optimal operation, ...)
Solution: Quantify these “running losses” and account for them
in the comparison with the model.
IcingTurbines
de-rated
High wind
hysteresis
Validation Methodology
Automatic
categorisation of
all 10-minute
SCADA data
Lessons Learned
• Total gross to net losses were reasonably accurate for most projects, but
accuracy of loss breakdown can be improved
• Example: default downtime losses can be set to match historic data
Turbine B.O.P. Grid
Measured Downtime (Time-Based)
Measured Downtime Loss (Energy-Based)
RES Standard Downtime Assumption
Lessons Learned
• Using a CFD flow model helps avoid over-predictions at specific turbines
• CFD flow model improves prediction accuracy by:
• 0.7% for the fleet
• Up to 5% for individual sites
• Up to 20% for individual turbines
0
20
40
60
80
100
120
-30% -25% -20% -15% -10% -5% 0% 5% 10% 15% 20%
Num
ber
of
Turb
ines
Energy Prediction Error
Linear CFD
Over prediction Under prediction
Lessons Learned
• TSMCP can have considerable diurnal and seasonal errors which also
show up in the energy validation results
• Downscaling of ReAnalysis data can help to some extent
• Analogue Ensemble (alternative to linear TSMCP) appears promising
-8%
-6%
-4%
-2%
0%
2%
4%
6%
8%
Jan Feb Mar Apr May Jun Jul Aug Sep Oct Nov Dec
Ene
rgy
Pre
dic
tio
n E
rro
r
Month of Year
Energy Prediction Error by Month of Year
UK&I Southern France
• A time series approach to energy yield validation has been developed
and implemented on a fleet of 31 wind farms
• This approach to validation has allowed RES to ensure an unbiased pre-
construction methodology
• Three key challenges to validation have been addressed:
1. Changes to the wind farm and its surroundings since construction
2. Weather during the operational period which differs to long-term
3. Operational losses which add noise to the comparison
• There is ample opportunity for further improvement to the model:
• Flow modelling
• Seasonal bias
Summary
Suspected
icingTurbines
de-rated
High wind
hysteresis
Any Questions?