DNV GL © 05 February 2019 SAFER, SMARTER, GREENERDNV GL ©
05 February 2019
Renewable Energy Analytics
WICE 2.0 – The new generation of ice loss models
1
Stefan Söderberg, Jon Collins, Till Beckford, and Carla Ribeiro
Winterwind 2019, Umeå
DNV GL © 05 February 2019
Contents
▪ Part 1: Introduction and Methodology
▪ Part 2: Validation of WICE model chain
– Weather data
– Model grid resolution and terrain elevation
– Long term correction
▪ Part 3: Improvements – WICE2.0
2
DNV GL © 05 February 2019
Part 1: Introduction and Methodology
3
DNV GL © 05 February 2019
Introduction
▪ In September 2018 DNV GL joined forces with WeatherTech.
4
DNV GL © 05 February 2019
Combined expertise
DNV GL
Empirical production loss model based on
considerable amount of
production and meteorological
data
WeatherTech
A combined atmospheric and machine
learning model to predict production losses, the
WICE model.
5
DNV GL © 05 February 2019
Nordic operational data – Icing vs Elevation
6
0
5
10
15
20
0 250 500 750 1000
An
nu
al
en
erg
y l
oss d
ue t
o I
cin
g [
%]
Effective hub-height elevation
South Sweden
Mid Sweden
North East Sweden
Norway
Finland
Anti/De icing
Terrain sheltering
DNV GL © 05 February 2019
Production loss model - WICE
7
A combination of physical and statistical modelling
ANN – Artificial Neural Network Following IEA Task 19
DNV GL © 05 February 2019
WICE model chain
8
• WRF model
• High resolution
• Microphysics
• Makkonen
• Turbine blade
• IPS
2016
• 30yrs
• Condensates
NWP Ice modelProduction
lossLong term
2014 201520122013 2013
2017
• WICE
• SCADA
2013
QUESTIONS? Contact Dorothy Henry or Ioana Petricean +32 2 740 22 39
Speaker Management [email protected]
(Conference Secretariat MCI Brussels) www.ewea.org/annual2013
GUIDELINES FOR ORAL PRESENTERS
CONTENTS
A. SPECIFIC INFORMATION FOR ORAL PRESENTERS
Before the event
1. Important dates
2. Session information
3. PowerPoint presentation submission
4. Biography submission
5. Full paper submission
6. PowerPoint presentation guidelines
7. Technical requirements
8. Substitutions
During the event
9. Speakers’ Room
10. Session briefing
11. Session rooms
12. Checklist
B. GENERAL PRACTICAL INFORMATION ABOUT EWEA 2013
13. Arrival at the venue
14. Registration
15. Networking and social events
16. Hotel & flight bookings
17. Contact
Thank you for agreeing to present at EWEA 2013. P lease take a moment to read the
speaker ’s guidelines below.
Updates to th is information wil l be sent to you by emai l .
If you are also preparing a poster presentation, please see the speci fic guidelines in
your poster presenter portal .
2018
DNV GL © 05 February 2019
Part 2: Validation of WICE model chain
9
DNV GL © 05 February 2019
Validation
10
▪ 10 sites
▪ Chosen to cover different regions
SCADA
▪ 1-7 years of operational data
▪ 263 turbines
Weather modelling
▪ WRF
▪ In-house setup
▪ 1000m / 333m model grid resolution
DNV GL © 05 February 2019
Validation – weather modelling
11
▪ Time series and statistics have been analysed
– In general a good agreement
DNV GL © 05 February 2019
Validation – WICE model chain: intra-farm variability
12
▪ Individual WTG losses
▪ SCADA vs WICE 1000m grid
resolution
DNV GL © 05 February 2019
Validation – WICE model chain: intra-farm variability
13
▪ Individual WTG losses
▪ SCADA vs WICE 1000m and
333m grid resolution
DNV GL © 05 February 2019
Validation – WICE model chain: intra-farm variability
14
▪ Individual WTG losses
– Terrain elevation vs SCADA
DNV GL © 05 February 2019
Validation – WICE model chain: intra-farm variability
15
▪ Individual WTG losses
– Terrain elevation vs SCADA
and WICE 1000m grid
resolution
DNV GL © 05 February 2019
Validation – WICE model chain: intra-farm variability
16
▪ Individual WTG losses
– Terrain elevation vs
SCADA, WICE 1000m and
333 grid resolution
DNV GL © 05 February 2019
Validation – WICE model chain: intra-farm variability
17
▪ Individual WTG losses
– Turbines arranged by
elevation.
SCADA WICE 1000m
WICE 333m
Mean 8.7 7.9 8.4
Max 14.6 9.4 10.5
Min 4.5 6.1 5.2
DNV GL © 05 February 2019
Validation – WICE model chain: long term correction
18
▪ Long term correction approach
– One year with high resolution model data.
– Long term reference modelled with coarser model grid resolution.
▪ Question:
– How sensitive is the estimated long term production loss to which year
that is modelled with high resolution?
DNV GL © 05 February 2019
Validation – WICE model chain: long term correction
19
▪ 4 years of WRF 1km and long term reference data
▪ Long term correction one season at a time
Year 1 2 3 4 Loss (%)
LT ref
High res
LT corr 7.1
High res
LT corr 6.6
High res
LT corr 6.1
High res
LT corr 6.0
High res 6.2
DNV GL © 05 February 2019
Validation – WICE model chain: production losses
20
▪ Total wind farm losses
– slope:1.0533
– std dev: 1.7064
– corr:0.9062
DNV GL © 05 February 2019
Validation – WICE model chain: production losses
21
▪ Total wind farm losses
– slope:1.0533
– std dev: 1.7064
– corr:0.9062
Without outlier:
– slope:1.0008
– std dev: 0.9667
– corr:0.9716
DNV GL © 05 February 2019
Part 3: Improvements – WICE2.0
22
DNV GL © 05 February 2019
WICE 2.0
23
▪ What is new
– Training data from new sites (and more coming soon)
– Machine learning improvements:
– ”Extreme Gradient Boosting” (XGBoost) – a tree-based model that typically performs better
than neural networks on this kind of multi-dimensional regression problem
– Refinements to feature selection and feature engineering
– Updated long term correction method
DNV GL © 05 February 2019
WICE 2.0
24
▪ Icing losses estimated by WICE2.0 are in line with previous DNV GL models
– Outliers are better represented
DNV GL © 05 February 2019
Conclusions
25
DNV GL © 05 February 2019
Conclusions
26
▪ In general a good agreement between modelled and observed wind and temperature
▪ Capture intra-farm variability quite well
– a higher model grid resolution is recommended for sites with complex terrain
▪ Estimated long term losses agree well with observed losses
▪ Icing losses estimated by WICE2.0 are in line with previous DNV GL models
DNV GL © 05 February 2019
SAFER, SMARTER, GREENER
www.dnvgl.com
The trademarks DNV GL®, DNV®, the Horizon Graphic and Det Norske Veritas®
are the properties of companies in the Det Norske Veritas group. All rights reserved.
Thanks for listening
27
Stefan Söderberg