View
3
Download
0
Category
Preview:
Citation preview
DNV GL © 02 October 2019
Wind farm controller design and testing using LongSim
5th Wind Energy Systems Engineering Workshop
Pamplona, Spain
Ervin Bossanyi 02 October 2019
DNV GL © SAFER, SMARTER, GREENER 1
Controlling wakes in wind farms
1. What is the optimum* distribution of power and yaw setpoints for all the turbines, in this wind condition?
2. How can we maintain optimum* performance in dynamically changing circumstances?
Traditional sector management • Reduced power! • increased loading!
Switch this turbine off?
Or reduce the power set-point of this one?
“Induction control”
Or maybe yaw the turbine slightly to steer its wake away from the
next turbine? “Wake steering”
* Optimum has to be defined – depends on energy and loading
2 2 DNV GL © 02 October 2019
Different approaches to control design
• Quasi-static open-loop control, or “Advanced Sector Management” – Optimised set-points pre-calculated for each wind condition (as a function of wind speed, direction, turbulence, …)
– Wind condition defined e.g. by met mast or SCADA data (filtered Þ slow response)
– OK as long as wind conditions are slowly-varying
– Re-optimise when something changes (e.g. energy price, turbine maintenance, etc., etc.)
• Dynamic closed-loop control (more advanced, many possible approaches) – e.g. MPC, with continuous feedback from measurements all over the wind farm
– Potentially rapid response
– In principle, should be capable of better performance … … … but is it practical?
• Machine learning approaches – Using domain knowledge (not just ‘black box’)
All can be tested in LongSim
DNV GL © 02 October 2019 3
Modelling requirements
• Detailed representation of turbine wakes in different atmospheric conditions
• Realistic, time-varying wind conditions • Accurate modelling of turbine control
dynamics
Fidelity of flow modelling
• Needs time-domain simulations – Long enough to capture low-frequency wind
variations (hours, days, weeks) – Short enough timestep (~1s) to capture
principal turbine and wind farm control dynamics
– Fast enough to run many repeat simulations for design iterations
Speed, computational resources
“Engineering” e.g. LongSim
RANS CFD
LES
DNV GL © 02 October 2019 4
Time-domain simulation - LongSim
• Choice of engineering wake models, embedded in stochastic flow field
• Wake meandering and advection
• Turbine details, including supervisory control
• Wind farm control algorithm – Estimation of wind conditions from turbine signals – Setpoint lookup – Setpoint implementation
Sedini example with wake steering
• Wind field generated from historical site data (met mast)
• Test & tune control algorithm details
• Test controller against different wake models
• Evaluate power increase, yaw actuator duty etc.
DNV GL © 02 October 2019 5
Wind farm control – design process
LongSim
LUT
Steady-state setpoint optimisation
Dynamic simulations Performance:
• Energy production • Yaw system duty • Loads • Etc.
Site data:
Wind
Compare!
SCADA
Robustness: Adjustment for uncertainty in wind conditions
Dynamic wind farm control algorithm
DNV GL © 02 October 2019
6
Sedini wind farm, Sardinia
• 43 GE 1.5MW turbines 1500 1 16 2 17 • Several experiments planned in CL-Windcon 3
1000 4 18 project
19 5 • Wake steering tests E7 20 500 6
21 7 28 • Wake steering and induction control field
22 8 E6 23 tests for a row of 9 turbines (this Mast 0
30 9 presentation) 10 24 11
12 25 – Preliminary control design and evaluation for -500 E5 13 wake steering 26 31
E4 32
– Final design for induction control tests (which are 33 -1000 34 14 currently in progress) 35
E3 15 36 – Preliminary field test results 37 E2 -1500 E1 38
-2000 -1000 -500 0 500 1000 1500 2000
DNV GL © 02 October 2019 7
235
Wake model makes a difference! (wake steering) Power gain (steady state) with optimised setpoints
Wake model: A1 E1 A1 but setpoints calculated with E1 Power ratio for WS Results1&2&3, wake A1 , TI=0.1 Power ratio for WS Results1&2&3, wake E1 , TI=0.1 PowerRatio WakeModel_A1.dat with WFC1&2&3_WS_E1.dat, TI=0.1
1.14 260 260 260
255 255 255 1.12
250 250 250 1.1
245 245 245
Dire
ctio
n, d
eg
Dire
ctio
n, d
eg
Dire
ctio
n, d
eg
1.08
240 240 240
1.06 235 235
1.04 230 230 230
1.02 225 225 225
220 220 1 220 6 6.5 7 7.5 8 8.5 9 9.5 10 6 6.5 7 7.5 8 8.5 9 9.5 10 6 6.5 7 7.5 8 8.5 9 9.5 10
Wind, m/s Wind, m/s Wind, m/s
DNV GL © 02 October 2019
8
35
30
25
0
Normalised power:A4-34/A4-38 3
2.5 Wake model investigation, using SCADA data RMS error summed over turbines for each wake model 2
1.5
P ar
get /
Pef
eren
ce [-
] P
arge
t / P
efer
ence
[-]
tr
tr
20.6
8 24.
7721
.45
20.2
5 24.
40
21.1
3
21.4
328
.88
24.1
2
23.4
632
.11
27.1
2
21.2
9 23.9
420
.52
21.6
423
.87
20.5
6
20.5
2 24.
2620
.22
19.3
0 23
.61
20.4
5 23.5
7 25.9
622
.65
19.4
5 25
.25
21.3
7
1
0.5
RM
S e
rror
20 0 180 200 220 240 260 280 300 320
15 Wind Direction [°]
Normalised power:A4-31/A4-38 3
10 Period 1
2.5 Period 2
Period 3 5
2
Ains
lieM
OL
Ains
lieM
OL_
QA
Ains
lieSP
4
Ains
lieSt
anda
rd
Ains
lieSu
mO
fDef
s
Ains
lie_H
_SoD
Ains
lie_H
_SoD
_Exa
ct
Ains
lieM
OL_
QA_
Exac
t
EPFL
Ains
lieM
OL_
QA_
WS_
Exac
t
1.5
1
0.5
0 180 200 220 240 260 280 300 320
Wake model Wind Direction [°]
DNV GL © 02 October 2019 9
Wake models
•Standard Ainslie: industry standard for energy calculations •Standard EPFL: popular for wake studies; several tunable parameters, so can always be made to fit reasonably well
•Modified Ainslie (selected): – Stability correction to eddy viscosity (using Obukhov length derived from historical mast data) – Wake superposition: sum-of-deficits instead of large wind farm corrections
– Removal of some approximations in the standard model
– General applicability: no tunable parameters – Allows possibility for control to track measured stability (e.g. using sonic anemometer), in
addition to wind speed, direction and turbulence.
10 DNV GL © 02 October 2019
Need for dynamic simulations
• How do wakes behave dynamically? – Meandering, advection
• How do wind conditions vary in practice, and how well can we follow this? – Robust (smoothed) setpoints to account for uncertainties in wind condition?
• How do wind conditions vary across the farm at any time? – Assume ambient conditions are the same everywhere in the farm at any time? – Smoothed setpoints again, or allow variation of estimated wind conditions?
• How to measure the wind condition? – Met mast if available? – Average of conditions from SCADA at unwaked turbines? – Filtering, to be representative of propagation through the farm – Variations across the farm?
• How often to update the control? – Tracking accuracy vs. smoothness of control action
• How to implement the setpoint changes at the turbine – Especially for yaw control. Consider overriding the turbine yaw logic. – How to handle ‘flipping’ of yaw offset as wind direction changes?
11 DNV GL © 02 October 2019
Induction control for Sedini: Dynamic simulation (5 hour period) 9 Wind conditions (from met mast) Setpoint example (#37) 8
11.0 (raw, smoothed) 10.5 7 10.0
FarmWindDirection
FarmWindSpeed
TurbulenceIntensity
[deg]
[m/s]
[%]
6 9.5
9.0 5 8.5
8.0 4
7.5
7.0 3 0 50 100 150 200 250 300 350
Time [min] 2 246
244 1
242 0 240 0 50 100 150 200 250 300 350
238 Time [min] 15 236
Dynamic 234 (60s ave) Power increase
232 Quasi-static 230 (9 turbines) Dynamic result 10 0 50 100 150 200 250 300 350
Time [min] 16 is actually 15 14 13 better than 12
Powe
r inc
reas
e, % 5
steady-state 0 11 10
8 7
expectation! -5 6 Mean increase: Dynamic: 1.58%, QS: 0.94% 0 50 100 150 200 250 300 350 -10
0 1 2 3 4 5 Time [min] Time, hours
12 DNV GL © 02 October 2019
9
Wake steering example – large wind farm
•Yaw setpoints optimised in steady state (whole wind farm)
•Dynamic simulations achieved 25-50% of steady-state expectation
•Can be improved: – Better handling of yaw logic – Consideration of wind direction variations
across farm
Annual Energy Production increase
Steady-state 2.8 – 6.6 %
Dynamic ~ 0.7 – 3.3 %
13 DNV GL © 02 October 2019
Wake steering example – Lillgrund wind farm
• Yaw setpoints optimised in steady state (whole wind farm)
• Dynamic simulations actually achieving more than steady-state expectation – 6.5% dynamic compared to 2.6% quasi-static in this example
Direction: wind from North 40
1500
1000 30
500
20 0
-500
10 -1000
-1500 0
0 1 2 3 4 5 6 7 8 9 10 -2000 -1500 -1000 -500 0 500 1000 1500 2000
A-01
A-02
A-03
A-04
A-05
A-06
A-07
B-01
B-02
B-03
B-04
B-05
B-06
B-07
B-08
C-01
C-02
C-03
C-04
C-05
C-06
C-07
C-08
D-01
D-02
D-03
D-04
D-06
D-07
D-08
E-01
E-02
E-03
E-04
E-06
E-07
F-02
F-03
F-04
F-05
F-06
G-02
G-03
G-04
G-05
H-02
H-03
H-04
Base case Wake steering Quasi-static
Win
d fa
rm p
ower
, MW
Time, hours
This project has received funding from the European Union’s Horizon 2020 research and innovation programme under grant agreement no. 727680 (TotalControl, website: www.totalcontrolproject.eu).
14 DNV GL © 02 October 2019
FIELD TEST RESULTS DATA FROM 11TH JULY TO 17TH SEPTEMBER
Sedini_InductionCtrl_test_July11-Sept9.mat, Chunks of 900s after first 300s, increase = 6.2051% 15
Sedini_InductionCtrl_test_July11-Sept9.mat, Chunks of 900s after first 300s, increase = 3.9424%
Mea
sure
d Po
wer /
Pow
er c
urve
pow
er
On (34.5944) Off (33.2823) On (193.0294)
Off (181.7514)
5 6 7 8 9 10 11 12
1
0.8
Powe
r [M
W] 10
5 0.6
0.4
0.2 200 0 210 220 230 240 250 260 270
Direction Wind speed Number of points per bin (total: 92)
Number of points per bin (total: 92) 25 20
15
10
5
0
1 0 1 4
6
14
21
2 0 0
5 5
12
21 On (47) Off (45)
0
5
17
10 10
4 1 0
8
13 12
7
3 2
On (47) Off (45)
20
15
10
5
0 200 210 220 230 240 250 260 270
5 6 7 8 9 10 11 12 Direction Wind speed 8
• Not enough points yet, but
Powe
r [M
W] 6
2
0 200 210 220 230 240 250 260 270
Direction
On (193.2046) Off (178.4753)
initial results are promising. 4 • Measurements are continuing
DNV GL © 02 October 2019
COMPANION SIMULATIONS USING LONGSIM TO TRY TO REPRODUCE MEASURED BEHAVIOUR
• SCADA measured 1-minute wind data used as wind conditions at the position of turbine #38.
• Use this to generate a wind field covering all the turbines.
• LongSim time-domain simulations run using that wind field.
• Simulated performance of the turbines compared to SCADA measurements.
Invaluable for understanding what’s happening on site!
Turbine Power: WTG-34 Turbine Power: WTG-33 Farm power Turbine setpoint: WTG-35
0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18 0 2 4 6 8 10 12 14 16 18 Hours Hours Hours Hours
0
2
4
6
8
10
12
14 LongSim SCADA
-0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4 LongSim SCADA
0 2 4 6 8 10 12 14 16 18 -0.2
0
0.2
0.4
0.6
0.8
1
1.2
1.4
1.6 LongSim SCADA
0
0.5
1
1.5
2
2.5
3 LongSim SCADA
DNV GL © 02 October 2019
LongSim to optimise wind farm yaw control • Each turbine makes use of information from its
neighbours
• Spatial averaging (reduces the need for time-averaging)
• Weighted average favouring the turbines nearest to the ‘focal point’ – Exponential decay of weighting with distance
• Position the focal point further upstream, to provide some useful preview
Tested in LongSim: • Using the layout of Horns Rev 1
• Correlated wind field generated from met mast data (actually from FINO-1)
• Wakes, with meandering (small uncertainty: is wind direction changed by wake effects?) -3000 -2000 -1000 0 1000 2000 3000
-2000
-1500
-1000
-500
0
500
1000
1500
2000 #1
#2
#3
#4
#5
#6
#7
#8
#9
#10
#11
#12
#13
#14
#15
#16
#17
#18
#19
#20
#21
#22
#23
#24
#25
#26
#27
#28
#29
#30
#31
#32
#33
#34
#35
#36
#37
#38
#39
#40
#41
#42
#43
#44
#45
#46
#47
#48
#49
#50
#51
#52
#53
#54
#55
#56
#57
#58
#59
#60
#61
#62
#63
#64
#65
#66
#67
#68
#69
#70
#71
#72
#73
#74
#75
#76
#77
#78
#79
#80
17 DNV GL © 02 October 2019
Simulation results Trade-off – power production vs yaw system duty
• Yaw duty represented by total yaw travel (circles) or number of yaw events (crosses).
• Purple: Turbine yaw, Red: Central yaw
• Central yaw slightly increases power production while significantly reducing yaw travel
Mea
n p
ower
, M
W
Central Yaw: Yaw travel, deg Turbine Yaw: Yaw travel, deg Central Yaw: Yaw events Turbine Yaw: Yaw events
72.90
C3 C3 C4 C4 T3 T3 72.85
C2 C1 C1 T2
72.80 C5 C5
72.75
T1 T1
C2
T2
Yaw travel: -24%
Energy: +0.206% (+0.219% with reduced yaw drive power consumption)
72.70 0 1000 2000 3000 4000 5000 6000 7000 8000
Yaw duty
18 DNV GL © 02 October 2019
Summary – use of LongSim
• Steady-state set-point optimisation for wake steering & induction control
• Time-domain simulation for testing any controllers: – Active wake control
– Wind turbine and wind farm yaw control algorithms
• Sub-second time step, but long simulations (hours ® many days) • Realistic varying wind conditions: site-specific
• Field tests of algorithms designed and tested with LongSim are in progress
• Companion simulations of field tests are proving invaluable to understand what’s going on!
19 DNV GL © 02 October 2019
Thanks for listening!
ervin.bossanyi@dnvgl.com
www.dnvgl.com
The trademarks DNV GL®, DNV®, the Horizon Graphic and Det Norske Veritas® SAFER, SMARTER, GREENER are the properties of companies in the Det Norske Veritas group. All rights reserved.
20 DNV GL © 02 October 2019
http:www.dnvgl.commailto:ervin.bossanyi@dnvgl.com
LongSim model – what is it?
• Range of different engineering wake models available
• Steady-state setpoint optimisation ® setpoint LUTs for different wind conditions, for each turbine
• Dynamic time-domain simulation: – Long simulations (hours, days or more), fast timestep (~1s)
– Correlated turbulent wind field across the wind farm, with time-varying mean conditions (e.g. from met mast data)
– Dynamic model of wakes (superimposed on the ambient flow) – Meandering, advection, deflection
– Turbine dynamics: rotor speed, pitch, speed & power control, supervisory control (including yaw control)
– Dynamic implementation of wind farm control algorithm – Estimation of wind conditions
– Setpoint lookup
– Implementation of setpoint changes
– Output of power, loads (indirectly, from database), supervisory control details (e.g. yaw manoeuvres)
21 DNV GL © 02 October 2019
Some next steps
• Full-scale measurements – Field tests need careful design – need to measure small changes – Some tests already reported in literature, beginning to show promise – Sedini (in progress, CL-Windcon project) – Other field tests in planning
• Power and loads optimisation – Demonstrated previously in simulation
Validation of wake models
Validation of control
effectiveness
– Not easy to define waked loads appropriately – Not easy to combine energy and loads into an economic cost function – Less immediate commercial interest in loads (but it is starting!)
• Load equalisation – During curtailment – Over lifetime
• More advanced control algorithms – Closed loop, e.g. MPC – AI / machine learning
Characterisation of waked turbine loads
22 DNV GL © 02 October 2019
Conclusions Can wind farms realistically benefit from active control of turbine wakes?
• Almost certainly, yes – Still many uncertainties in modelling – More field testing required – Wake steering may be more effective, but more problematic to implement – Induction control: the jury is still out (evidence in favour is beginning to return). More straightforward to implement. – Dynamic induction control: still at the research stage – could be promising but still many questions
• Modest energy gains – Very dependent on the situation (especially wind farm layout and wind rose) – Could be several percent – very valuable – Even small gains (
Recommended