Wind Farm Controller Design and Testing Using LongSim€¦ · Wind farm controller design and testing using LongSim 5th Wind Energy Systems Engineering Workshop Pamplona, Spain

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


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.


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


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


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


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 [°]


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?


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


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 %


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).


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


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


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


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


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!


Thanks for listening!

[email protected]

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.


http:www.dnvgl.commailto:[email protected]

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)


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


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 (

Documents

Wind Farm Controller Design and Testing Using LongSim€¦ · Wind farm controller design and testing using LongSim 5th Wind Energy Systems Engineering Workshop Pamplona, Spain