Farm Control Interest & (Selected) Activities in DTU Wind

Farm Control Interest & (Selected) Activities in DTU Wind

Tuhfe Göçmen & Gregor Giebel & Jonas Kazda& Paul Hulsman & Søren Juhl Andersen

DTU Wind Energy, RisøRoskilde DENMARK

DTU Wind Energy, Technical University of Denmark

DTU Wind vision

9/14/20182

Vision

• DTU Wind Energy is a globally leading department for wind energy with scientific and engineering competences in the international front and with a unique integration of research, education, innovation and public/private sector consulting.

• DTU Wind Energy is a key contributor to the realization of the vision of Denmark as a world leading wind power development centre and the activities support and develop the global wind energy sector with a special effort on national industrial development and innovation.


DTU Wind organisation

9/14/20183

Big Challenges – Big group

• Ten sections

• 5 “inter-sectional” programmes

o 3 technical programmes

• Many internal and external projects

o DK national & international

DTU Wind involvement in WFCExternal Projects

• TotalControl – EU (2018)

• DemoWind – EU (late 2017)

• Concert – PSO Energinet.dk (2016)

o Successor of PossPOW – PSO (2012)


DTU Wind involvement in WFC

9/14/20184

TotalControl

• EU project, 4 years, 4.8M€: DTU, KU Leuven, SINTEF, DNV GL, ORE Catapult, Equinor, Vattenfall, SGRE

• Open loop incl. testing at Lillgrund, control of electric system, turbine control testing at Levenmouth, closed loop, plus high-fidelity open toolbox - Website: totalcontrolproject.eu

Lillgrund tests Levenmouth tests

ORE Catapult’s Samsung 7MW prototype turbine at Levenmouth, to be used for turbine control trials.

Vattenfalls Lillgrund wind farm, where a full-scale test will be run, monitored with two synchronized lidars.Lillgrund image © www.siemens.com/press,

Lidar scan pattern from Windscanner.eu.



9/14/20185

DemoWind – Wind Farm Control Trials

• EU funded €2.3 million project, Wind Farm Control Trials (WFCT), with the aim of demonstrating new control strategies to improve energy yield and reduce operational and maintenance (O&M) costs. The WFCT trials involve the demonstration of new control strategies on an operational wind farm to understand their impact on improving the energy generation of a whole wind farm.

• Keywords: Full size offshore wind farm, wind scanner, staring lidars, strain gauges.

• Trials start (hopefully) spring 2019.

• Investigates mostly yaw steering, but also induction control.

https://www.carbontrust.com/offshore-wind/owa/demonstration/wfct/

https://www.carbontrust.com/offshore-wind/owa/demonstration/wfct/



9/14/20186

Concert - Control and Uncertainties in Real-Time Power Curves of Offshore Wind Power Plants

• PSO Energinet.dk national funded 4.8mDKK WP3 : Multi-Objective Wind Farm Control

• A real-time load estimation procedure

o Better load management under down-regulation

• Development of WPP controller

o Including up-regulation and yaw steering

• Integration and implementation of the dynamic WPP controller

o Taking uncertainties into account

WP2 : Estimation and Mitigation of Uncertainty in (Available/Possible) Active Power

• Uncertainty Quantification for the real-time possible power

• Convolution of the uncertainty in the real-time power & forecast available power

• Enhancement of the available power algorithm (both forecast and real-time) using machine learning uncertainty reduction techniques


DTU Wind – (~ first) involvement in WFC

9/14/20187

PossPOW - Possible Power Estimation of Down-regulated Offshore Wind Farms

• Power System integration / Electricity Market / Wake Modelling

PossPOW aims to correct the reduced wake during down-regulation!

• Different Problems / Different time scales

• Ultimately, regulated by the grid codes (especially offshore)

o DK : Energinet.dk Data collection @ 5mins, quality check @ 15mins. The error should be within ±5% of the actual power

• Focus on operational wind farms

o Purely SCADA based modelling 1Hz data

• Modular structure

o Resources, available models, quality criteria, etc.

• Validation via field experiments in Horns Rev-I


8

PossPOW – Rotor Effective Wind Speed Estimation

9/14/2018

DTU Wind Energy, Technical University of Denmark 9/14/20189

PossPOW – Real-time WakeRe-calibration of Larsen Model

𝒖𝒖𝒙𝒙 𝒙𝒙, 𝒓𝒓 = −𝑼𝑼∞𝟗𝟗 𝒄𝒄𝑻𝑻𝑨𝑨 𝒙𝒙𝟎𝟎 + ∆𝒙𝒙 −𝟐𝟐

𝟏𝟏𝟑𝟑 𝒓𝒓

𝟑𝟑𝟐𝟐 𝟑𝟑𝒄𝒄𝟏𝟏𝟐𝟐𝒄𝒄𝑻𝑻𝑨𝑨 𝒙𝒙𝟎𝟎 + ∆𝒙𝒙

−𝟏𝟏𝟐𝟐 −𝟑𝟑𝟑𝟑𝟐𝟐𝟐𝟐

𝟑𝟑𝟏𝟏𝟎𝟎

𝟑𝟑𝒄𝒄𝟏𝟏𝟐𝟐−𝟏𝟏𝟑𝟑

𝟐𝟐

• 2 variables to adjust:• First stage of calibration

o Nonlinear LSE fit in Thanet prior parameter distribution, no uncertainty Horns Rev-I test case

• Simple time delay based advection velocity in the wakeo Updated @ every 15-mins

• Pragmatic meanderingꜞ �𝑢𝑢𝑒𝑒𝑒𝑒𝑒𝑒 = 𝑢𝑢𝑒𝑒𝑒𝑒𝑒𝑒 1 + 7.12 𝜎𝜎𝜃𝜃𝑏𝑏

2 −12

𝒙𝒙𝟎𝟎 = 𝒑𝒑𝟏𝟏 � 𝒄𝒄𝐓𝐓𝒑𝒑𝟐𝟐 +𝒑𝒑𝟑𝟑 � 𝑻𝑻𝑻𝑻 𝒄𝒄𝟏𝟏 = 𝒑𝒑𝟒𝟒 � 𝒄𝒄𝐓𝐓

𝒑𝒑𝟑𝟑 + 𝒑𝒑𝟔𝟔 � 𝑻𝑻𝑻𝑻

“Gunner Chr Larsen.A simple wake calculation procedure. Risø National Labaratory, Roskilde, Denmark,1988”

“Gunner C Larsen. A simple stationary semi-analytical wake model. Technical Report August, Risø DTU,2009”

ꜞJohn F Ainslie. Wake modelling and prediction of turbulence properties. In Wind Energy Conversion, 1986: Proceedings of the 8th British Wind Energy Association Conference, volume 8 of Lecture Notes in Computer Science, UK, 1986. Cambridge


PossPOW – Real-time WakeRe-calibration of Larsen Model

• Better recovery with the re-calibration


Photo courtesy : Vattenfall & Hasager, Charlotte Bay, et al. "Wind farm wake: The Horns Rev photo case." Energies 6.2 (2013): 696-716.

in normal operation:

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃 𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 = 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑵𝑵𝑵𝑵𝑵𝑵𝑻𝑻𝑻𝑻𝑵𝑵𝑵𝑵 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻due to upstream curtailment:

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝒄𝒄𝑻𝑻𝑻𝑻𝑨𝑨 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 > 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑹𝑹𝑻𝑻𝑨𝑨𝑻𝑻𝒓𝒓𝑻𝑻𝑻𝑻𝒄𝒄𝑻𝑻 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻

what should be in normal operation:

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝒄𝒄𝑻𝑻𝑻𝑻𝑨𝑨 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 = 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑹𝑹𝑻𝑻𝑨𝑨𝑻𝑻𝒓𝒓𝑻𝑻𝑻𝑻𝒄𝒄𝑻𝑻 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻

PossPOW Validation - Down-Regulation Experiments in HR-I

9/14/2018


PossPOW Validation - Down-Regulation Experiments in HR-I

• Assume similarity between the rows• Avoid constant rated power

Normal wind direction Below rated wind speed


PossPOW Validation - Down-Regulation Experiments in HR-IWind Speed


PossPOW Validation - Down-Regulation Experiments in HR-IWind Speed


PossPOW Validation - Down-Regulation Experiments in HR-ITSO Requirements

15


PossPOW Validation - Down-Regulation Experiments in HR-IError Distribution• Total 6 experiments

% 𝑻𝑻𝒓𝒓𝒓𝒓𝑵𝑵𝒓𝒓 𝑨𝑨𝑨𝑨𝑵𝑵𝑻𝑻𝑵𝑵𝑵𝑵𝑻𝑻𝑵𝑵𝑻𝑻 𝑷𝑷𝑵𝑵𝑷𝑷𝑻𝑻𝒓𝒓 =𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃 𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝒄𝒄𝑻𝑻𝑻𝑻𝑨𝑨 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻 − 𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑹𝑹𝑻𝑻𝑨𝑨𝑻𝑻𝒓𝒓𝑻𝑻𝑻𝑻𝒄𝒄𝑻𝑻 𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝒄𝒄𝑻𝑻𝑻𝑻𝑨𝑨 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻

𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴𝐴 𝑃𝑃𝑃𝑃𝑃𝑃𝐴𝐴𝑃𝑃𝑹𝑹𝑻𝑻𝑨𝑨𝑻𝑻𝒓𝒓𝑻𝑻𝑻𝑻𝒄𝒄𝑻𝑻 𝑨𝑨𝑨𝑨𝑨𝑨𝑻𝑻𝒄𝒄𝑻𝑻𝑻𝑻𝑨𝑨 𝑻𝑻𝒖𝒖𝒓𝒓𝑻𝑻𝑻𝑻𝑻𝑻𝑻𝑻� 100


PossPOW to CONCERT – Uncertainty Focus

9/14/201817

PossPOW - Possible Power Estimation of Down-regulated Offshore Wind Farms

• Power System integration / Electricity Market / Wake Modelling




o DK : Energinet.dk Data collection @ 5mins, quality check @ 15mins. The error should be within ±5% of the actual power









o DE : TSO Consortium Data collection @ 1min, quality check @ 1mins. The std of the 1min error < 5% (pilot phase)







• Re-calibrated Larsen WF (and time) specific parameters

CONCERT – Uncertainty Focus Re-Re-calibration of Larsen Model


• Gaussian Deficit in 1Hz SCADA?

𝑘𝑘∗ = 𝐴𝐴 𝑇𝑇𝑇𝑇 + 𝐴𝐴

𝐴𝐴 = 0.3837𝐴𝐴 = 0.003678

CONCERT – Uncertainty Focus Gaussian Deficit Model

Majid Bastankhah and Fernando Port e-Agel. A new analytical model for wind-turbine wakes. Renewable Energy, 70:116–123, 2014

Mahdi Abkar and Fernando Porte-Agel. Influence of atmospheric stability on wind-turbine wakes: A large-eddy simulation study. Physics of Fluids, 27(3):1–20, 2015


• Gaussian Deficit Bayesian Re-calibration

𝑘𝑘∗ = 𝐴𝐴 𝑇𝑇𝑇𝑇 + 𝐴𝐴

𝐴𝐴 = 0.3837𝐴𝐴 = 0.003678

CONCERT – Uncertainty Focus Re-calibration of Gaussian Deficit Model


• Machine Learning Platform – TensorFlow– With Keras wrapper in Python– Fast & easy to apply

• The deep learning algorithm – LSTM– Long Short-term Memory– Special building unit for RNN– Shown to perform faster & better for

highly fluctuating time series

http://colah.github.io/posts/2015-08-Understanding-LSTMs/

• The inputs from the upstream turbines– Defined at every minute (WD dependent)– WD, Ueff, std(Ueff), ct + uncertainties– Data fed for the previous 1-hour

• Time window of 1-hour shiftedforward at every minute

Input intervalOutput interval

CONCERT – Uncertainty Focus Machine Learning for short-term wakes



• Machine Learning Platform – TensorFlow– With Keras wrapper in Python– Fast & easy to apply

• The deep learning algorithm – LSTM– Long Short-term Memory– Special building unit for RNN– Shown to perform faster & better for

highly fluctuating time series


• The inputs from the upstream turbines– Defined at every minute (WD dependent)– WD, Ueff, std(Ueff), ct + uncertainties– Data fed for the previous 1-hour



Input intervalOutput interval


DTU Wind Energy, Technical University of Denmark23

• Machine Learning Platform – TensorFlow• The inputs from the upstream turbines

– Defined at every minute (WD dependent)– WD, Ueff, std(Ueff), ct + uncertainties– Data fed for the previous 1-hour


• The output – Ueff at the downstream turbine

• New network (or model) per WF per turbine per minute

– Still feasible real time! • 20 epochs• Batch size = 64• Single hidden layer with 18 neurons



• Also modular, operational wind farm controller containing variety of

• We’re working on a validation database!– Reference wind farm / digital twin – (Hopefully) research wind farm in Risø

CONCERT – Wind farm ControlDTU Wind Farm Controller

Basic DTU Wind Energy

Controller (open-source)

& more

Online dataprocessing procedures

Wind farm operation models, wake models,

load estimators & reference WF

FugaPossPOW(ers)

HAWC2

Data storage

DWM

FLEX5

Ellipsys3D


• SWIFT Facility

• Lubbock Texas• 15th of December 2016, 20:09 to 22: 52• Measurement data from

– METa1, WTGa1, DTU Spinner Lidar (2.5D downstream)– 𝜇𝜇32𝑚𝑚 = 8.3 𝑚𝑚/𝑃𝑃, 𝑇𝑇𝑇𝑇32𝑚𝑚 = 0.14

CONCERT – Wind farm ControlWake Steering


• SWIFT & LES - inflow

• Inflow for LES matched with the measurement data



• SWIFT & LES - downstream

• Stability? • FLORIS agrees better – deeper wake.



• LES + FLEX5 – fatigue loads : e.g. TBBMy



• Optimisation – Power– FLORIS (Gaussian Deficit)

o Not realiable near wakeo Bias could be fitted

o Std critical– Surrogates?

o PCEo Benchmark & Validation for loads


Nikolay Dimitrov, Mark Kelly, Andrea Vignaroli, and Jacob Berg, From wind to loads: wind turbine site-specific load estimation using databases with high-fidelity load simulations, Wind Energy Science Journal, 2018.

Juan PabloMurcia, Pierre- Elouan Réthoré, NikolayDimitrov, AnandNatarajan, John Dalsgaard Sørensen, Peter Graf, Taeseong Kim, Uncertainty propagation through an aeroelastic wind turbine model using polynomial surrogates, Renewable Energy, 2018

https://www.wind-energ-sci-discuss.net/wes-2018-18/

https://www.sciencedirect.com/science/article/pii/S0960148117306985?via%3Dihub


• Optimisation – Power– Surrogate : Polynomial Chaos Expansion




• Large yaw angle & highest power gain at close spacing• Slight positive angle for the downstream turbine• App. 2.5% gain @2.5D mean model error = 5%, std model error = 4% - how to read that?



• Increasing weight in DEL– Weights = [Power, FlapM1, FlapM2, TBBMy1, TBBMy2]– Normalised wrt upstream DEL, zero yaw.


Conclusions – WFC in DTU Wind

•Focus on application–Bringing the tools together–Coupling, Calibration, Verification & Validation–Digitalization & Big data

•System perspective–Wake reduction and grid services–WFC in siting & layout optimization

•Stay tuned for the outcomes of the new projects!–Let’s make more together


Thanks for your [email protected]

Documents

Farm Control Interest & (Selected) Activities in DTU Wind