1. FUSED-Wake
Framework for Unified System Engineering and
Design of wind farm Wake models
Pierre-Elouan Réthoré
Senior Researcher
DTU-Wind Energy, Risø
2. DTU Wind Energy, Technical University of Denmark
FUSED-Wind
• Collaborative effort between
DTU and NREL to create a
Framework for Unified
System Engineering and
Designed of Wind energy
plants.
• Based on OpenMDAO, a
python based Open source
framework for Multi-
Disciplinary Analysis and
Optimization.
PythonPython
H
A
W
C
2
FAST
Wind
Resource
Model
Flow
M
odel
Wake
Model
Cost
Model
3. DTU Wind Energy, Technical University of Denmark
FUSED-Wake
• The heart (and brain) of TopFarm II
• Based on FUSED-Wind
• Can run all the wake models of DTU with the same inputs and
outputs
FUGA
DWM
GCL NOJ(s)
EllipSys
AD RANS
EllipSys
AD LES
EllipSys
AL LES
EllipSys
FR LES
4. DTU Wind Energy, Technical University of Denmark
Research tool: Modularized concept
• The wind farm wake models are split into a generalized
workflow
Inflow
Generator
Inflow
Generator
WS positionsWS positions Wake
Accumulation
Wake
Accumulation Hub WSHub WS WT ModelWT Model
Wake
Model
Wake
Model
Stream wise
WTs
Stream wise
WTs
Upstream WTsUpstream WTs
RecorderRecorder
RecorderRecorder
Power
Curve
CFD ALFUGA
BEMLog LawMann
Precursor
CFD
LinearQuadraticMixed HAWC2
FUGA
EllipSys
AL/LES
5. DTU Wind Energy, Technical University of Denmark
Potential applications of the framework
• Model automatic selection
• Machine learning (model recalibration)
• Uncertainty quantification
• Model Averaging (combining the information of several
models)
• Multi-fidelity optimization
• Standard way to run wind farm models
• Bridging the gap between researchers and industry
• “Companion” to WindBench
–Automatically running all the benchmarks with the
same inputs / post processing
–Robust benchmarking (no expert user required)
6. DTU Wind Energy, Technical University of Denmark
WindBench companion
6 July 8, 2013
Benchmark manager
Post-processing
script
Post-processing
script
FUSED-Wake
Input
FUSED-Wake
Input
WakeBench
benchmark
WakeBench
benchmark
Cloud Cluster
Report
Web-graphs
For all models
Model manager
Wake ModelWake Model
FUSED-Wake
Wrapping
FUSED-Wake
Wrapping
7. DTU Wind Energy, Technical University of Denmark
Multi-fidelity & Machine Learning
• Each subcomponent, or several of them together can be
swapped to a different level of fidelity.
• Each subcomponent level of fidelity produces an intrinsic
uncertainty, dependent of the input-region.
• Swapping to higher fidelity might involve a
computation cost and offer a reduction in intrinsic
uncertainty in return.
• Running a higher fidelity can potentially re-calibrate the
lower fidelity models, and lowering its intrinsic
uncertainty within the specific input-region.
8. DTU Wind Energy, Technical University of Denmark
Input-region - Re-calibration cascade
FUGA
DWM
GCL NOJ
EllipSys
AD RANS
EllipSys
AD LES
EllipSys
AL LES
Time[log]
Intrinsic Uncertainty [-]
EllipSys
FR LES
SCADA
data
SCADA
data
9. DTU Wind Energy, Technical University of Denmark
Parameter calibration
• Re-analysis of SCADA data using LES
• Inverse uncertainty quantification
–Bias correction and Parameter calibration
–Bayesian inference
–Optimal maps
–Kalman filters
Experiment
Variables ParametersModel
Experimental uncertainty
Bias function
10. DTU Wind Energy, Technical University of Denmark
Towards a higher level of science
• Including the uncertainty of the models in the results:
–Parameter uncertainty estimation
–Input uncertainty elicitation
–Uncertainty propagation to the outputs
–Model inaccuracy
–Code inaccuracy
• Deterministic model => stochastic model
–The framework could automatize this workflow
11. DTU Wind Energy, Technical University of Denmark
Next steps
• Gathering interest group
• Alpha release to interest group
• Public release of beta version
• Forming a project portfolio to coordinate the efforts
Status
• Framework in alpha version is ready for testing
–N.O. Jensen
–G.C. Larsen
–FUGA
–EllipSys
–DWM