isope topfarm
DESCRIPTION
http://windenergyresearch.org/?p=2836A wind farm layout optimization framework based on a multi-fidelity model approach is applied to the offshore test case of Middelgrunden, Denmark. While aesthetic considerations have heavily influenced the famous curved design of Middelgrunden wind farm, this work focuses on testing a method that optimizes the profit of offshore wind farms based on a balance of the energy production, the electrical grid costs, the foundations cost, and the wake induced lifetime fatigue of different wind turbine components. The multi-fidelity concept uses cost function models of increasing complexity (and decreasing speed) to accelerate the convergence to an optimum solution. In the EU TopFarm project, three levels of complexity are considered. The first level uses a simple stationary wind farm wake model to estimate the Annual Energy Production AEP, a foundations cost function based on the water depth, and an electrical grid cost function. The second level calculates the AEP and adds a wake induced fatigue cost function based on the interpolation of a database of simulations done for various wind speed and wake setups with the aero-elastic code HAWC2 and the Dynamic Wake Meandering (DWM) model. The third level includes directly the HAWC2 and DWM models in the optimization loop in order to estimate both the fatigue costs and the AEP.The novelty of this work is the implementation of the multi-fidelity, the inclusion of the fatigue costs in the optimization framework, and its application on an existing offshore wind farm test case.Paper presented at ISOPE 19-14 June 2011, Maui, Hawaii, USAP.-E. Réthoré, P. Fuglsang, G.C. Larsen, T. Buhl, T.J. Larsen, H.A. Madsen. TopFarm: Multi-fidelity Optimization of Offshore Wind FarmTRANSCRIPT
TOPFARM: Multi-fidelity Optimization of Offshore Wind Farm
Pierre-Elouan Réthoré, Peter Fuglsang,
Gunner C. Larsen, Thomas Buhl,
Torben J. Larsen and Helge A. Madsen
Wind Energy Division
Risø DTU - National Laboratory for Sustainable Energy
Technical University of Denmark
ISOPE – 19-24 June 2011
Outline
• Introduction – Vision and Philosophy• The TOPFARM Modules• The Wake “engine”• The Load and Production “Engine”• Cost Functions ... and Objective Function• Results
o Generic Offshore Wind Farmo Middelgrunden Offshore Wind Farm
• Conclusion• Outlook
Conclusions
• Vision: A “complete” wind farm topology optimization taking into account • Loading- and production aspects in a realistic and
coherent framework• Financial costs (foundation, grid infrastructure, ...)
... and subjected to various constraints (area, spacing,...)
• Philosophy: The optimal wind farm layout reflects the optimal economical performance as seen over the lifetime of the wind farm
Introduction – Vision and Philosophy
ConclusionsThe TOPFARM Modules
Conclusions
• Only variable costs (i.e. costs depending on the wind farm topology) are included in the objective function (OF)
• OF formulated as a financial balance expressing the difference between the wind farm income (power production (WP)) and the wind farm expenses (i.e. O&M expenses (CM), cost of turbine fatigue load degradation (CD), and financial expenses (C) – in this case including grid costs (CG) and foundation costs (CF))
Cost Functions … and Objective Function (1)
LXN
L
icn N
rrCWPFB
11
LXN
L
ic
N
rrCWPFB
11
Gunner Larsens cost model
CGCFC
Foundation cost
Electrical infrastructure cost
Degradation cost
• Maximize Finance balance• Maximize power production• Minimize cost
• Minimize foundation cost
• Minimize cable cost
• Minimize maintenance
• Minimize degradation
Maintenance cost
Cost Functions … and Objective Function (2)
,CMCDWPWPn
Electrical Production Sales
Simple wake model
From: G. C. Larsen, ”A simple stationary semi-analytical wake model. Risø-R-1713
Fast and Simple Wake Model
Conclusions
• The dynamic wake meandering (DWM) model (“poor mans LES”) is an efficient model providing a realistic description of wake affected flow fields within wind farms
• The core of the DWM model is a split in scales in the wake flow field, with large turbulent scales being responsible for stochastic wake meandering (passive tracer analogy), and small scales being responsible for wake attenuation and expansion in the meandering frame of reference as caused by turbulent mixing (… LES sub-scale analogy)
The Wake ”Engine” (1)
Conclusions
DWM main ingredients (2):
• Wake down-stream transportation (meandering) by passive tracer analogy (... LES large-scale analogy)
• A sequence of “deficit-releases” is considered ... and described in space and time
The Wake ”Engine” (2)
Conclusions
• Does it work? ... a full-scale verification (based on LiDAR measurements)
The wake ”engine” (4)
Conclusions
• Dynamic wake meandering (DWM) model (“poor mans LES”) – module1 – combined with aeroelastic simulations of the individual turbines – modules 2, 3
The load and production ”engine” (1)
Conclusions
• Does it work? ... a full-scale verification (based on WT7)
The load and production ”engine” (2)
• Damages costs (CD) are accumulated damage of an element ’s’ (Ds) X the price of the element (Ps).
• Ds is a linear function of the fatigue loads.
• Maintenance Costs are based on the probability of occurrence of a failure multiplied by the replacement cost.
Damages Costs (CD) & Maintenance Costs (CM)
,TN S
SSDPCD
,TN S
SCMCM
,,;,;1
,,1,1,
R
jjSjSSSRSRSSSRS DFPDFPNCM
rr
,Sd
SaS L
LD
• Different models investigated.
• Discontinuous model chosen (green):• 20 % of turbine price• For each meter above 8 meters
an additional cost of 2%• Above 20 meters very
expensive (20% pr m)
• Simple and fast
• Could be refined easily
Foundation Costs (CF)
• Simple model for electrical grid costs:• Finding the shortest
connection between all the turbines.
• Fixed cost per meter: • C=675 €/m
Electrical Grid Costs (CG)
• Genetic Algorithm (SGA)• Structured grid • Slow (Many iterations necessary)• Global optimum
• Gradient Based Search (SLP)• Unstructured grid (good for refinements)• Faster (Few iterations for converging)• Local minimum
• SGA+SLP:• Good combination for finding a refined
global optimum.
Optimization Algorithms
Example of Structured grid
LXN
L
icsf N
rrCpWPFB
111
Multifidelity Approach
Fidelity Level 1st 2nd 3rd
Electricity salesStationary wake +
Power curveHAWC2-DWM
DatabaseHAWC2-DWM
Simulations
Fatigue costs NoHAWC2-DWM
DatabaseHAWC2-DWM
Simulations
Foundation costs Yes Yes Yes
Electrical Grid costs Yes Yes Yes
Optimization algorithm SGA SLP or SGA+SLP SLP
Domain discretization Coarse Fine Fine
Wind speed and direction bin size Coarse Fine Fine
Results – Generic Offshore Wind Farm (1)
0.002 0.004
0.006
30
210
60
240
90270
120
300
150
330
180
0Wind rose
Gray color: Water depth [m]Yellow line: Electrical grid
Wind direction probabilitydensity distribution
• 6 5MW offshore wind turbines.• Water depth between 4 and 20 m
Results – Generic Offshore Wind Farm (2)
Result of a gradient based optimization (SLP)
Results – Generic Offshore Wind Farm (3)
Result of a genetic algorithm + gradient based optimization (Simplex)
Results - Middelgrunden (1)
Photo: Andreas Johansen
Results - Middelgrunden (2)
Allowed wind turbine region
Middelgrunden layout
Results - Middelgrunden (3)
• Ambient wind climate:
Results - Middelgrunden (4)
Optimum wind farm layout (left) and financial balance cost distribution relative to baseline design (right).
Iterations: 1000 SGA + 20 SLP
Results - Middelgrunden (5)
Before
After
Conclusions
• The baseline layout was largely based on visual considerations
• The optimized solution is fundamentally different from the baseline layout ... the resulting layout makes use of the entire feasible domain, and the turbines are not placed in a regular pattern
• The foundation costs have not been increased, because the turbines have been placed at shallow water
• The major changes involve energy production and electrical grid costs ... both were increased
• A total improvement of the financial balance of 2.1 M€ was achieved compared to the baseline layout
Results - Middelgrunden (4)
Conclusions
• A new optimization platform has been developed that allow for wind farm topology optimization in the sense that the optimal economical performance, as seen over the lifetime of the wind farm, is achieved
• This is done by:o Taking into account both loading (i.e. degradation),
operational costs (i.e. O&M) and production of the individual turbines in the wind farm in a realistic and coherent framework .... and by
o Including financial costs (foundation, grid infrastructure, etc.) in the optimization problem
• The performance of the platform has been demonstrated for offshore and on-shore example sites
Conclusion(s)
Conclusions
• (Possible) future model extensions:o Inclusion of atmospheric stability effects (… important for
production; cf. L. Jensen & R. Barthelmie)o Full DWM based optimization (presently only the closest
upstream turbine is considered load generating) ... o Better foundation cost modelo Parallization of the code … improved computational speedo Aeroelastic computations in the frequency domain …
improved computational speedo Cheapest rather than shortest cabling between turbineso Inclusion of extreme load aspects o Eddy viscosity model consistent with the DWM split in
scales … improved description of the wake deficit
Outlook