Is the AEP discrepancy mainly due to probability weighting (WEIS vs. WISDEM)?

[EVElove444]

Subject: Is the AEP discrepancy mainly due to probability weighting (WEIS vs. WISDEM)?

Hi team,

With identical design variables/constraints/objectives, I see large AEP/LCOE differences between WISDEM and WEIS.
I know time-domain vs. steady-state modeling differs; my question is only about weighting:

WISDEM computes AEP by integrating the steady-state power curve against a Weibull PDF (per the RotorSE AEP docs). Could you confirm?
wisdem.readthedocs.io

WEIS (OpenFAST AEP/DLC) appears to compute AEP as
AEP

are the probability weights stored in pCrunch (Crunch.prob). pCrunch also supports user-specified probability distributions. Could you confirm this is the default mechanism?
GitHub
+1

If my WEIS AEP DLC does not explicitly pass Weibull/sector probabilities, does WEIS fall back to non-Weibull (e.g., equal or template) weights, thereby explaining the discrepancy? What is the recommended way to pass the same Weibull (k, A) used in WISDEM to pCrunch/WEIS so both pipelines use identical probability weighting?

Thanks!
​

[dzalkind]

Hi,

Your understanding is correct.

You posted an issue specific to the NREL-5MW on one of our forums/pages. Can you link to that for further details that will help us debug? I think you are correct that something is amiss in the WEIS AEP calculation.

Here is where AEP is calculated. The wind speed distribution parameters should be consistent between WISDEM and WEIS.

One potential issue is that this factor might result in an AEP in kW*s units rather than the more standard kW*h, which WEIS also specifies.

What is the difference you are seeing between WISDEM and WEIS currently?

Best, Dan

[gbarter]

Thank you for raising this issue and quickly identifying the root cause. Hopefully this is resolved in this PR.

[dzalkind]

Thanks @gbarter !

[EVElove444]

@gbarter @dzalkind I’m aligning AEP calculations between WISDEM and WEIS and have a couple of targeted questions:

I’ve already corrected an AEP-units detail in my local WEIS setup so that reported AEP is in kWh/MWh as intended.

WISDEM Weibull parameter source
In my WindIO model file (e.g., NREL-5MW), I provide environment.weib_shape_parameter: 2. (i.e., Rayleigh). Can you confirm WISDEM’s AEP module directly reads this WindIO field and uses it to build the wind-speed distribution for integrating the power curve? (WISDEM docs describe AEP as the power curve integrated against the wind-speed distribution, and WindIO lists weib_shape_parameter under environment.)

WEIS default AEP probabilities and how to set Weibull
In WEIS’s AEP workflow (OpenFAST DLC “AEP” + pCrunch post-processing), what is the default for the occurrence weights 𝑝𝑖 across the wind-speed bins if I do not specify any probabilities—are they uniform by default?
Also, what is the recommended way to pass a Weibull weighting (e.g., Rayleigh with 𝑘=2) to WEIS?

Is there a supported key in modeling_options.yaml under the DLC_driver → "AEP" block to provide a probabilities array matching my wind_speed bins, or a way to provide Weibull parameters that WEIS converts to bin probabilities internally?

Alternatively, should I compute bin probabilities externally from k/scale (or mean wind) and pass them directly?
(For context: WEIS uses pCrunch to compute AEP from OpenFAST runs with user-defined probabilities; pCrunch itself supports user probabilities.)

If there’s a canonical example YAML showing how to (a) set uniform vs. Weibull probabilities for the AEP DLC in WEIS, or (b) map WindIO’s weib_shape_parameter into WEIS/pCrunch automatically, I’d really appreciate a pointer.

[dzalkind]

Hi,

In my WindIO model file (e.g., NREL-5MW), I provide environment.weib_shape_parameter: 2. (i.e., Rayleigh). Can you confirm WISDEM’s AEP module directly reads this WindIO field and uses it to build the wind-speed distribution for integrating the power curve? (WISDEM docs describe AEP as the power curve integrated against the wind-speed distribution, and WindIO lists weib_shape_parameter under environment.)

I can confirm this, but you can check the distribution here: https://github.com/WISDEM/WISDEM/blob/master/wisdem/rotorse/rotor_power.py#L1000

Here is where AEP is computed in WEIS:

WEIS/weis/aeroelasticse/openmdao_openfast.py

Line 2618 in e497f9b

AEP, _ = self.cruncher.compute_aep("GenPwr", idx=idx_pwrcrv)

There is not currently an option for setting a uniform distribution, but you could adapt the distribution here:

WEIS/weis/aeroelasticse/openmdao_openfast.py

Line 2612 in e497f9b

self.cruncher.set_probability_turbine_class(U, discrete_inputs['turbine_class'], idx=idx_pwrcrv)

Alternatively, you could alter the source code to use the second output from the pCrunch compute_aep function: https://github.com/NREL/pCrunch/blob/d586bf09148ddc7188b42b4509da2063b49ddca4/pCrunch/crunch.py#L497

I hope this helps.

Best, Dan

[EVElove444]

@gbarter @dzalkind I’m using WEIS with OpenFAST enabled to evaluate Annual Energy Production (AEP) cases and then optimize Levelized Cost of Energy (LCOE). I would greatly appreciate pointers to any official theory notes, equations, or documentation that formalize this workflow in WEIS—especially when OpenFAST is the aero-servo-elastic solver.

AEP computation with OpenFAST

The definition of the AEP “DLC/AEP” case set (wind-speed bins, yaw misalignment handling, turbulence class/TI usage, seeding strategy).

The probability weighting and loss factors applied (availability, array, soiling), and the exact aggregation formula used to convert time-domain power to annual energy (i.e., the probability-weighted energy/time ratio multiplied by seconds per year).

Any recommended settings for simulation length, transient trimming, and convergence for robust AEP estimation.

Blade mass / structural properties

Equations/models used to compute blade mass theory.

Implementation references

Documentation pages, , or code references (module/function names) that describe the AEP function, the OpenFAST AEP driver, and the LCOE objective.

Any published papers or validation notes that use this exact WEIS+OpenFAST AEP-to-LCOE chain.

For context, I’m optimizing twist, chord , and spar-cap thickness at several spanwise stations, with AEP computed from OpenFAST time-domain outputs and constraints on strains/tip deflection.

[gbarter]

WEIS is an open source project, so all of the documentation that we have is posted online already. We are working to expand the documentation in the coming months, but we don't have any text to copy-and-paste here that isn't already online. Some of the questions you ask have answers in documentation of WEIS components, such as WISDEM, or fundamental libraries, such as OpenMDAO.

For specifics, I can point you to:

• AEP estimation with DLC 1.1: This is not a WEIS question. I suggest you do some web research on IEC Design Load Cases and specifically DLC 1.1. I understand many of the official IEC reports are paywalled, but there are older versions and summaries available that would suffice. The WEIS DLC driver has enough customization features to select alternate wind speed binning, number of seeds, etc.
• In WEIS, the AEP is calculated by pCrunch. See code here and example here.
• Recommendations for best practices can be problem dependent. Most simulations might default to 10min of simulation time with 2min of transient time, but it is up to the user to make final determination.
• For blade mass calculation and stiffness, see the WISDEM notes on PreComp.
• For LCOE, see the WISDEM module, PlantFinanceSE.

For your specific workflow optimizing blade parameters to maximize AEP or minimize LCOE, we would recommend starting with a steady-state optimization in WISDEM, which will converge much more quickly. WISDEM provides good examples to follow in this regard. One you have a final design in WISDEM, then bring the solution to a higher level of fidelity with optimization around the OpenFAST simulations.

[gbarter]

One more comment- The DLCs are meant for performance and loads certification and standards. They were never devised to be in the inner loop of an optimization. Giving the optimizer a strong signal requires computational efficiency and minimal noise or stochasticity. Meaning, blade design under a full DLC 1.1 with all the seeds and wind speed bins is probably computational overkill and will be a challenge to converge.