GCH versus Empirical Gauss - mirror wakes in vertical plane

[MarkJamesSpring]

I have run the example: 003_tilt_driven_vertical_wake_deflection.py with GCH and empirical wake models

1. the original code implements the Empirical Gauss wake model, which demonstrates vertical deflection downwind of a tilted WTG (on a heeling floating platform for instance)
2. if the Gauss Curl Hybrid wake model is used, there is no vertical deflection, the wakes interact more significantly downstream and velocity deficits are greater - power diminishes along the row when the wind is aligned with the row
3. if the TurbOPark model is implemented the reduction in power along the row in a streamwise direction is greater
4. if the TurbOPark model is implemented with wake mirroring, there is an erroneous ghost wake above the blade tip - the power is further diminished in the streamwise direction

the mirror plane ought to be the ground (z=0m) but appears to be the blade tip (z={hub height + ½ · rotor diameter}m)

what parameter do I need to set to recalibrate to 0m?

[misi9170]

Hi @MarkJamesSpring ,

I didn't totally follow the titles of the some of the figures in your post, but I've tried running example examples_visualizations/002_visualize_cut_plane.py using gch.yaml (the configuration that is in the example on github), and then replacing gch.yaml with emgauss.yaml and turboparkgauss.yaml, and I get the following figures:

Using gch.yaml (as in the example):

Using emgauss.yaml:

Using turboparkgauss.yaml:

Using turboparkgauss.yaml but with the include_mirror_wakes flag set to false:

There certainly seems to be a bug for the turboparkgauss case where include_mirror_wakes is true (which is the default). I'll look into that and see if I can track it down. (It seems that the inclusion of the mirror wakes shifts everything up, including the "actual" hub heights---I'm guessing this is limited to the visualization solver, but I'll need to check and confirm.)

Is that what you were asking about? I didn't totally follow your question regarding tilt.

Cheers,
Misha

[MarkJamesSpring]

thanks so much for writing back so quickly

you have managed to replicate the strange results I have been getting - thanks!

my guess is that the second wake pattern above the blade tip ought to have been derived at ground level (shifted down by hub height + ½ · rotor diameter) before superposition on the velocity field resulting directly downstream of the rotor

in addition to these ghost wakes, the power deficit is extreme - this is indicated by the bar charts: recovery is shown along the row of WTGs when both the empirical Gauss and Gauss Curl Hybrid models are used (but only without wake mirroring in the case of GCH) - however in contrast when the mirror wake (ghost wake) is enabled, the power continues to decrease along the WTG row in a downstream direction

the point I made about tilt is an observation not a question: with GCH, when the tilt is increased there is no vertical deflection of the wake

this is not surprising since FLORIS flags a warning to this effect:-
"floris.logging_manager.LoggingManager WARNING The current model does not account for vertical wake deflection due to tilt. Corrections to power and thrust coefficient can be included, but no vertical wake deflection will occur."

[misi9170]

Ok, that makes sense regarding tilt, thanks!

I've taken a quick pass and my first test indicates that this "shifting" of the wakes only occurs in the visualization solve, not the "usual" turbine solve. I'm not totally sure what you mean by "only without wake mirroring in the case of GCH"---the GCH model does not have a mirror wakes option (it raises an error if include_mirror_wake is added the the input file).

EDIT: after implementing a bug fix #1206, I realized that the mirror wake bug does affect the normal turbine solve, but the differences are very small so not likely to change results significantly.

If you mean that there is a strong wake deficit observed from the TurbOPark model (compared to the GCH and Empirical Gauss models), yes, this is a known result for the TurbOPark model with the default wake parameter of $A = 0.04$. You will find that changing the value of $A$ alters this deficit significantly. I'm would also expect that the deficit is "worsens" when the mirror wakes are included, since this is effectively adding back some extra wake loss that would otherwise "pass through the ground (or sea) surface".

I'm working on a fix for the visualization solve and will let you know when it's done, hopefully within the next week or so.

Misha

[MarkJamesSpring]

yes - you are correct - sorry!

mirror switch is available only with TurbOPark Gauss model

thanks for taking an action to work on the visualisation solve ambiguity

which result for the power deficit is likely to be closest to reality?

1. Empirical Gauss
2. Gauss Curl Hybrid
3. Gauss Curl Hybrid TurbOPark without mirror
4. Gauss Curl Hybrid TurbOPark with mirror
   what value of A is most physically representative of real-world wakes?

[misi9170]

That's the million dollar question! :)

Different wake models perform better or worse in different conditions, and most or all models will need to be calibrated to best represent the local circumstances of a wind farm. With that all being said, if you are just looking for something "reasonable" for a smallish (say, fewer than 50 turbines), onshore wind farm, I would start with the GCH model.

The Empirical Gauss model will produce similar results to GCH, but is specifically designed to be calibrated to observations (hence the "Empirical" part of the name).

If you are looking at large, offshore farms, you might have better luck with the Cumulative Curl or TurbOPark model. Bay et al. (2023) compare GCH to the Cumulative Curl model---their work might give you some further insights.

If you do have observations you'd like to calibrate to, I'd recommend checking out FLASC (and probably using the Empirical Gauss model), where we've built some automated model calibration tools.

Misha

[misi9170]

I've now opened a bug fix pull request here: #1206

[MarkJamesSpring]

thanks @misi9170

the modifications to the geometry you've described in the bug fix report all make sense to me

when I switch from using the GCH wake model with TurbOPark Gauss to the empirical Gauss wake, there is quite a jump in AEP estimates, as illustrated for this candidate WF layout, in which the AEP reductions are indicated [%] going from EmGauss -> GCH (lower numerically)

[misi9170]

Hi @MarkJamesSpring , I don't exactly follow which two models you are comparing, but if you are comparing the TurboParkGauss ("turboparkgauss") model to either the Empirical Gauss ("empirical_gauss") or GCH ("gauss"with the secondary effects flags set toTrue`), then I would expect the AEP to be lower with TurboParkGauss than the other two (because TurboParkGauss in the default configuration has a strong wake deficit compared to Empirical Gauss and GCH).