Matching FAST.Farm vs FLORIS for inline array (5D, 0.6D offset) at TI≈6%: downstream rotor-avg deficit too high in FAST.Farm?

[fbelli997]

Dear all,

We are building a model-based controller and want to reduce mismatch between FAST.Farm (FF) and FLORIS. Case: 4 floating turbines inline at 5D streamwise spacing with a 0.6D lateral offset per turbine, aligned yaw (greedy). Hub-height TI ≈ 6%. No wake steering.
Fast farm is set using the curl model and is set as suggested in the manual and the results obtained are presented in the table here below:

](http://www.w3.org/TR/REC-html40%22%3E) <style> </style>

Turbine	Power (kW)	WindDiskVelX (m/s)	Blade pitch (deg)
T1	9349	8.98	0.156
T2	4304.8	6.82	0.39
T3	4815	7.13	0.255
T4	5045.5	7.22	0.241

As I mentioned before, there are 4 wind turbines in-line (5D) with an offset of 0.6D for each turbine.
we are using default parameter of WAT.

Questions:

1. Reasonbleness of FF deficit:
Do these numbers look plausible to you? are they reasonable? Specifically, do you know of LES studies where the rotor-averaged velocity deficit for a second inline turbine at ~5D (TI ≈ 6%, slight 0.6D lateral offset) is around 0.25–0.30? Our T2 rotor-avg deficit is ~0.28, giving ~half the upstream power.

2. From FLORIS point of view:
The procedure we are using for building model is using look up table of power and thrust coming from 11000 seconds of simulation from FASTfarm at 6% of Turbulence intensity (exactly the same we are using in the farm simulation) and each wind speed from 3 to 14 m/s to be sure the power calculated in FLORIS is coherent with the one in FASTfarm. what is weird for us is that the results obtained in FLORIS are much higher than those obtained in FASTfarm. Also tuning FLORIS parameters decreasing at the maximum the wake expansion parameters, is not possible to match the power of the downstream wind turbine, neither the second one, which should be the easiest.

results from Floris with the default wake parameters using cumulative curl wake model

<style> </style>

	T1	T2	T3	T4
power	9.218	5.843	6.441	6.158	MW
wind speed	8.974	7.661	7.94	7.808	m/s

Trying to get close in the results, with all the wake models calibtrated to not expand the wakes, I can get closer in the power results losing physical behaviour in the wake field as shown here in the picture:

but still is not possible to reach FATfarm results since also with "tuned" parameters the results are:

<style> </style>

	T1	T2	T3	T4
power	9.218	5.442	5.931	5.893	MW
wind speed	8.973	7.475	7.703	7.684	m/s

[misi9170]

Hi @fbelli997 ,

On the FAST.Farm side, a relative velocity deficit of ~0.28 (and a power loss of 50%) seems high but not impossibly high. I can't point you directly to an LES paper to confirm that, but for me at least, that doesn't seem outside of the realm of possibility. Of course, the velocity deficit depends strongly on the spacing between turbines.

It looks like you're using IEA 15MW turbines with 240m rotor diameter and 1200m (5D) distance between T1 and T2. Is that correct? For some reason, the FF image you've provided makes the turbines look quite closely spaced (more so than the FLORIS image), but perhaps that is an illusion. Clearly, closer turbine spacing will lead to larger losses.

Regarding FLORIS: are you asking how to tune FLORIS to match the FAST.Farm wind speed and power predictions? i.e., you're wondering about how to make the wakes "deeper" in FLORIS? As you're doing, you can try tuning the wake parameters to achieve this. You might have better luck with other wake models, such as the Gauss-Curl-Hybrid or Empirical Gaussian models (in particular, the Empirical Gaussian model will let you directly control the wake expansion downstream of the turbine).

It's also worth thinking about FLORIS as representing time-averaged wakes, rather than "instantaneous" wakes. In particular, wake models in FLORIS will include the average effect of wake meandering (which appears as a widening of the Gaussian wake shape). My understanding is that FAST.Farm, on the other hand, is more representative of instantaneous wake locations. To compare them directly, you'd perhaps be better to run FAST.Farm with some wake meandering; and then look at the time-averaged output from FAST.Farm. Is that what you're doing here?

Misha

[fbelli997]

Hi @misi9170,

Yes I confirm you that we are using 4 IEA15MW wind turbines spaced of 5D.
Exactly, I'm trying to match FASTfarm and FLORIS to make the FLORIS controller more accurate and giving FASTfarm better setpoints during the FASTfarm simulation.

I tried every wake model in floris, and with all of them I tried to reduce the wake expansion parameters to increase the deficit but it seems the results saturate at higher power values compared to those of FAST farm.
I send you some results using GCH here below in the picture by using wake expansion values around 0:

Here, power values are still higher:

T1 = 9.218 MW
T2 = 5.359 MW
T3 = 5.510 MW
T4 = 5.569 MW

These are results with empirical gauss:

and still powers are higher:
T1 = 9.218 MW
T2 = 5.988 MW
T3 = 6.548 MW
T4 = 6.509 MW

To solve, what I did was changing the two parameters inside the Gaussian wake model in the gaussian defict. The "r" and "n" parameters of the exponential as in this two following lines:

near_wake_deficit = gaussian_function(C, r, 1.12, np.sqrt(0.85))
# near_wake_deficit = gaussian_function(C, r, 1, np.sqrt(0.5))
 far_wake_deficit = gaussian_function(C, r, 1.12, np.sqrt(0.85))
# far_wake_deficit = gaussian_function(C, r, 1, np.sqrt(0.5))

In this way I was able to match pretty well the results in this layout configuration cause the deficit got stronger in the wake field slightly tuning the other parameters in the .yaml file. The problem is that now the results of FLORIS are not reliable when turbines are completely aligned (I haven' t tried to simulate this configuration on FASTfarm but downstream powers turbines now look pretty low in aligned conditions with this parameters).

is it possible that FLORIS is accurate when turbines are aligned but is not when the turbines are misaligned?
Do you think what i did is reasonable?

thank you very much for your feedbak.

Federico

[misi9170]

Hi Federico,

Yes, it's certainly possible that FLORIS works better in aligned cases than slightly misaligned cases. One thought that comes to mind is that FLORIS (normally) represents the rotor using a rectangular grid of points across the turbine rotor (see also here). It is possible that in misaligned conditions, where some points may be inside of the wake of the upstream turbine and some may be outside, that the "average" inflow to the turbine is too high.

You could see if specifying the solver type as a turbine_cubature_grid on the FLORIS input file (see here for an example using the TurbOPark model). I'm not sure how this will change your results, but the cubature grid uses a radial pattern of grid points, rather than a square pattern, which hopefully will improve the way that the average wind speed is evaluated over the rotor.

Best,
Misha

[fbelli997]

Hi Misha,
I wanted to try the cubature grid but I'm using third version of FLORIS...I think it doesn't work in that version, do you confirm or I'm doing something wrong?

Thank you for your help,

Federico

[misi9170]

Hi Federico, the cubature grid was actually introduced in FLORIS v3.4, so you may be in luck. You could take a look at #649 for details