Zoom Logo

How Wind Energy Research can make the most of advances in High Performance Computing (HPC) - Shared screen with speaker view
Julien Balsen
04:05
Good morning everyone ! Sorry being late 🙂 technical problems ahah
JAKE BADGER
15:57
Questions can be added here in the chat.
Maria Ramalho
18:02
Several EU wind researchers are waiting for the Exascale codes being developed in the US. Do you think this is a good enough strategy? or should EU try to develop its own codes?
Luis Baquero
19:02
Hi. Are the slides of the presentations going to be shared?
Maria Ramalho
19:25
There are two DOE funded projects to push wind energy towards exascale, A2E and Exawind, both directed by Michael Sprague, with a total budget of 6 Million US$/per year.How does this compare to the situation in Europe?
JAKE BADGER
20:09
Thanks for questions Maria. I'll ask them on your behalf. Luis - we'll ask for that with permission of the authors.
Maria Ramalho
20:34
ok
Mikel Iribas
34:00
Michelle, have you faced to any simulation case that could not be done to computing limitation??
Maria Ramalho
38:29
For more on EoCoE, the Energy-oriented Centre of Excellence see www.eocoe.eu
Maria Ramalho
43:32
The project EoCoE is there … together with the network build through the EERA Transversal Joint Programming … to assist in code porting to Exascale using the network of experts available in different HPC centres. This directly addresses the issue stated by Michele on the lack of personnel to tackle Exascale Challenges. So we are not starting from scratch in Europe after all ...
JAKE BADGER
01:04:20
Please indicate the questions you have for Richard in the chat.
Maria Ramalho
01:07:13
How many collaborators are working in dedicated mode to this research using codes for HPC systems?
esteban ferrer
01:07:20
I would like to know what is the effect of using actuator line modelling in the turbulence generation of the turbine wake. Is inflow turb inlcuded in the airfoil tables? Does this have an effect in the overall wind farm simulation?
esteban ferrer
01:10:24
what type of more local subgrid model, can you ellaborate please?
herbert owen
01:17:30
Is the code suitable for GPUs? How do you plan to take advantage of machines like LUMI? Do you think staying on CPUs is a viable alternative?
Hendrik Heißelmann
01:19:07
Very nice overview, Richard. How do you handle different atmospheric stability and large scale effects like Coriolis force in you wind farm simulations?
Maria Ramalho
01:19:19
Thanks Richard for your informative answer
herbert owen
01:21:52
THANKS!! very nice talk!!
Richard Stevens
01:23:36
THANKS!! very nice talk!! —> Thanks a lot!
Richard Stevens
01:26:53
Very nice overview, Richard. How do you handle different atmospheric stability and large scale effects like Coriolis force in you wind farm simulations? —> Thanks a lot! We consider a geostrophic wind to force the simulations and can set the thermal boundary conditions at the bottom and top of the domain. This will allow the atmospheric flow to develop naturally. For details see: https://doi.org/10.1007/s10546-020-00570-5
Hendrik Heißelmann
01:27:25
Thx RIchard.
Richard Stevens
01:32:31
Is the code suitable for GPUs? How do you plan to take advantage of machines like LUMI? Do you think staying on CPUs is a viable alternative? —> At the moment this code does not use GPU yet. We did already explore GPU for DNS I showed, see https://doi.org/10.1016/j.cpc.2018.03.026. There the GPU code is used on PiZ-Daint supercomputer in Switzerland. When there is shared nodes CPU/GPU one of the main challenges is to ensure that communication is performed at the same time as calculations, so trying to perform computations in a batched way and send over data when part of the calculations are done.
Richard Stevens
01:34:05
what type of more local subgrid model, can you ellaborate please? --> Anisotropic Minimum Dissipation model, for details see https://doi.org/10.1007/s10546-020-00570-5. The main advantage is even the simplification of the communication patterns that better fit what the remainder of the code uses, compared to the previous subgrid model (lagragian Scale Dependent model) we used
esteban ferrer
01:37:07
thank you Richard, very nice talk!
Richard Stevens
01:37:50
I would like to know what is the effect of using actuator line modelling in the turbulence generation of the turbine wake. Is inflow turb inlcuded in the airfoil tables? Does this have an effect in the overall wind farm simulation? —> The inflow is a fully developed atmospheric boundary layer, using concurrent precursor method (http://dx.doi.org/10.1016/j.renene.2014.01.024). The inflow turbulence has a very large effect on the wind farm performance. Essentially higher turbulence leads to more mixing in the atmosphere, which helps the wind turbine wakes to recover faster
esteban ferrer
01:41:01
thanks!
JAKE BADGER
01:43:29
Please indicate the questions you have for Stefan in the chat.
esteban ferrer
01:47:31
thank you Stefan, nice talk. You mentioned wake control using ML, do you mean reinforcement learning? if yes what is the reward policy like?
herbert owen
01:49:38
What about mesh anisotropy (factor 1:10 in LES) for complex terrain? Or do you think isotropic grids would be fine?
esteban ferrer
01:51:13
thank you Henry!
herbert owen
01:53:34
Thanks!!!
JAKE BADGER
02:18:24
Please indicate the questions you have for Mikel in the chat.