BEGIN:VCALENDAR
VERSION:2.0
PRODID:Linklings LLC
BEGIN:VTIMEZONE
TZID:Europe/Stockholm
X-LIC-LOCATION:Europe/Stockholm
BEGIN:DAYLIGHT
TZOFFSETFROM:+0100
TZOFFSETTO:+0200
TZNAME:CEST
DTSTART:19700308T020000
RRULE:FREQ=YEARLY;BYMONTH=3;BYDAY=-1SU
END:DAYLIGHT
BEGIN:STANDARD
TZOFFSETFROM:+0200
TZOFFSETTO:+0100
TZNAME:CET
DTSTART:19701101T020000
RRULE:FREQ=YEARLY;BYMONTH=10;BYDAY=-1SU
END:STANDARD
END:VTIMEZONE
BEGIN:VEVENT
DTSTAMP:20210916T132455Z
LOCATION:Ernesto Bertarelli
DTSTART;TZID=Europe/Stockholm:20210709T153000
DTEND;TZID=Europe/Stockholm:20210709T160000
UID:submissions.pasc-conference.org_PASC21_sess151_msa305@linklings.com
SUMMARY:Predicting Wakes Behind Buildings: A Machine Learning Approach for
  Extracting Physics Informed Low-Order Models from Highly Resolved Flow-Fi
 eld Datasets
DESCRIPTION:Minisymposium\n\nPredicting Wakes Behind Buildings: A Machine 
 Learning Approach for Extracting Physics Informed Low-Order Models from Hi
 ghly Resolved Flow-Field Datasets\n\nFytanidis, Maulik, Balakrishnan, Kota
 marthi\n\nResolved simulations of the atmospheric boundary layer (ABL) pro
 vide valuable information for optimizing the siting of wind turbines in a 
 “distributed wind” scenario, where wind turbines located amids
 t buildings and vegetation, in an urban/semi-urban/rural setting, supply e
 lectricity, directly to the consumer. Such computationally intensive simul
 ations, for estimating the velocity deficit in the leeward side of buildin
 gs, are impractical for rapid/real-time estimation of the wind energy pote
 ntial. Thus, there is a need for alternative low-order/analytical wake mod
 els, that can be built into a phone app to provide quick estimates of the 
 wind velocity characteristics in the wake of a building. In our work, LES 
 and uRANS simulations were performed using the highly scalable spectral el
 ement-based solver, Nek5000, to build a flow database, for buildings of di
 fferent aspect-ratios, with various wind angles of attack, and for inflow 
 conditions with varying mean shear. The database was then used to generali
 ze the formulation, and calibration of wake models. The generalization was
  achieved through the construction of a physics-informed machine learning 
 framework which was interfaced with the analytical wake models. The simula
 ted flow fields were utilized to train neural networks, which then predict
 ed model form parameters as a function of geometric information and spatia
 l location. Constraints were embedded into the neural architecture to ensu
 re the preservation of desirable properties for the flow profile predictio
 ns. In this talk, we present the implementation of these generalized low-o
 rder models in TensorFlow, and demonstrate its ability to make flow field 
 predictions, at a fraction of the computational cost.\n\nDomain: CS and Ma
 th, Emerging Applications, Climate and Weather, Physics, Engineering
END:VEVENT
END:VCALENDAR