BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:IEM special seminar series in energy storage systems: Online Batte
 ry Control in Active Distribution Grids Using Lyapunov Optimization and Re
 inforcement Learning
DTSTART:20220322T100000
DTEND:20220322T110000
DTSTAMP:20260429T060606Z
UID:86c12414d269effc60cd2deb40fa4c8f54df85b4212495b74f01308a
CATEGORIES:Conferences - Seminars
DESCRIPTION:Eleni Stai\, (Post-​doctoral Researcher\, ETHZ\, Zurich\, Sw
 itzerland)\nAbstract: Energy storage in the form of batteries will likely 
 play a crucial role in the operation of distribution grids that are charac
 terized by a continuously growing integration of uncertain distributed ene
 rgy resources. In this talk\, we present computationally efficient real-ti
 me control schemes for batteries in active distribution grids with lookahe
 ad state-of-energy constraints. The goal is to follow a previously compute
 d dispatch plan or to optimize a monetary cost from buying and selling pow
 er at the point of common coupling. However\, the lookahead constraints re
 nder the battery decisions non-trivial. The common practice in literature 
 to solve this problem is Model Predictive Control (MPC)\, which does not s
 cale for large grids.\nWe first propose iterative Lyapunov Real-time Contr
 ol (iLypRC)\, a fast online algorithm\, which does not need forecasts. ILy
 pRC is designed via Lyapunov optimization and requires only bounds on the 
 uncertain quantities for each real-time interval. It efficiently accounts 
 for grid losses\, battery efficiency and grid constraints via iterative li
 nearizations of the power flow equations. We compute a theoretical upper b
 ound on the difference between the cost of iLypRC with the cost of an orac
 le. We show via numerical examples that iLypRC achieves a cost very close 
 to that of MPC with good forecast\, but iLypRC needs no forecast and has m
 uch lower run time complexity. In addition\, when the MPC forecast is inac
 curate\, iLypRC outperforms MPC.\nSecond\, we investigate a reinforcement 
 learning approach based on the Deep Deterministic Policy Gradient (DDPG) a
 lgorithm. To satisfy the lookahead battery constraints we adapt the experi
 ence replay technique used in DDPG. To guarantee the satisfaction of the h
 ard grid constraints\, we introduce a safety layer that performs constrain
 ed optimization. We show that it can achieve costs close to MPC and Lyapun
 ov\, while reducing the computational time by multiple orders of magnitude
 .\n\nBio: Eleni Stai is a Postdoctoral researcher with the Power Systems L
 aboratory at ETHZ. She received the Diploma in Electrical and Computer Eng
 ineering from the National Technical University of Athens (NTUA)\, Greece\
 , in 2009\, the B.Sc. in Mathematics from the National and Kapodistrian Un
 iversity of Athens\, Greece\, in 2013\, the M.Sc. in Applied Mathematical 
 Sciences from NTUA in 2014 and the Ph.D. in Electrical Engineering from NT
 UA in 2015. From 2016 to 2020\, she was a Postdoctoral researcher with the
  Laboratory for Communications & Applications 2 at EPFL. She has received 
 the Chorafas Foundation Best Ph.D. Thesis award\, the Thomaidis Foundation
  Best M.Sc. Thesis award and a Best Paper Award in ICT 2016. Her main rese
 arch interests include battery management and control in active distributi
 on grids\, smart-grid control and applications\, network design and optimi
 zation. She has co-authored the book “Evolutionary Dynamics of Complex C
 ommunications Networks” (Taylor and Francis Group\, CRC Press).\n 
LOCATION:BM 5202 https://plan.epfl.ch/?room==BM%205202 https://epfl.zoom.u
 s/j/63474947803
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR