Date: 2:30pm, March 8, 2016
Location: EE Conference Room, Mudd
Speaker: Andrey Bernstein and Niek J. Bouman, postdoctoral researchers at EPFL
Abstract: The Commelec framework:
We design an agent-based framework for real-time and explicit control of electrical grids with a heterogeneous mix of intermittent and uncertain generation (like PV), storage capacity and loads. The method is scalable to grids of any size, and solves the problems of quality of service and energy balance without major investment. Our approach avoids solving stochastic mixed-integer optimization at each time step. Instead, the control method is based on projected gradient descent. The framework provides a "Grid Operating System" that allows device controllers for intelligent buildings, e-car charging systems, etc. to be easily connected and provide real time support to the grid.
Design of Commelec Resource Agents with Guaranteed Tracking Properties:
In a basic setup of Commelec, one centralized 'grid agent' (GA) controls a number of resources, where each resource is either a load, a generator, or a combination thereof, like a battery. The GA periodically computes new power setpoints for the resources based on the estimated state of the grid and an overall objective, and subject to safety constraints. Each resource is augmented with a resource agent (RA), which takes care of communicating with the grid agent and implements the power-setpoint requests sent by the GA on the resource.
In this talk, we focus on the resource agents (including the resources that they manage) and their impact on the overall system's behavior. Intuitively, for the system to converge to the overall objective, the RAs should be obedient to the requests from the GA, in the sense that the actually implemented setpoint should be close to the requested setpoint, at least on average.
We propose a formal notion of obedience called $c$-bounded accumulated-error for some constant $c >= 0$. We then demonstrate the usefulness of our notion, by presenting theoretical results (for a simplified scenario) as well as some simulation results (for a more realistic setting) that indicate that, if all resource agents in the system have $c$-bounded accumulated-error, the closed-loop system converges on average to the objective.
Also, we show resource agents that (provably) have $c$-bounded accumulated-error for resources with uncertainty (e.g., PV panels) and for resources with a discrete set of implementable setpoints (e.g., heating systems with heaters that each can either be switched on or off).
Biography: Andrey Bernstein received the B.Sc. and M.Sc. degrees in Electrical Engineering from the Technion -- Israel Institute of Technology in 2002 and 2007 respectively, both summa cum laude. He received the Ph.D. degree in Electrical Engineering from the Technion in 2013. Between 2010 and 2011, he was a visiting researcher at Columbia University. During 2011-2012, he was a visiting Assistant Professor at the Stony Brook University. He is currently a postdoctoral researcher at the Laboratory for Communications and Applications of Ecole Polytechnique Federale de Lausanne (EPFL), Switzerland. His research interests are in the decision and control problems in complex environments and related machine learning methods, with particular application to intelligent power and energy systems. His current focus is on the real-time control of electrical grids.
Niek J. Bouman was born in Son en Breugel, the Netherlands, on September 2, 1983. He received the B.Sc. degree and M.Sc. degree in Electrical Engineering from Universiteit Twente, Enschede, NL, in 2005 and 2007, respectively, and the Ph.D. degree in Mathematics from Universiteit Leiden, Leiden, NL, in 2012. From 2008 to 2012, he was with the cryptology group of Centrum Wiskunde & Informatica (CWI), Amsterdam, NL. The topic of his Ph.D. thesis is quantum cryptography and (quantum) information theory. During 2013-2014, he was an independent research consultant for a crypto-currency-analysis project. In 2014 he joined the Laboratory for Communications and Applications of EPFL, Lausanne, Switzerland as a postdoctoral researcher. His research interests include information and communication theory, quantum cryptography, statistical inference, as well as applied research related to grid control and the Internet of Things.
Joint EE and SCMD seminar, hosted by Prof. Gil Zussman