Events Calendar

Online Optimization under Uncertainty: Intelligence in the smart grid and a connection to Model Predictive Control

Speaker: Murali Narayanaswamy, IBM Research, India

Talk Title: Online Optimization under Uncertainty: Intelligence in the smart grid and a connection to Model Predictive Control

Abstract: Incorporating large quantities of intermittent renewable power into the grid highlights the need for intelligent scheduling of generation, loads and storage. Recent advances in solar and wind power prediction offer hope that a reduction in the uncertainty of renewable availability will lead to an increase in its value. However, incorporating these predictions effectively turns out to be a non-trivial problem.

In this talk we show how to model these (and many other) problems as a Markov Decision Process with short term predictions of (or lookahead into) future rewards. In each time step more information is revealed to the algorithm as predictions are updated, leading to what we call dynamic uncertainty. We first show that the natural Model Predictive Control (MPC) based algorithm for this class of problems can perform arbitrarily badly because of _temporal_ uncertainty. We then describe online algorithms, both randomized and deterministic, to handle time varying uncertainty in future reward structures and values. We establish that, in the deterministic case, discounting future rewards is a method to effectively de-randomize against possible futures, thus providing a theoretical justification for discounting in MPC. Time permitting we will also talk about recent work on multi-agent models for power systems and highlight important problems that require increased intelligence in the smart grid.

This talk will be accessible to a wide audience since we will give examples and intuition in lieu of detailed proofs. It may be of particular interest to those interested in AI, control theory, machine learning and smarter energy systems.

Biography: Balakrishnan Narayanaswamy received his PhD from the department of Electrical and Computer Engineering (ECE) at Carnegie Mellon University in 2011, after which he joined the IBM Research Lab in Bangalore, India. His research interests lie at the intersection of AI, optimization, learning and inference particularly using them to understand, model and combat noise and uncertainty in real world applications.

His current research centers on the application of novel, theoretically well motivated optimization algorithms to resource allocation problems that arise in next generation smarter energy management systems. His thesis research at Carnegie Mellon was in the application of information and coding theory, detection, probability theory and inference algorithms to a variety of sensing systems such as sensor networks, mobile robots, biological screening and drug discovery. During his graduate studies he also worked on problems ranging from target tracking, iris recognition, speaker recognition, multi-source separation to codes for next generation memory systems.

A system partly based on some of his work scored near the top of the DARPA iris grand challenge. He is a proud recipient of the National Talent Search (NTSE) and the Jawaharlal Nehru (JNCASR) scholarships from the government of India during my undergraduate studies. He is currently on the TPC of the Energy in Communication, Information and Cyber-Physical Systems (E6) workshop at COMSNETS 2013.

Host: Professor Raghu Raghavendra

Location: Hughes Aircraft Electrical Engineering Center (EEB) - 248

Audiences: Everyone Is Invited

Contact: Janice Thompson