Personal tools
Home Events Events Archive 2013 Efficiency and Risk Tradeoffs in Dynamic Multi-agent Networked Systems

Efficiency and Risk Tradeoffs in Dynamic Multi-agent Networked Systems

— filed under:

  • Seminar Series
When Feb 25, 2013
from 01:00 PM to 02:30 PM
Where Engr. IV Bldg., Shannon Room 54-134
Contact Name
Add event to calendar vCal

Mardavij Roozbehani

Massachusetts Institute of Technology  



In this talk, we first introduce a few problems that arise in the context of modeling, analysis, and design of future power networks, and in characterization of the trade-offs that exist among different performance and robustness objectives in such networks. We will then introduce a specific model of a dynamic oligopolistic energy market, in which, a set of distributed agents with market power dynamically update their output (consumption or production) decisions. In this model, the agents have complete knowledge of how their decisions affect the market price, and are fully rational in strategizing their decisions to minimize their expected cost. By characterizing the statistics of the stationary aggregate output process across a spectrum of networks from fully cooperative to fully non-cooperative, we show that a tradeoff exists between efficiency (aggregate system cost) and risk (tail probability of aggregate output). Although the non-cooperative network leads to an efficiency loss - widely known as the "price of anarchy" - the stationary distribution of the corresponding aggregate output process has a smaller tail, whereas, the cooperative network achieves higher efficiency at the cost of a higher probability of output spikes. Furthermore, the cooperative network has a smaller output variance, which can be interpreted as higher robustness to disturbances, but it also has a higher probability of large output spikes, which can be interpreted as higher fragility to certain disturbances. We then establish the connection between these tradeoffs and some results from the classical control literature, and conclude with suggestions for future research directions.


Mardavij Roozbehani is a principal research scientist at the Laboratory for Information and Decision Systems (LIDS) at the Massachusetts Institute of Technology (MIT). He received the Ph.D. degree in Aeronautics and Astronautics from MIT in 2008. His Ph.D. research focused on developing a control theoretic framework for verification and implementation of software systems. Between 2008 and 2011 he held postdoctoral and research scientist positions at MIT, LIDS, focusing on applications of control and optimization in power systems and energy networks. His current research interests and activities include distributed and networked control systems, software and finite-state control systems, and dynamics and economics of energy networks with an emphasis on robustness and risk. Dr. Roozbehani is the recipient of the 2007 AIAA graduate award for safety verification of real-time software systems.

Document Actions