Control Techniques for Complex Networks

University Of Illinois At Urbana-Champaign, Urbana IL

Investigators

Abstract

This proposal concerns policy synthesis and performance evaluation for complex systems. Research will focus on the development of methods to achieve reliability in complex networks that are subject to substantial variability. Speci.c topics include resource allocation, performance evaluation, and new approaches to fast simulation and adaptation. Power transmission networks are taken as a primary application of this research, and will guide the directions for theoretical research. Intellectual Merits The proposed research will build upon recent advances in stochastic systems and stochastic networks to bring new approaches to model reduction and policy synthesis for complex, interconnected systems. The project will also serve as a bridge between researchers in economics and control systems to synthesize recent innovations from these .elds. The most basic issue addressed in this project is complexity management. In a variety of applications one seeks control solutions of reasonable complexity in spite of the complexity of the system to be controlled. It is very important that the solution provide some intuition to the user for the purposes of both prediction and policy improvement. Centralized optimal control solutions will be used as a benchmark in analyzing decentralized policies, and as a tool for constructing appropriate market mechanisms to ensure reliability in a decentralized setting. Traditional economic analysis to address reliability in power distribution systems is based on a static (equilibrium) model. While much insight has been gained, just as in the design of a control system for an airplane, a static equilibrium model is inadequate to address robustness and reliability. The PIs have shown recently that it is possible to construct tractable dynamic models of power distribution systems that reveal insight on the sources of low reserves in a decentralized setting. The discovery of new methods to combat these de.ciencies is a major goal of this research. Broader Impact Fundamental research on accelerated algorithms for learning and simulation has potential impact in various .elds, including computer vision and data mining. Similarly, the control techniques to be developed are based on minimal structural assumptions, and consequently there is high probability of crossover into various .elds. Research on large-scale electric power has tremendous potential commercial impact. A clear understanding of the dynamics of deregulated markets will provide for the .rst time a proper evaluation of the social value of reliability, and the construction of e.ective incentives and policy tools to ensure e.cient market outcomes. These results will provide forecasting tools for policy makers to predict behavior of users and suppliers, and design mechanisms to ensure reliable and e.cient service. A CSL-ECE-Economics Department seminar series on distributed networks will be initiated. Professors will be invited from across the country, and as far away as India to present recent work. This will strengthen existing collaborations within the University of Illinois as well as the broader research community in the networks area. The seminar series will also provide excellent training for the students involved.