Performance, information pattern trade-offs and computational complexity analysis of a consensus based distributed optimization method

Abstract:

This talk is concerned with a cross-disciplinary approach for the convergence of computer science and control in large scale networked control systems. In this talk a simple consensus based distributed optimization method is presented which approximates the solution of a linear quadratic optimal control problem subject to constraints using distributed decision makers subject to non-classical information pattern. This method can be seen as a mechanism for distributing the computational load of the centralized control to distributed decision makers. Distributed decision makers are constrained in terms of the pattern of local computation and information exchange, as a mechanism for managing the corresponding overheads. Feasibility (i.e., constraints satisfaction by the approximated solutions), convergence, and optimality of the method are shown. The computational complexity of this method is compared with the centralized method for Australia’s automated irrigation networks. It is shown that for automated irrigation networks, the computational complexity of the centralized method in terms of the number of subsystems is of the order of six; while the computational complexity of the proposed consensus based distributed optimization method is quadratic. Therefore, as will be shown, there is a significant advantage in terms of computational complexity in using the proposed consensus based distributed optimization method in large scale networks. Trade-offs between number of subsystems, computational complexity, tuning parameters of the consensus based distributed optimization method, and communication pattern are also illustrated for Australia’s automated irrigation networks.

Presentation Slides

Biography:Currently, from August 2011, Dr. Alireza Farhadi works as a Research Fellow in the Department of Electrical and Electronic Engineering of the University of Melbourne. From February 2010 to July 2011 he worked as a Post Doctoral Fellow at INRIA (the French national institute for research in computer science and control). Before that he worked as a Post Doctoral Fellow at the University of Ottawa, Canada (May 2008-August 2009) where he received his PhD degree in Electrical Engineering on Systems & Control and Telecommunications. His current research interest is concerned with cross-disciplinary approaches for the convergence of computer science and control in large scale networked control systems. His main research interest is concerned with integrated frameworks for control/communication/computation/scalability co-design for emerging complex systems and networks.