BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Distributed Model Predictive Control: Theory and Algorithms
DTSTART:20130521T111500
DTEND:20130521T120000
DTSTAMP:20260407T021105Z
UID:a7cf26e9b188323029059c6db1b07d29d0a216b2f62ce388ec1976ac
CATEGORIES:Conferences - Seminars
DESCRIPTION:Pontus Giselsson\nBio: I am a postdoc researcher at the Depart
 ment of Automatic Control since January 2013. My research interests includ
 e distributed optimization\, distributed control\, and model predictive co
 ntrol.\nIn this talk\, we discuss various topics related to distributed mo
 del\npredictive control (DMPC). DMPC is a control methodology for large-sc
 ale\ndynamical systems that consist of several subsystems with a sparse\nd
 ynamic interaction structure. In this context\, an optimization problem\nt
 hat takes the system-wide performance into account is solved in\ndistribut
 ed fashion in each sample of the DMPC controller. Two main\ntopics are dis
 cussed in this talk: Theory and Algorithms.\nTheory: Traditional methods t
 o show stability and recursive feasibility\nin model predictive control (M
 PC) include a terminal cost and a terminal\nconstraint set that usually in
 volve variables from all subsystems. This\nhinders distributed implementat
 ion of the solution algorithm and sets\nrequirements for new stability the
 ory in DMPC. We will briefly present\nsome results in this direction. We w
 ill also show that these results\ncan\, for some examples\, increase the r
 egion of attraction significantly\,\ncompared to using traditional MPC met
 hods.\nAlgorithms: To enable distributed implementation of the DMPC contro
 ller\,\ndual decomposition is used to solve the optimization problem in\nd
 istributed fashion. In dual decomposition\, a gradient method is\ntraditio
 nally used to maximize the dual problem. However\, gradient\nmethods are k
 nown for their slow rate of convergence. In this talk\, we\nwill present d
 ifferent ways to improve the slow convergence rate in dual\ndecomposition\
 , e.g.\; by using fast gradient methods\, by preconditioning\nthe problem 
 data\, and by incorporating second order information into the\ndistributed
  algorithm. We will\, by an example\, show that these methods\ncan reduce 
 the number of iterations in dual decomposition by several\norders of magni
 tude\, compared to if gradient methods are used.
LOCATION:ME C2 405
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR