Back to the program.
Distributed Supervision Strategies for Constrained Multi-agent Systems
Emanuele Garone
Service d'Automatique et D'Analyse des Systemes
ULB
egarone@ulb.ac.be

Abstract

In this talk we will present a family of distributed supervision strategies for multi-agent coupled linear systems connected via data networks and subject to coordination constraints. The problem of distributed supervision consists in locally modifying the desired reference of each subsystem in such a way that both local and global constraints are not violated. Interestingly, the use of the so-called Feed-Forward Command Governor reduces the problem into the one of solving, at each sampling time, a static problem. More precisely, each agent has to select an applied reference (possibly approximating the desired one) in such a way that the aggregation of all the local decisions does not violate a set of static global geometrical constraints. The ability to deal with such a problem may be seen as the key-issue for the development of any distributed supervision strategy. During the presentation we will focus on the development of distributed strategies to solve the latter static problem and we will analyze what are their control properties when applied in a Receding Horizon fashion (i.e. when this strategy is repeated at each time step). Different distributed strategies will be described and analyzed and their convergence properties, either to Pareto Optimal or to Nash Equilibria, pointed out. In order to show the effectiveness of the proposed methods, some applications are finally presented.

Keywords

Command Governor, Model Predictive Control, Distributed Control, Constraints