Asymptotic minimal communication for decentralized discrete-event control

Abstract

Algorithms for finding minimal communication policies for decentralized discrete-event control use the following notion for optimality: remove any one of the elements from the communication set and either the control problem can no longer be solved correctly or a condition of observational equivalence is violated. In contrast, a strategy for finding a globally optimal communication policy, based not on structural properties of the system, but rather, in a behavioral sense, is presented. The problem of finding a minimal communication set is reduced to an optimization problem for a set of Markov chains.