Does size matter? The effects of supervisor reduction on minimal communication between distributed discrete-event agents.

摘要

When applying control to a large and complex discrete-event system, traditional distributed supervisors may only be able to enforce a subset of the desired behaviour. If that subset is unacceptable, the agents must communicate with each other to achieve the required control action. However, the potential for delayed or lost messages and negative effects on the system demands that transmissions must be minimized. An algorithm that determines a minimal communication scheme which guarantees each supervisor is always certain of its current state does exist, but any options that may result in a further reduction in transmissions are worthy of exploration. One of these is to perform a state-size reduction on the agents, reducing their complexity and (hypothetically) the amount of communication they require to perform their control tasks.; This research examines several approaches to supervisor reduction and defines a new relation between agents known as comparability that encapsulates most of the concepts found in those methods. It then explores what occurs during the execution of the minimal communication algorithm for two pairs of agents, one pair of which is comparable to the other. The conjecture is that the original supervisors will communicate at least what the reduced supervisors communicate and most likely more. Unfortunately, this statement is not always correct. Both a counterexample and a "proof of concept" example are examined to illustrate both possible outcomes. Although there is no definitive result, performing a state-size reduction can lessen or even eliminate the need for communication. This fact would prompt the designer of any such system to perform such a reduction, but compare results before implementing the smaller agents.