Model Predictive Control(MPC) is one of the leading advanced control
technologies employed today in the process industries. The ability to
incorporate complex objectives as well as constraints in a unified
framework make it an extremely attractive and handy tool. MPC
technology has evolved tremendously over the years. However most
advancements in MPC are geared towards its implementation in a single
processing unit. Any complex system, a chemical plant for instance, is
comprised of a number of subsystems. These subsystems are invariably
linked through shared resources, material and energy flows
and/or information links. This interaction complicates controller design.
In the traditional unit operations approach to system design, each
subsystem is constructed with
its individual control system forming a pure
decentralized control setup. Decentralized control assumes that each
subsystem functions independently and is not influenced by the
subsystems around it. Since this assumption is not valid, it would be
reasonable to assume that control system performance could be improved
if some information about the nature of the interactions between the
subsystems could be used in making control decisions. Our research is
aimed at utilizing the inherent characteristics of the MPC scheme;
modifying and gearing these
towards obtaining improved plantwide control. However in this process,
we also strive to minimize structural changes in the current
industrial control
configurations thereby enhancing the possibility of swift implementation.
Communication based Decentralized MPC
First Level of Information Exchange
As a first step towards enhancing plantwide control, we employ the
concept of information-exchange between the various model predictive
controllers through various levels. Communication constitutes the most
basic level of information exchange, each controller receives
information about the most likely course of action for the set of
subsystems that influence it. This information is then used by the
regulator to generate a suitable control move. The information
exchange is made possible by the introduction of an additional
interface viz. the Communicator. The communicator is akin to a
supervisory block for the set of controllers. It receives information
from each of the individual controllers, processes it and sends back
relevant information that pertains to the effect of the various
subsystems on the subsystem under consideration. The various
controllers then generate control moves taking this additional
information into account.
