Supervisory Control and Optimization

Abstract

Listen

Translate article

In the first part of this chapter, structures and algorithms of Model-based Predictive Control (MPC) and on-line process set-point optimization are presented, corresponding to implementations in the DiaSter system. The principle of MPC is first recalled, as a special case of the general principle of ”open loop with feedback optimal control”.MPC is now the most important advanced feedback control technique, widely used for supervisory feedback control and also implemented as a direct control algorithm, mainly where classical PID structures cannot deliver required control performance, due to difficult dynamics, strong interactions, active constraints. It is well known that a prerequisite for a successful MPC application is the use of a sufficiently accurate process model. Different models can be used, leading to different realizations of MPC algorithms. Two basic algorithms with linear models are presented: the Dynamic Matrix Control (DMC) algorithm, being one of the first to be implemented in the industry and still very popular there, and the Generalized Predictive Control (GPC) algorithm. The algorithms are given both in analytical versions (control laws, suitable for simple controllers) and in full numerical versions with constrained optimization problems solved at every sampling instant. A selected MPC algorithm with a non-linear process model, based on non-linear prediction and optimization using linearized models, is presented, considered by the authors to be both simple and very efficient in practice. Next, on-line set-point optimization for the MPC controller, integrated with its operation, is described. In the final part of the section, two example applications using the DiaSter modules implementing the presented techniques are discussed (obtained in the simulation mode, using the PExSim package): DMC control of concentration in a Continuous Stirred-Tank Reactor (CSTR) and two-dimensional GPC control of composition and temperature.