ABSTRACT
In this chapter, the authors highlight the key principles behind the Model predictive control (MPC) strategy and introduce the traditional dynamic matrix control implementation as a means to demonstrate why MPC is such an appealing technology for the process industries. For linear MPC applications, most algorithms use one of the three model forms: impulse-response models, step-response models, and state-space models. One alternative approach is to employ a discrete response (convolution) model. The advantage of these models is that the model coefficients can be obtained directly from the experimental step response. MPC approach allows to design a controller that satisfies the constraints on current and future values of the controlled and manipulated variables. There are many commercial algorithms, some involving shortcut procedures, that can solve the constrained optimization problem in real-time with reasonable efficiency. The, typically quadratic, cost function of MPC ensures a stable and satisfactory closed-loop response but has no direct bearing on cost of the real-time operation.
