4-1 Introduction

Because Model Predictive Control (MPC) uses an experimental model it can create a future trajectory of the process response based on multiple measured process inputs. Unknown disturbances, noise, limit cycles, random error, and incorrect model parameters result in a bias correction and a shift of the trajectory. Since MPC seeks to minimize the squared error of the trajectory over the time horizon, the short-term effects of unknowns and erratic signals are minimized. In contrast, a PID only knows what it sees as a change or rate of change for one process feedback measurement in the current scan. Additionally, MPC can simultaneously manipulate multiple variables, whereas a PID control block is restricted to one controller output and one feedback measurement. Consequently, PID must use various downstream blocks, such as the "Splitter" block for split-range control of multiple manipulated variables and the "Control Selector" block for override control of multiple controlled variables. With all of these PID techniques, a single controlled variable is matched up with a single manipulated variable at any given time with no inherent knowledge of the dynamics of the pairing. PID uses sequential pairing whereas MPC offers simultaneous manipulation of multiple process inputs for control of multiple process outputs. Finally, MPC is inherently better at optimization because it can predict future violations of constraints; has built-in features for maximization, minimization, and setting priorities; has a tuning adjustment to smooth out the optimization; and has hooks to a linear program.

Section 4-2 of this chapter addresses how...