Step-Response Models

A key featui-e of MPC is that a dynamic model of the pi ocess is used to pi-edict futui e values of the contmlled outputs. Thei-e is considei--able flexibihty concei-ning the choice of the dynamic model. Fof example, a physical model based on fifst principles (e.g., mass and energy balances) or an empirical model coiild be selected. Also, the empirical model could be a linear model (e.g., transfer function, step response model, or state space model) or a nonhnear model (e.g., neural net model). However, most industrial applications of MPC have relied on linear empirical models, which may include simple nonlinear transformations of process variables. 

The original formulations of MPC (i.e., DMC and IDCOM) were based on empirical hnear models expressed in either step-response or impulse-response form. For simphcity, we will consider only a singleinput, single-output (SISO) model. However, the SISO model can be easily generalized to the MIMO models that are used in industrial applications. The step response model relating a single controlled variable y and a single manipiilated variable u can be expressed as... 

The step-response model is also referred to as a finite impulse response (FIR) model or a discrete convolution model. 

The step-response model in Eq. (8-63) is equivalent to the following impulse response model ... 

The minimization of the quadratic performance index in Equation (16.2), subject to the constraints in Equations (16.5-16.7) and the step response model such as Equation (16.1), forms a standard quadratic programming (QP) problem, described in Chapter 8. If the quadratic terms in Equation (16.2) are replaced by linear terms, a linear programming program (LP) problem results that can also be solved using standard methods. The MPC formulation for SISO control problems described earlier can easily be extended to MIMO problems and to other types of models and objective functions (Lee et al., 1994). Tuning the controller is carried out by adjusting the following parameters ... 

There is one method that is based on a time-domain model. It was developed at Shell Oil Company (C, R. Cutler and B. L. Kamaker, Dynamic Matrix Control A Computer Control Algorithm, paper presented at the 86th National AlChE Meeting, 1979) and is called dynamic matrix control (DMC). Several other methods have also been proposed ihat are quite similar. The basic idea is to use a time-domain step-response model of the process to calculate the future changes in the manipulated variable that will minimize some performance index. Much of the explanation of DMC given in this section follows the development presented by C. C. Yu in his Ph.D. thesis (Lehigh University, 1987). 

Dynamic matrix control uses time-domain step-response models (called convolution models). As sketched in Fig. 8.18, the response (x) of a process to a unit step change in the input (Ami = ) made at time equal zero can be described by the values of x at discrete points in time (the fc, s shown on the figure). At r nTJ, the value of X is h r,. If Affii is not equal to one, the value of x at f = n7 is b j Aibi, The complete response can be described using a finite number (NP) values of b coefficients. NP is typically chosen such that the response has reached 90 to 95 percent of its final value. 

Step 4 Model Development The dynamic model is developed from the plant test data by selecting a model form (e.g., a step response model) and then estimating the model parameters. However, first it is important to eliminate periods of test data where plant upsets or other abnormal situations have occurred. Decisions to omit portions of the test data are based on visual inspection of the data, knowledge of the process, and experience. Parameter estimation is usually based on least squares estimation. 

In the step response model, a change in MV of size u causes a corresponding change in the CV, y. In matrix notation, this is described as... 

---