Model Predictive Control

Charlie Cutler and Ramaker caused quite a stir in 1980 when they presented a paper that provided information on a new technique known as Dynamic Matrix Control. Shell Oil had developed and deployed this technique for the control of large interactive multiple input-multiple output (MIMO) processes such as refinery distillation columns. This work by Shell was the first version of what is commonly referred to today as Model Predictive Control(MPC). Since that time Model Predictive Control has been installed in thousands of plants. However, the computation power and memory needed to implement model predictive control may require a computer that is layered on top of an older distributed control system. Thus, the, application of MPC in the past has often been limited to large, high throughput processes.

In a modern DCS, advances in the processors and memory of the controller make it possible to embed MPC in the controller. If the DCS controller supports MPC, then it is possible to apply MPC to small processes that have historically been controlled using multi-loop techniques based on PID. Also, MPC may be used to more effectively control processes that are dominated by deadtime and difficult dynamics such as inverse response than is possible with PID. In Chapter 13 of Control Loop Foundation – Batch and Continuous Processes we addresses the application of MPC to small processes and to processes that are dominated by deadtime and difficult dynamics, and examine the benefits of this approach versus traditional control techniques based on PID.

When a process output is a constraint parameter in PID feedback control, this constraint parameter measurement is simply added as an input to the MPC block as illustrated in the following figure. The design and implementation of an override control strategy using an MPC block is thus much simpler than implementing override control using two PID blocks and a control selector block.

MPC Constaint Control.jpg

In many cases, the use of MPC in place of PID control is a new concept to a control engineer who has designed and commissioned control systems and used traditional control strategies based on PID. An easy way to initially learn about MPC and to gain experience commissioning MPC blocks is to layer MPC blocks on top of PID-based traditional control strategies. For example, the following figure shows how MPC can be used to replace a single PID in a controller.

By changing the AO block mode to RCas, the engineer can allow the MPC block to control the process. When the AO mode is switched back to Cascade, the PID block may be used to control the process. Switching between Cascade and RCas mode is bumpless since the PID and MPC blocks both have back calculation connections.

Several model predictive control application examples are detailed in chapter 13 of Control Loop Foundation – Batch and Continuous Processes. Also, the model predictive control workshop included in that chapter provides several exercises that may be used to further explore model predictive control. By accessing the book’s web site, you may complete this model predictive control workshop using your web browser. The viewer below may be used to see the solution to this exercise.