Have you found the perfect application for Model Predictive Control (MPC) but don’t know how to get started? Or perhaps you have implemented an MPC but you are not getting the expected results. This workshop will explain how the DeltaV MPC works in easy-to-understand terms. Field proven practical tips and techniques to implement MPC will be explained. This information will be worth years of experience in applying DeltaV MPC!
He opened by defining DeltaV MPC as a multivariable, Model Predictive Controller. The MPCPro block contains a dynamic controller and a linear optimizer. The DeltaV MPC block has only the dynamic controller. A model predictive controller has a model, learned from history, to be able to predict what’s going to happen in the future. Process inputs are manipulated variables (MV) and disturbance variables (DV). MVs write to valves or controller setpoints. DVs are measured variables that affect the value of controlled variables.
“Process” Outputs include controlled variables (CV) and Constraints (limiting variables – LV). CVs are process variables, which are to be maintained at a specific value (i.e. setpoint). LVs are variables that must be maintained within an operating range (a special type of CV).
The steady start part of the model is a matrix with coefficients of the process variables. MVs and DVs are inputs to the process model. The output of the model feeds the CVs and LVs. Process models are derived from observed step tests of the variables. James did an example with a shower where the MVs were the hot and cold-water controls. The output was the water temperature and shower flow rate. To get the right temperature and control rate you were doing complex matrix math and you didn’t even know it!
In DeltaV, PredictPro is the application to determine the process models, setup, and to tune the MPCPro block. By switch to expert mode automatically, it selects the variables to be in the dynamic controller. It will try to pick the best settings.
James noted a truism in applying models in process control, “All models are wrong, but some are useful.” He noted some ways to tune the dynamic controller portion of MPC control by adjusting the Penalty on Error for the CV and LV. Usually the change is plus or minus 20% although it can be cut in half for integrating variables and be 1/10 the value for some special optimization applications. The MV can be adjusted through the Penalty on Move parameter.
James highlighted another control analogy doing automobile speed control by manipulating the accelerator and the break. To achieve 50% speed, one could have the accelerator at 50% and the brake at 0%, or the accelerator at 100% and the brake at 50% (for the fans of red-hot glowing brakes!), or the accelerator at 80% and the brake at 30%, etc. If we optimize for braking the best spot to operate the model would be with the accelerator at 50% and the brake at 0%. The Linear Programming (LP) optimizer helps to calculate these optimum points for the variables with linear relationships.
James shared the interaction between the optimizer and dynamic controller portions of the MPC controller. The optimizer first calculates the target value for the MVs at the end of steady state time (Tss) based on the selected Objective Function. Based on these target values, the optimizer calculates the CVs and LVs at the end of Tss, which are now the target setpoints for the CVs, and LVs. The dynamic controller moves the MVs to the target setpoint for the CVs and LVs that are in the dynamic controller.
Take a look at the presentation for some tips James gives on troubleshooting and where to get more information on DeltaV model predictive control.