초록 ▼

A method of creating and using an adaptive DMC type or other MPC controller includes using a model switching technique to periodically determine a process model, such as a parameterized process model, for a process loop on-line during operation of the process. The method then uses the process model

A method of creating and using an adaptive DMC type or other MPC controller includes using a model switching technique to periodically determine a process model, such as a parameterized process model, for a process loop on-line during operation of the process. The method then uses the process model to generate an MPC control model and creates and downloads an MPC controller algorithm to an MPC controller based on the new control model while the MPC controller is operating on-line. This technique, which is generally applicable to single-loop MPC controllers and is particularly useful in MPC controllers with a control horizon of one or two, enables an MPC controller to be adapted during the normal operation of the process, so as to change the process model on which the MPC controller is based to thereby account for process changes. The adaptive MPC controller is not computationally expensive and can therefore be easily implemented within a distributed controller of a process control system, while providing the same or in some cases better control than a PID controller, especially in dead time dominant process loops, and in process loops that are subject to process model mismatch within the process time to steady state.

대표청구항 ▼

What is claimed is: 1. A process controller for controlling a process, the process controller comprising: a dynamic matrix control controller having a controller algorithm that uses a gain vector to produce a control signal based on an input signal, the control signal being developed for controllin

What is claimed is: 1. A process controller for controlling a process, the process controller comprising: a dynamic matrix control controller having a controller algorithm that uses a gain vector to produce a control signal based on an input signal, the control signal being developed for controlling a process variable; a process model estimation unit coupled to the process to collect process data during the on-line operation of the process and that determines a process model representing the operation of the process from the collected process data, the process model including multiple model parameters including a process dead time parameter; and a controller adaptation unit that uses the process model including the process dead time parameter to calculate a new gain vector for use in the dynamic matrix controller algorithm and that adapts the controller algorithm of the dynamic matrix controller to use the new gain vector, wherein the controller adaptation unit adapts the controller to use the new gain vector while the controller is operating on-line to control the process. 2. The process controller of claim 1, wherein the dynamic matrix control controller is a model predictive control (MPC) controller. 3. The process controller of claim 2, wherein the MPC controller is a single-loop controller. 4. The process controller of claim 3, wherein the MPC controller includes one or more feedback paths and one or more feedforward paths. 5. The process controller of claim 4, wherein the control algorithm uses a control horizon equal to two scans. 6. The process controller of claim 4, wherein the control algorithm uses a control horizon equal to one scan. 7. The process controller of claim 2, wherein control algorithm includes a penalty on move variable that is altered during operation of the MPC controller. 8. The process controller of claim 2, wherein the MPC controller includes a set point target vector filter having a set point target vector filter parameter, and wherein the set point target vector filter parameter is altered during operation of the MPC controller. 9. The process controller of claim 2, wherein the process model is a first order plus dead time process model. 10. The process controller of claim 2, wherein the process model is a parametric process model. 11. The process controller of claim 2, wherein the controller adaptation unit determines a prediction horizon for the control algorithm based on the process model. 12. The process controller of claim 2, wherein the controller adaptation unit determines an execution time for the MPC controller based on the process model. 13. The process controller of claim 2, wherein the controller adaptation unit determines a time to steady state from the process model and determines a combination of a prediction horizon for the control algorithm and an execution time for the MPC controller based on the time to steady state. 14. The process controller of claim 2, wherein the controller adaptation unit determines a value for a penalty on move variable used in the controller algorithm based on the process model. 15. The process controller of claim 14, wherein the process model defines a process dead time and the controller adaptation unit determines the value of the penalty on move variable based on the process dead time. 16. The process controller of claim 1, wherein the dynamic matrix control controller includes a prediction unit that determines a prediction vector for a set of controlled variables, a prediction error unit that combines the prediction vector with a set point vector to produce a prediction error vector and wherein the controller algorithm multiplies the prediction error vector by the gain vector to produce a change in a control signal to be used to develop the control signal. 17. The process controller of claim 1, wherein the controller adaptation unit calculates the new gain vector using a closed form equation. 18. The process controller of claim 17, wherein the closed form equation is a function of one or more of the multiple model parameters. 19. The process controller of claim 18, wherein the closed form equation includes a term related to a penalty on move. 20. The process controller of claim 19, wherein the controller adaptation unit determines the penalty on move as a function of the process dead time parameter. 21. The process controller of claim 19, wherein the process model includes a process time constant as one of the process parameters and wherein the controller adaptation unit determines the penalty on move as a function of the process time constant. 22. The process controller of claim 18, wherein the controller adaptation unit includes a step response model generation unit that generates a step response model from the multiple model parameters of the process model, the step response model having a set of step response model gains, and wherein the closed form equation is a function of the step response model gains. 23. The process controller of claim 1, wherein the new gain vector comprises a matrix. 24. The process controller of claim 1, wherein the new gain vector comprises a scalar. 25. A process controller for controlling a process, the process controller comprising: a dynamic matrix control controller having a controller algorithm that uses a gain vector to produce a control signal based on an input signal, the control signal being developed for controlling a process variable; a process model estimation unit coupled to the process to collect process data during the on-line operation of the process and that determines a process model representing the operation of the process from the collected process data; and a controller adaptation unit that uses the process model to calculate a new gain vector for use in the dynamic matrix controller algorithm and that adapts the controller algorithm of the dynamic matrix controller to use the new gain vector, wherein the controller adaptation unit adapts the controller to use the new gain vector while the controller is operating on-line to control the process; wherein the dynamic matrix control controller is a single-loop, model predictive control (MPC) controller that includes one or more feedback paths and one or more feedforward paths and wherein the control algorithm uses a control horizon at least ten times shorter than a prediction horizon.