Method and apparatus for controlling multivariable nonlinear processes

• US 5,566,065 A
• Filed: 11/01/1994
• Issued: 10/15/1996
• Est. Priority Date: 11/01/1994
• Status: Expired due to Term

• Current Assignee: Schneider Electric Systems USA, Inc. (Schneider Electric SE)
• Original Assignee: The Foxboro Company (Schneider Electric SE)
• Inventors: Hansen, Peter D., Badavas, Paul C.
• Primary Examiner(s): Mai, Tan V.
• Assistant Examiner(s): Garland, Steven R.
• Application Number:
  US08/333,539
• Time in Patent Office: 714 Days
• Field of Search: 364/148, 364/137, 364/149-154, 364/160-165, 395/903, 395/904, 395/906, 395/21, 395/22
• US Class Current: 700/44
• CPC Class Codes: G05B 13/026 using a predictor
• CPC Class Codes: G05B 13/027 using neural networks only

Abstract

A method and apparatus for a robust process control system which utilizes a neural-network based multivariable inner-loop PD controller cascaded with decoupled outer-loop controllers with integral action, the combination providing a multivariable nonlinear PID and feedforward controller. The inner-loop PD controller employs a quasi-Newton iterative feedback loop structure whereby the manipulated variables are computed in an iterative fashion as a function of the difference between the inner loop setpoint and the predicted controlled variable as advanced by the optimum prediction time, in order to incorporate the downstream limiting effects on the non-limited control loops. The outer-loop controllers compensate for unmodeled process changes, unmeasured disturbances, and modeling errors by adjusting the inner-loop target values.

Claims

1. A method for controlling a process having process inputs, process outputs comprising of at least one controlled variable, and an actuator affecting the process, the method comprising computing at least one manipulated variable for use by the actuator to affect the controlled variable in relation to a setpoint, the method comprising the steps of:

a) obtaining a set of manipulated variables;

b) selecting a first manipulated variable from the set of manipulated variables that has not been previously selected;

c) computing a predicted controlled variable having an associated setpoint and corresponding to said first manipulated variable as a function of the process inputs and outputs, said predicted controlled variable representing the process output as advanced by an optimum prediction time, said optimum prediction time representing the effective response time of the controlled variable to the setpoint;

d) determining sensitivity from a change in value of the predicted controlled variable to a change in value of the corresponding first manipulated variable;

e) determining a difference between the predicted controlled variable and the associated setpoint;

f) incrementing the value of the first manipulated variable in relation to the sensitivity and the difference;

g) repeating steps b) through f) for each manipulated variable from the set of manipulated variables; and

h) repeating steps b) through g) until a predetermined condition is satisfied, said predetermined condition being the determination of the optimum prediction time wherein a minimal change in the manipulated variable is required to affect a change in the controlled variable that minimizes the difference between the controlled variable and the associated setpoint so that the difference is within a predetermined tolerance .

2. A method as in claim 1 wherein the step of incrementing the value of the first manipulated variable further comprises modifying the value by utilizing a partial derivative of the first manipulated variable with respect to the corresponding predicted controlled variable.

3. A method as in claim 1 wherein the function defined in the step of computing a predicted controlled variable comprises the process inputs comprising the set of manipulated variables, and the process outputs comprising the controlled variables.

4. A method as in claim 3 wherein the function defined in the step of computing a predicted controlled variable further comprises the process outputs comprising derivatives, with respect to time, of the controlled variables.

5. A method as in claim 3 wherein the function defined in the step of computing a predicted controlled variable further comprises the process inputs comprising measured load variables representing measured load disturbances.

6. A method as in claim 5 wherein the function defined in the step of computing a predicted controlled variable further comprises the process inputs comprising derivatives, with respect to time, of the measured load variables.

7. A method as in claim 1 wherein the function defined in the step of computing a predicted controlled variable comprises the process inputs comprising a set of inferential variables.

8. A method as in claim 7 wherein the function defined in the step of computing a predicted controlled variable comprises the process outputs comprising the controlled variables.

9. A method as in claim 8 wherein the function defined in the step of computing a predicted controlled variable further includes the process outputs comprising derivatives, with respect to time, of the controlled variables.

10. A method as in claim 7 wherein the function defined in the step of computing a predicted controlled variable further includes the process inputs comprising the measured load variables representing measured load disturbances.

11. A method as in claim 10 wherein the function defined in the step of computing a predicted controlled variable further includes the process inputs comprising derivatives, with respect to time, of the measured load variables.

12. A method as in claim 1 wherein said step of computing a predicted controlled variable is performed by a nonlinear function generator previously trained to determine the optimum prediction time and said predicted controlled variable on historical values of the process inputs and outputs.

13. A method as in claim 12 wherein said nonlinear function generator is selected from a neural network, fuzzy logic processor, or inferential logic machine.

14. A method as in claim 1 wherein the step of incrementing the value of the first manipulated variable further comprises the step of limiting the first manipulated variable thereby preventing the value from exceeding its limits.

15. A method as in claim 1 wherein said predetermined condition is when the difference between the setpoint and the predicted controlled variable for each manipulated variable is within a predetermined tolerance.

16. A method as in claim 1 wherein said predetermined condition is achieved within a finite number of iterations.

17. A method as in claim 1 wherein said step of computing a predicted controlled variable further comprises generating the optimum prediction time for further control processing.

18. A system for controlling a process, said process responsive to a set of manipulated variables for affecting the process as a function of a controlled variable in order to place the process at a desired setpoint, said system comprising:

a) means for retrieving a set of current values of the manipulated variables;

b) selector means for choosing a selected manipulated variable from the set of current values of the manipulated variables that has not been previously selected;

c) a nonlinear function generator for generating a predicted controlled variable as advanced by an optimum prediction time, the optimum prediction time representing the effective response time of the controlled variable to the setpoint;

d) a sensitivity processor for determining the sensitivity of a change in value of the predicted controlled variable to a change in the corresponding manipulated variable;

e) a first computation means for determining the difference between the predicted controlled variable and the setpoint corresponding to the selected manipulated variable;

f) a second computation means for recomputing the value of the selected manipulated variable as a function of said difference, said sensitivity, and said current value;

g) a limiter for preventing the selected manipulated variable from exceeding its limits; and

h) a processing element for operating said system for all values of the manipulated variable and for completing the operation of said system when a predetermined condition is satisfied.

19. A system as in claim 18 wherein said nonlinear function generator is selected from a neural network, fuzzy logic processor, or inferential logic machine.

20. A system as in claim 18 wherein said nonlinear function generator operates with inputs comprising current values of the manipulated variables and controlled variables representing a measured process output.

21. A system as in claim 20 wherein said nonlinear function generator operates with inputs further comprising derivatives, with respect to time, of the controlled variables.

22. A system as in claim 20 wherein said nonlinear function generator operates with inputs further comprising measured load variables representing measured load disturbances.

23. A system as in claim 22 wherein said nonlinear function generator operates with inputs further comprising derivatives, with respect to time, of the measured load variables.

24. A system as in claim 18 wherein said nonlinear function generator operates with inputs comprising the current values of the manipulated variables and inferential variables representing an indication of the controlled variables.

25. A system as in claim 24 wherein said nonlinear function generator operates with inputs further comprising derivatives, with respect to time, of the inferential variables.

26. A system as in claim 24 wherein said nonlinear function generator operates with inputs further comprising measured load variables representing measured load disturbances.

27. A system as in claim 26 wherein said nonlinear function generator operates with inputs further comprising derivatives, with respect to time, of the measured load variables.

28. A system as in claim 18 wherein said nonlinear function generator generates the optimum prediction time for use in further control processing.

29. A system as in claim 18 wherein said nonlinear function generator was previously trained to compute the predicted controlled variable as advanced by the optimum prediction time on historical values of process inputs and process outputs.

30. A system as in claim 18 wherein said predetermined condition is when the difference between the setpoint and the predicted controlled variable for each manipulated variable is within a predetermined tolerance.

31. A system as in claim 18 wherein said predetermined condition is a finite number of iterations.