Multivariable nonlinear process controller

• US 5,570,282 A
• Filed: 11/01/1994
• Issued: 10/29/1996
• 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): RUGGIERO, JOSEPH F
• Application Number:
  US08/333,161
• Time in Patent Office: 728 Days
• Field of Search: 364/148, 364/152, 364/153, 364/157-159, 364/160-163, 364/164, 364/165, 364/176, 364/177, 318/561, 395/21-24, 395/68, 395/903, 395/906
• US Class Current: 700/41
• CPC Class Codes: C08F 2400/02 Control or adjustment of polymerization parameters
• 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 that utilizes a neural-network 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 neural-network controller is trained to achieve optimal performance behavior when future process behavior repeats the training experience. The outer-loop controllers compensate for process changes, unmeasured disturbances, and modeling errors. In the first and second embodiments, the neural network is used as an inner-loop controller in a process control system having a constraint management scheme which prevents integral windup by controlling the action of the outer-loop controllers when limiting is detected in the associated manipulated-variable control path. In the second and third embodiments, the neural-network controller is used without the integral controllers or the constraint management scheme as a simple PD feedforward controller.

Claims

1. An apparatus for controlling a process having process inputs and process outputs comprising a controlled variable, the process responsive to a manipulated variable to vary the process in relation to the controlled variable, said apparatus comprising:

a) measuring means for measuring values of one of said process outputs generating a process output signal representative thereof;

b) setpoint means for receiving a setpoint which represents a target value of the controlled variable;

c) control means, coupled to said measuring means and the process, for generating a selected value of the manipulated variable as a function of the setpoint and the process output signal, said control means utilizing a nonlinear function generator previously trained to compute the selected value of the manipulated variable in accordance with an optimum prediction time, the optimum prediction time representing the effective response time of the process to a change in the setpoint, and the selected value of the manipulated variable representing a change in the process needed to move the controlled variable towards the setpoint as advanced by the optimum prediction time; and

d) actuator means, coupled to said control means and said process, for applying the selected value of the manipulated variable to the process.

2. An apparatus as in claim 1

wherein said measuring means generates a derivative process output signal representative of the derivative, with respect to time, of the process output;

and

wherein said control means computes the manipulated variable in accordance with said function also including the derivative process output signal, said control means thereby providing a proportional and derivative control action.

3. An apparatus as in claim 1

further comprising a means for receiving measurements of process inputs;

and

wherein said control means computes the manipulated variable, in accordance with said function also including the process inputs, thereby providing additional feedforward control action.

4. An apparatus as in claim 3 wherein said process inputs further comprise a measured load variable representative of measured load disturbances.

5. An apparatus as in claim 4 wherein said process inputs further comprise a derivative measured load variable representative of the derivative, with respect to time, of the measured load variable.

6. An apparatus as in claim 1 further comprising

an integral control means for providing integral control action in response to the process output signal, the setpoint, and an integral time constant, thereby calculating an inner setpoint in order to compensate for process changes, said integral control means producing the inner setpoint for use by the control means.

7. An apparatus as in claim 6 wherein said control means further generates the optimum prediction time for use in generating the integral time constant by said integral control means.

8. An apparatus as in claim 6 wherein said integral control means is selected from a PID controller, an integral controller, or a PI controller.

9. An apparatus as in claim 6 wherein said integral control means further comprises:

a) means for detecting limiting of the manipulated variable; and

b) wherein said integral control means performs a control action in response to the limiting of the manipulated variable.

10. An apparatus as in claim 1

wherein the process outputs further comprise an inferential variable representative of an approximate indication of the controlled variable.

11. An apparatus as in claim 1

wherein said controlled variable is a measured process output.

12. An apparatus as in claim 1

wherein said apparatus is a multivariable controller and the process output signal further comprises a plurality of process output signals, the manipulated variable further comprises a plurality of manipulated variables, the setpoint further comprises a plurality of setpoints, and the optimum prediction time further comprises a plurality of optimum prediction times.

13. An apparatus as in claim 1 wherein said nonlinear function generator is a neural network.

14. An apparatus for controlling a process having process inputs and process outputs comprising at least one controlled variable, the process being responsive to at least one manipulated variable for changing the process in relation to the controlled variable, said apparatus comprising:

a) measuring means for measuring values of at least one of said process outputs and generating at least one process output signal representative thereof;

b) setpoint means for receiving at least one setpoint which represents a target value of the controlled variable;

c) control means, coupled to said measuring means and the process, for generating a selected value of the manipulated variable and at least one optimum prediction time as a function of the setpoint and the process output signal, said control means utilizing a nonlinear function generator previously trained to determine a selected value of the manipulated variable and the optimum prediction time, the optimum prediction time representing the effective response time of the process to a change in the setpoint, and the selected value of the manipulated variable representing a change in the process needed to produce a value of the controlled variable that approaches the setpoint as advanced by the optimum prediction time;

d) an integral control means for providing integral control action in response to the process output signal, the setpoint, and the optimum prediction time, thereby calculating an inner setpoint in order to compensate for process changes, the optimum prediction time being used in generating an integral time constant, said integral control means producing the inner setpoint for use by the control means; and

e) actuator means, coupled to said control means and the process, for applying the manipulated variable to the process.

15. An apparatus as in claim 14

wherein said measuring means generates a derivative, with respect to time, of each process output signal;

and

wherein said control means computes each manipulated variable in accordance with said function also including the derivative of the process output signal, said control means thereby providing a proportional-and-derivative control action.

16. An apparatus as in claim 14

further comprising a means for receiving measurements of the process inputs;

and

wherein said control means computes each manipulated variable in accordance with said function also including the process inputs, thereby providing additional feedforward control action.

17. An apparatus as in claim 16 wherein said process inputs further comprise at least one measured load variable representative of measured load disturbances.

18. An apparatus as in claim 17 wherein said process inputs further comprise at least one derivative, with respect to time, of each measured load variable.

19. An apparatus as in claim 14 wherein said integral control means is selected from a PID controller, an integral controller, or a PI controller.

20. An apparatus as in claim 14 wherein said integral control means further comprises:

a) means for detecting limiting of each manipulated variable; and

b) wherein said integral control means performs a control action in response to the limiting of each manipulated variable.

21. An apparatus as in claim 14 wherein said controlled variable is a measured process output.

22. An apparatus as in claim 14 wherein said nonlinear function generator is a neural network.

23. An apparatus for controlling a process having process inputs and process outputs comprising at least one controlled variable and at least one inferential variable, the process responsive to at least one manipulated variable for changing the process in relation to the controlled variable, said apparatus comprising:

a) a first measuring means for measuring values of a measured process output generating at least one controlled variable representative thereof;

b) a second measuring means for measuring a process output thereby producing an approximate indication of the controlled variable and generating at least one inferential variable representative thereof;

c) setpoint means for receiving at least one setpoint which represents a target value of the controlled variable;

d) control means, coupled to said measuring means and the process, for generating a selected value of the manipulated variable and at least one optimum prediction time as a function of the setpoint and the inferential variable, said control means utilizing a nonlinear function generator previously trained to determine the selected value of the manipulated variable and the optimum prediction time, the optimum prediction time representing the effective response time of the process to a change in the setpoint, and the selected value of the manipulated variable representing a change in the process needed to produce the controlled variable that approaches the setpoint as advanced by the optimum prediction time;

e) an integral control means for providing integral control action in response to the controlled variable, the setpoint, and the optimum prediction time, thereby computing an inner setpoint in order to compensate for process changes, the optimum prediction time being used for generating the integral time constant, said integral control means producing the inner setpoint for use by the control means; and

f) an actuator means coupled to said control means and the process, for applying the manipulated variable to the process.

24. An apparatus as in claim 23

wherein said measuring means generates at least one derivative, with respect to time, of each inferential variable;

and

wherein said control means generates each manipulated variable in accordance with said function also including the derivative of the inferential variable, said control means thereby providing a proportional-and-derivative control action.

25. An apparatus as in claim 24

further comprising a means for receiving measurements of the process inputs;

and

wherein said control means produces each manipulated variable in accordance with the function including the process inputs, thereby providing additional control action for the process inputs.

26. An apparatus as in claim 25 wherein said process inputs further comprise at least one measured load variable representative of measured load disturbances.

27. An apparatus as in claim 26 wherein said process inputs further comprise at least one derivative, with respect to time, of each measured load variable.

28. An apparatus as in claim 24 wherein said integral control means is selected from a PID controller, an integral controller, or a PI controller.

29. An apparatus as in claim 24 wherein said integral control means further comprises:

a) means for detecting limiting of each manipulated variable; and

b) wherein said integral control means performs a control action in response to the limiting of each manipulated variable.

30. An apparatus as in claim 23 wherein said nonlinear functin generator is a neural network.

31. A multivariable nonlinear controller for controlling a process having process outputs comprising of at least one controlled variable, and an actuator affecting the process, the controller comprising:

a) means for receiving a signal representative of at least one of the process outputs;

b) means for receiving at least one setpoint indicating a target value of the controlled variable;

c) control means, in communication with the process output signal and the setpoint, said control means, utilizing a nonlinear function generator to produce at least one selected value of the manipulated variable to affect the process, said control means performing the operations of;

i) performing an integral control action using the process output signal, the setpoint, and an integral time constant, and generating at least one adjusted setpoint representative thereof;

ii) operating the nonlinear function generator as a function of the process output signal and the adjusted setpoint, and producing an optimum prediction time representing a response time of the controlled variable to the adjusted setpoint, the optimum prediction time used to compute the integral time constant for the integral control action, the nonlinear function generator producing the selected value of the manipulated variable for affecting the controlled variable as advanced by the optimum prediction time; and

iii) transmitting the selected value of the manipulated variable to the actuator for affecting the process.

32. A controller as in claim 31 wherein said means for receiving at least one process output signal

further comprises a means for computing and generating at least one derivative, with respect to time, of the process output signal and

wherein the step of operating the nonlinear function generator is performed in accordance with said function also including the derivative of the process output signal.

33. A controller as in claim 31 wherein said means for receiving at least one process output signal further comprises a means for receiving at least one measured load variable representative of measured load disturbances and

wherein the step of operating the nonlinear function generator is performed in accordance with said function also including the measured load variable.

34. A controller as in claim 33 wherein said means for receiving at least one measured load variable further comprises a means for computing and generating at least one derivative measured load variable representative of the derivative, with respect to time, of the measured load variable and

wherein said operation of the nonlinear function generator is performed in accordance with said function also including the derivative measured load variable.

35. A controller as in claim 31 wherein said step of performing integral control action is selected from a PID controller, an integral controller, or a PI controller.

36. A controller as in claim 31 wherein the step of performing integral control action further comprises the step of detecting limiting of each manipulated variable and performing the integral control action in response to the limiting.

37. A controller as in claim 31 wherein the nonlinear function generator is a neural network.

38. A multivariable nonlinear controller for controlling a process having a process output comprising controlled and inferential variables, and an actuator affecting the process, the controller comprising:

a) means for receiving at least one process output signal;

b) means for generating at least one inferential variable representing approximate indications of the controlled variable;

c) means for receiving at least one setpoint indicating a target value of the controlled variable;

d) control means, in communication with the process output signal and the setpoint, said control means, utilizing a nonlinear function generator to produce at least one selected value of the manipulated variable to affect the process, said control means performing the operations of;

i) performing an integral control action using the process output signal, the setpoint, and an integral time constant, and generating at least one adjusted setpoint representative thereof;

ii) operating the nonlinear function generator in response to the inferential variable and the adjusted setpoint, and producing an optimum prediction time representing a response time of the controlled variable to the adjusted setpoint, the optimum prediction time used to generate the integral time constant for the integral control action, the nonlinear function generator producing the selected value of the manipulated variable for affecting the controlled variable to approach the setpoint as advanced by the optimum prediction time; and

iii) transmitting the manipulated variable to the actuator for affecting the process.

39. A controller as in claim 38

further comprising a means for computing and generating at least one derivative, with respect to time, of the inferential variable and

wherein the step of operating the nonlinear function generator is performed in accordance with said function also including the derivative of the inferential variable.

40. A controller as in claim 38

further comprising a means for receiving at least one measured load variable representative of measured load disturbances and

wherein the step of operating the nonlinear function generator is performed in accordance with said function also including the measured load variable.

41. A controller as in claim 38

further comprising a means for computing and generating at least one derivative measured load variable representative of the derivative, with respect to time, of the measured load variable and

wherein said operation of the nonlinear function generator is performed in accordance with said function also including the derivative measured load variable.

42. A controller as in claim 38 wherein said integral control action of the step performing integral control action is selected from a PID controller, an integral controller, or a PI controller.

43. A controller as in claim 38 wherein the step of performing integral control action further comprises the step of detecting limiting of each manipulated variable and performing the integral control action in response to the limiting.

44. A controller as in claim 38 wherein the nonlinear function generator is a neural network.

45. 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 utilizing a nonlinear function generator for computing at least one manipulated variable for use by the actuator to affect the controlled variable, the method comprising the steps of:

a) measuring the process output generating at least one process output signal representative thereof;

b) receiving at least one setpoint representing the target value of the controlled variable;

c) operating the nonlinear function generator as a function of the process output signal and the setpoint to generate a selected value of the manipulated variable, the nonlinear function generator previously trained to generate the selected value of the manipulated variable based on an optimum prediction time, the optimum prediction time representing a response time of the controlled variable to the setpoint, and the selected value of the manipulated variable representing a change made to the process in order to affect the controlled variable to approach the setpoint as advanced by the optimum prediction time; and

d) transmitting the manipulated variable to the actuator for applying the manipulated variable to the process.

46. A method as in claim 45

wherein the step of measuring the process output further comprises the step of computing at least one derivative, with respect to time, of the process output generating at least one derivative process output signal representative thereof;

and

wherein the step of operating the nonlinear function generator is performed with said function also including the derivative of the process output signal in order to generate the manipulated variable.

47. A method as in claim 45

wherein the step of measuring the process output further comprises the step of measuring load disturbances of the process generating at least one measured load variable representative thereof;

and

wherein the step of operating the nonlinear function generator is performed with said function also including the measured load variable in order to generate the manipulated variable.

48. A method as in claim 45

wherein the step of measuring load disturbances further comprises the step of computing the derivative, with respect to time, of at least one measured load variable and generating at least one derivative measured load variable representative thereof;

and

wherein said step of operating the nonlinear function generator is performed with said function also including the derivative measured load variable in order to generate the manipulated variable.

49. A method as in claim 45 wherein the process output is at least one controlled variable representing a measured process output.

50. A method as in claim 45 wherein the process output is at least one inferential variable representing an approximate indication of the controlled variable.

51. A method as in claim 45 wherein the nonlinear function generator is a neural network.

52. A system for controlling a process, the process having process inputs, process outputs comprising of a plurality of controlled variables, and an actuator affecting the process, the system comprising:

a) first control means for generating a plurality of selected values of the manipulated variables for use by the actuator for affecting the process, said first control means performing the steps of;

i) measuring load disturbances of the process and generating a plurality of measured load variables representative thereof and generating the derivative, with respect to time, of the measured load disturbances, thereby producing a plurality of derivative measured load variables representative thereof;

ii) measuring a plurality of process outputs generating a plurality of process output signals representative thereof and generating the derivative of the process output signal producing a plurality of derivative process output signals representative thereof;

iii) receiving a plurality of adjusted setpoints representing values of the controlled variables for use by the first control means in determining the selected values of the manipulated variables;

iv) producing the selected values of the manipulated variables and a plurality of optimum prediction times as a function of the adjusted setpoints, the measured load variables, the derivative measured load variables, the inferential variables, and the derivative inferential variables, the optimum prediction time representing a response time of the controlled variable to the adjusted setpoint, and the selected values of the manipulated variable representing changes made to the process in order to affect the controlled variables to approach the adjusted setpoints as advanced by the optimum prediction time;

v) transmitting the selected values of the manipulated variable to the actuator for application to the process;

b) second control means for generating the adjusted setpoint by performing the steps of;

i) receiving a desired setpoint representing a target value of the process;

ii) measuring the process thereby generating a controlled variable representative thereof;

iii) determining whether the manipulated variable has been limited and generating a logic signal representative thereof; and

iv) calculating the adjusted setpoint by performing an integral control action as a function of the desired setpoint, the controlled variable, the logic signal, and the optimum prediction time produced from said first control means, said calculation step producing the adjusted setpoint for use by the first control means in generating the manipulated variables which will affect the process.

53. A system as in claim 52 wherein said process output signals are controlled variables representing measured process outputs.

54. A system as in claim 52 wherein said process output signals are inferential variables representing approximate indications of the controlled variables.