Method for implementing indirect controller

• US 7,400,935 B2
• Filed: 01/12/2007
• Issued: 07/15/2008
• Est. Priority Date: 06/05/2003
• Status: Expired due to Term

• Current Assignee: GE Digital Holdings LLC (GE Vernova, Inc.)
• Original Assignee: NeuCo Incorporated (General Electric Company)
• Inventors: Kohn, Daniel W., Lefebvre, Wesley Curt
• Primary Examiner(s): BARNES-BULLOCK, CRYSTAL JOY
• Application Number:
  US11/622,702
• Publication Number:
  US 20070118238A1
• Time in Patent Office: 550 Days
• Field of Search: 700/28, 700/29, 700 31- 33, 700 47- 60, 700108-110, 700/173, 700/174, 700/177, 702 81- 84, 702182-185, 706 14- 16, 706/19, 706/23, 706/25, 706/903, 706/904
• US Class Current: 700/47
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G05B 7/02 electric
• CPC Class Codes: G06F 18/24133 Distances to prototypes
• CPC Class Codes: G06V 10/82 using neural networks
• CPC Class Codes: Y10S 706/903 Control
• CPC Class Codes: Y10S 706/906 Process plant
• CPC Class Codes: Y10S 706/914 Process plant

Abstract

A system and method for implementing an indirect controller for a plant. A plant can be provided with both a direct controller and an indirect controller with a system model or a committee of system models. When the system model has sufficient integrity to satisfy the plant requirements, i.e., when the system model has been sufficiently trained, the indirect controller with the system model is automatically enabled to replace the direct controller. When the performance falls, the direct controller can automatically assume operation of the plant, preferably maintaining operation in a control region suitable for generating additional training data for the system model. Alternatively, the system model incorporates a committee of models. Various types of sources for errors in the committee of models can be detected and used to implement strategies to improve the quality of the committee.

Claims

1. A computer-implemented method for implementing an indirect controller in a plant operable for performing a process, the plant having a plurality of plant inputs, the inputs being adjustable according to control settings received from a controller, the method comprising:

operating the plant using a direct controller;

using the direct controller to generate training data for a model of the process during operation of the plant by the direct controller;

training the model of the process using the training data generated by the direct controller; and

operating the plant using said indirect controller after the model has been trained, thereby replacing the direct controller.

2. The computer-implemented method of claim 1, wherein said training data includes plant operating conditions, plant inputs, and actual plant outputs for given plant operating conditions and plant inputs.

3. The computer-implemented method of claim 1, wherein said indirect controller uses said model to search for a set of optimal plant inputs, said model predicting plant outputs for given plant inputs.

4. The computer-implemented method of claim 3, where said step of searching for a set of optimal plant inputs using said model includes the steps of:

individually moving each of said plant inputs to a new position, and evaluating the resulting performance of said process in accordance with the plant outputs predicted by the model, and

if improved performance of said process results from the move to the new position, accepting the new position for the plant inputs and searching for a new set of optimal plant inputs.

5. The computer-implemented method of claim 3, wherein said step of searching for the set of optimal plant inputs is implemented as a zero or first order search method.

6. The computer-implemented method of claim 1, wherein said model is implemented by a parametric adaptive modeling method.

7. The computer-implemented method of claim 6, wherein said parametric adaptive modeling method includes at least one of the following:

linear regression and neural network model.

8. The computer-implemented method of claim 7, wherein said neural network model is trained by backpropagating an error between a predicted output and an actual output.

9. The computer-implemented method of claim 1, wherein said model is implemented by a non-parametric adaptive modeling method.

10. The computer-implemented method of claim 9, wherein said non-parametric method includes evolutionary or genetic programming.

11. The computer-implemented method of claim 1, wherein the method includes training a plurality of models using the training data generated by the direct controller.

12. The computer-implemented method of claim 11, wherein said plurality of models have the same architectures.

13. The computer-implemented method of claim 11, wherein said plurality of models have different architectures.

14. The computer-implemented method of claim 1, wherein said plant uses a fossil-fuel furnace.

15. The computer-implemented method of claim 1, wherein the step of training the model includes:

providing to said model current operating conditions and current inputs selected by the direct controller;

determining predicted plant outputs using said model;

comparing the predicted plant outputs to actual plant outputs; and

establishing the model as trained if the error in the predicted plant outputs is below a predetermined threshold.

16. A computer program product, residing on a computer readable medium, for use in implementing an indirect controller with a model in a plant operable for performing a process, the plant having a plurality of inputs, the inputs being adjustable according to control settings received from a controller, the computer program product comprising instructions for causing a computer to:

select a direct controller to operate the plant;

generate training data from the plant during operation of the plant by the direct controller, wherein the direct controller generates the training data;

train the model of the process using the training data generated by the direct controller; and

select the indirect controller to operate the plant after the model has been trained, thereby replacing the direct controller.

17. The computer program product of claim 16, wherein said training data includes plant operating conditions, plant inputs, and actual plant outputs for given plant operating conditions and plant inputs.

18. The computer program product of claim 16, wherein said indirect controller uses said model to search for a set of optimal plant inputs, said model predicting plant outputs for given plant inputs.

19. The computer program product of claim 18, wherein said search for a set of optimal plant inputs using said model includes instructions for causing the computer to:

individually move each of said plant inputs to a new position, and evaluate the resulting performance of said process in accordance with the plant outputs predicted by the model, and

if improved performance of said process results from the move to the new position, accept the new position for the plant inputs and search for a new set of optimal plant inputs.

20. The computer program product of claim 18, wherein said search for the set of optimal plant inputs is implemented as a zero or first order search method.

21. The computer program product of claim 16, wherein said model is implemented by a parametric adaptive modeling method.

22. The computer program product of claim 21, wherein said parametric adaptive modeling method includes at least one of the following:

linear regression and neural network model.

23. The computer program product of claim 22, wherein said neural network model is trained by backpropagating an error between a predicted output and an actual output.

24. The computer program product of claim 16, wherein said model is implemented by a non-parametric adaptive modeling method.

25. The computer program product of claim 24, wherein said non-parametric method includes evolutionary or genetic programming.

26. The computer program product of claim 16, wherein the instructions cause a computer to train a plurality of models using the training data.

27. The computer program product of claim 26, wherein said plurality of models have the same architectures.

28. The computer program product of claim 26, wherein said plurality of models have different architectures.

29. The computer program product of claim 16, wherein said plant uses a fossil-fuel furnace.

30. The computer program product of claim 16, wherein the instructions for causing the computer to train the model of the process includes instructions causing the computer to:

provide to said model current operating conditions and current inputs selected by the direct controller;

determine predicted plant outputs using said model;

compare the predicted plant outputs to actual plant outputs; and

establish the model as trained if the error in the predicted plant outputs is below a predetermined threshold.

31. A system for implementing indirect control in a plant operable for performing a process, the plant having a plurality of plant inputs, the inputs being adjustable according to control settings received from a controller, the system comprising:

direct control means for directly controlling operation of the plant and generating training data from the plant during operation of the plant by the direct control means;

training means for training a model of the process using the training data generated by the direct control means; and

indirect control means for indirectly controlling operation of the plant after the model has been trained, thereby replacing the direct control means.

32. The system of claim 31, wherein said training data includes plant operating conditions, plant inputs, and actual plant outputs for given plant operating conditions and plant inputs.

33. The system of claim 31, wherein said indirect control means includes search means for searching for a set of optimal plant inputs using said model, wherein said model predicts plant outputs for given plant inputs.

34. The system of claim 33, wherein said search means includes:

means for individually moving each of said plant inputs to a new position, and means for evaluating the resulting performance of said process in accordance with the plant outputs predicted by the model, and

if improved performance of said process results from the move to the new position, accepting the new position for the plant inputs and searching for a new set of optimal plant inputs.

35. The system of claim 33, wherein said search means implements a zero or first order search method.

36. The system of claim 31, wherein said model is implemented by a parametric adaptive modeling method.

37. The system of claim 36, wherein said parametric adaptive modeling method includes at least one of the following:

linear regression and neural network model.

38. The system of claim 37, wherein said neural network model is trained by said training means by backpropagating an error between a predicted output and an actual output.

39. The system of claim 31, wherein said model is implemented by a non-parametric adaptive modeling method.

40. The system of claim 39, wherein said non-parametric modeling method includes evolutionary or genetic programming.

41. The system of claim 31, wherein said training means trains a plurality of models using the training data.

42. The system of claim 41, wherein said plurality of models have the same architectures.

43. The system of claim 41, wherein said plurality of models have different architectures.

44. The system of claim 31, wherein said plant uses a fossil-fuel furnace.

45. The system of claim 31, where said training means includes:

means for providing to said model current operating conditions and current inputs selected by the direct controller, said model determining predicted plant outputs;

means for comparing the predicted plant outputs to actual plant outputs; and

means for establishing the model as trained if the error in the predicted plant outputs is below a predetermined threshold.