Method and apparatus for training a system model including an integrated sigmoid function

• US 7,213,006 B2
• Filed: 11/04/2005
• Issued: 05/01/2007
• Est. Priority Date: 05/06/1996
• Status: Expired due to Fees

• Current Assignee: Rockwell Automation Technologies Incorporated (Rockwell Automation, Inc.)
• Original Assignee: Pavilion Technologies Incorporated (Rockwell Automation, Inc.)
• Inventors: Piche, Stephen, Hartman, Eric Jon, Gerules, Mark
• Primary Examiner(s): Hirl; Joseph P
• Application Number:
  US11/267,812
• Publication Number:
  US 20060074501A1
• Time in Patent Office: 543 Days
• Field of Search: 706/16, 706/15, 706/23, 700/29
• US Class Current: 706/16
• CPC Class Codes: G05B 13/027 using neural networks only
• CPC Class Codes: G05B 13/042 in which a parameter or coefficient is automatically adjusted to optimise the performance
• CPC Class Codes: G05B 13/048 using a predictor
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G06N 3/02 Neural networks

Abstract

Method and apparatus for training a system model with gain constraints. A method is disclosed for training a steady-state model having an input and an output and a mapping layer for mapping the input to the output, the model comprising a stored representation of a plant or process, and including a linear portion and a non-linear portion, where the non-linear portion includes a function. Input is received to the model, and predicted output computed corresponding to attribute(s) of the plant or process. The predicted output is stored, and is usable to manage the plant or process. The model is trained to optimize a specified objective function subject to one or more constraints, e.g., via a non-linear programming (NLP) optimizer, the constraints including, hard constraint(s) comprising strict limitations on the training in optimizing the objective function, and/or soft constraint(s) comprising a weighted penalty function included in the objective function.

Claims

1. A system for managing a plant or process, comprising:

a computer system, comprising;

a processor; and

a memory medium coupled to the processor, wherein the memory stores program instructions executable by the processor to implement a model, wherein the model comprises a representation of a plant or process, and wherein the plant or process comprises linear and non-linear aspects;

an input, coupled to the computer system, wherein the input is operable to receive an input vector comprising one or more inputs for the model, each input representing an attribute of the plant or process; and

an output, coupled to the computer system;

wherein the model comprises;

a linear portion, representing the linear aspects of the plant or process; and

a non-linear portion, representing the non-linear aspects of the plant or process, wherein the non-linear portion comprises a function, wherein for each of the one or more inputs, x_i, the function is of the form;

2. The system of claim 1, wherein the model is operable to be trained by a training process, wherein the training process is operable to:

a) receive training input data representative of inputs to the plant or process;

b) execute the model to generate predicted training output data for the plant or process corresponding to the training input data;

c) compare the predicted training output data to training output data representative of output of the plant or process to determine an error;

d) update the model in accordance with the determined error; and

e) perform a)–

d) in an iterative manner to minimize the error.

3. The system of claim 2, wherein, to train the model, the training process is operable to train the model to optimize a specified objective function.

4. The system of claim 3, wherein the training process is operable to train the model to optimize the specified objective function subject to one or more constraints.

5. The system of claim 4, wherein the system further comprises an optimizer coupled to the model, wherein the one or more constraints comprise one or more of:

one or more hard constraints, wherein the one or more hard constraints comprise strict limitations on the training process in optimizing the objective function; and

/or

one or more soft constraints, wherein the one or more soft constraints comprise a weighted penalty function comprised in the objective function.

6. The system of claim 5, wherein the optimizer comprises a non-linear programming (NLP) optimizer.

7. The system of claim 4,

wherein the model comprises one or more gains, wherein each gain comprises a partial derivative of an output with respect to a respective input, wherein the one or more constraints comprise one or more gain constraints.

8. The system of claim 7, wherein each of the one or more inputs has a lower bound and an upper bound, wherein the gain constraints comprise respective constraints on the gain at the lower bound and/or the upper bound of the respective input.

9. The system of claim 4, wherein the one or more constraints comprise one or more soft constraints, and wherein the training process is operable to train the model via gradient descent.

10. The system of claim 1, wherein the model comprises a neural network.

11. The system of claim 1, wherein the system further comprises:

a module coupled to the plant or process, wherein the module is operable to;

receive plant or process input data;

provide the plant or process input data to the model as input;

execute the model to generate predicted plant or process output based on the plant or process input data; and

wherein the predicted plant or process output is useable to manage the plant or process.

12. The system of claim 1, wherein the system further comprises:

an optimizer coupled to the plant or process;

wherein the optimizer is operable to utilize the model for optimization of the plant or process, wherein, to utilize the model for optimization of the plant or process, the optimizer is operable to;

receive desired plant or process output;

use the desired plant or process output to generate optimized plant or process inputs;

provide the optimized plant or process inputs to the plant or process to operate the plant or process to produce optimized plant or process output substantially in accordance with the desired plant or process output.

13. The system of claim 12, wherein, to optimize the plant or process, the optimizer is operable to optimize a specified objective function.

14. The system of claim 13, wherein the optimizer is operable to optimize the specified objective function subject to one or more constraints.

15. The system of claim 14, wherein the one or more constraints comprise one or more of:

one or more hard constraints, wherein the one or more hard constraints comprise strict limitations on the training process in optimizing the objective function; and

/or

one or more soft constraints, wherein the one or more soft constraints comprise a weighted penalty function comprised in the objective function.

16. The system of claim 15, wherein the optimizer comprises a non-linear programming (NLP) optimizer.

17. The system of claim 14, wherein the one or more constraints comprise one or more soft constraints, and wherein the optimizer is operable to optimize the plant or process via gradient descent.

18. A computer-accessible memory medium that stores program instructions implementing a model of a plant or process, wherein the program instructions are executable by a processor to perform:

receiving an input vector comprising one or more inputs, each input representing an attribute of a plant or process, wherein the plant or process comprises linear and non-linear aspects, wherein the model comprises a linear portion representing the linear aspects of the plant or process, and a non-linear portion, representing the non-linear aspects of the plant or process, and wherein the non-linear portion comprises a function, and wherein for each of the one or more inputs, x_i, the function is of the form;

19. The memory medium of claim 18, wherein the program instructions are further executable to perform:

executing an optimizer coupled to the plant or process to utilize the model for optimization of the plant or process, wherein the optimizer is executable to perform;

receiving desired plant or process output;

using the desired plant or process output to generate optimized plant or process inputs;

providing the optimized plant or process inputs to the plant or process to operate the plant or process to produce optimized plant or process output substantially in accordance with the desired plant or process output.

20. The memory medium of claim 18, wherein the model comprises a neural network.

21. A method for managing a plant or process, comprising:

receiving one or more inputs to a model, each input representing an attribute of a plant or process, wherein the process comprises linear aspects and non-linear aspects, wherein the model comprises a representation of the plant or process, wherein the model comprises a linear portion representing the linear aspects of the plant or process, and a non-linear portion representing the non-linear aspects of the plant or process, and wherein the non-linear portion comprises a function, and wherein for each of the one or more inputs, x_i, the function is of the form;

22. The method of claim 21, further comprising:

an optimizer coupled to the plant or process executing to utilize the model for optimization of the plant or process, comprising;

receiving desired plant or process output;

using the desired plant or process output to generate optimized plant or process inputs;

providing the optimized plant or process inputs to the plant or process to operate the plant or process to produce optimized plant or process output substantially in accordance with the desired plant or process output.

23. The method of claim 21, wherein the model comprises a neural network.