System and method for automatically determining a set of variables for use in creating a process model

• US 5,727,128 A
• Filed: 05/08/1996
• Issued: 03/10/1998
• Est. Priority Date: 05/08/1996
• Status: Expired due to Term

• Current Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Original Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Inventors: Morrison, Steven Michael
• Primary Examiner(s): HAFIZ, TARIQ R
• Assistant Examiner(s): HAFIZ, TARIQ R
• Application Number:
  US08/646,919
• Time in Patent Office: 671 Days
• Field of Search: 358/75, 395/22, 395/26, 395/76, 395/68, 395/11, 395/21
• US Class Current: 706/45
• CPC Class Codes: G05B 17/02 electric
• CPC Class Codes: G06N 3/08 Learning methods

Abstract

A process modeling system and method develop a set of process model inputs for a process model, such as a neural network, from values for a number of process input variables and at least one process output variable. The system and method first determine a correlation measurement between each of the process input variables and the process output variable and select a set of potential model input variables based on the correlation measurements. The system and method then iteratively determine a succession of sets of potential model input variables by performing a regression analysis on the selected set of potential model input variables and the model output variable and by then refining the set of potential model input variables based on the result of the regression analysis and on the correlation measurements. After a number of iterations, the system and method choose a set of potential model input variables as the set of model inputs and develop a process model from the chosen set of model inputs.

Claims

1. A computer program embodied on a computer readable medium for implementation on a computer that automatically develops a set of actual model input variables using values for a multiplicity of process input variables and at least one process output variable, comprising the steps of:

determining a correlation measurement between each of the multiplicity of process input variables and the at least one process output variable;

automatically selecting a set of potential model input variables from the multiplicity of process input variables based on the correlation measurements;

performing a regression analysis on the set of potential model input variables to produce a regression analysis result for the set of potential model input variables; and

choosing the set of potential model input variables as the set of actual model input variables based on the regression analysis result;

wherein the selected set of potential model input variables includes less than all of the multiplicity of process input variables.

2. The computer program according to claim 1, further performing the steps of picking a new set of potential model input variables and providing the new set of potential model input variables for use in the step of performing as the set of potential model input variables.

3. The computer program according to claim 2, wherein the step of performing calculates a sensitivity measurement related to the correlation between each of the set of potential model input variables and the at least one process output variable and wherein the step of picking includes a step of selecting the new set of potential model input variables based on the sensitivity measurement.

4. The computer program according to claim 2, wherein the step of picking includes a step of selecting the new set of potential model input variables from the multiplicity of process input variables using the correlation measurements.

5. The computer program according to claim 4, wherein the step of choosing includes a step of comparing the regression analysis result for the set of potential model input variables to a predetermined threshold.

6. The computer program according to claim 1, wherein the step of determining includes a step of calculating a linear correlation coefficient between one of the process input variables and the process output variable.

7. The computer program according to claim 6, wherein the step of calculating calculates the linear correlation coefficient between an input variable A and an output variable O from records of data pertaining to the input and output variables according to the equation:

##EQU23## wherein;

CC_A,O =the correlation coefficient between the input variable A and the output variable O;

n=the number of records for which the values of the variables A and O are determined;

A_i =the ith value of the input variable A;

O_i =the ith value of the output variable O;

Avar=the variance of the input variable A over the n records; and

Ovar=the variance of the output variable O over the n records.

8. The computer program according to claim 1, wherein the step of determining includes a step of calculating a non-linear correlation coefficient between one of the process input variables and the at least one process output variable.

9. The computer program according to claim 1, wherein the step of determining includes a step of time aligning data from a plurality of runs of a batch process.

10. The computer program according to claim 9, wherein the step of determining includes a step of scaling data from a plurality of runs of a batch process to have zero mean and unit variance.

11. The computer program according to claim 1, wherein the step of selecting selects the set of potential model input variables as a predetermined number of the process input variables having the highest correlation measurements associated therewith.

12. The computer program according to claim 1, wherein the step of selecting selects the set of potential model input variables as all of the process input variables having an associated correlation measurement greater than a predetermined threshold.

13. The computer program according to claim 1, wherein the step of performing includes a step of implementing a partial least squares analysis on the set of potential model input variables.

14. The computer program according to claim 1, further performing a step of producing a process model using the set of actual model input variables.

26. The computer program of claim 1, wherein the process input variables comprise batch process input variables and wherein the step of selecting selects a variable type and a time associated with each of the potential model input variables.

15. A process modeling system that automatically develops a set of actual model input variables using values for a multiplicity of process input variables and at least one process output variable, comprising:

a data processor including;

a memory for storing the values for the multiplicity of process input variables and at least one process output variable;

means for determining a correlation measurement between each of the multiplicity of process input variables and the at least one process output variable;

means for selecting a set of potential model input variables from the multiplicity of process input variables; and

means for developing the set of actual model input variables from the selected set of potential model input variables including;

means for performing a regression analysis on the set of potential model input variables to produce a regression analysis result for the set of potential model input variables;

means for automatically creating a new set of potential model input variables;

means for providing the new set of potential model input variables to the performing means as the set of potential model input variables; and

means for choosing one of the sets of potential model input variables developed by the selecting means or the creating means as the set of actual model input variables based on a selection criterion related to the regression analysis result;

wherein one of the selecting means and the creating means develops one of the sets of potential model input variables based on the correlation measurements and wherein at least one of the sets of potential model input variables includes less than all of the multiplicity of process input variables.

16. The process modeling system according to claim 15, wherein the selecting means selects the set of potential model input variables from the multiplicity of process input variables based on the correlation measurements.

17. The process modeling system according to claim 15, wherein the determining means includes means for calculating a non-linear correlation coefficient between one of the process input variables and the process output variable.

18. The process modeling system according to claim 15, wherein the performing means includes means for implementing a partial least squares analysis on the set of potential model input variables.

19. The process modeling system according to claim 18, wherein the partial least squares implementing means includes means for generating a set of sensitivity measurements related to the correlation between the set of potential model input variables and the at least one process output variable and the creating means creates the new set of potential model input variables based on the sensitivity measurements.

20. The process modeling system according to claim 15, further including means for generating a process model from the set of actual model input variables.

27. The process modeling system of claim 15, wherein the process input variables comprise batch process input variables and wherein the creating means selects a variable type and a time associated with each of the new set of potential model input variables.

21. A method for implementation by a computer program embodied on a computer readable medium used with a computer for developing a process model from values for each of a multiplicity of process input variables and at least one process output variable, comprising the steps of:

storing the values for each of the multiplicity of process input variables and the at least one process output variable on a computer readable memory;

determining a correlation measurement between each of the multiplicity of process input variables and the process output variable from the stored values;

automatically selecting a set of potential model input variables from the multiplicity of process input variables based on the correlation measurements;

performing a regression analysis on the set of potential model input variables to produce a regression analysis result for the set of potential model input variables;

choosing the set of potential model input variables as a set of actual model input variables based on the regression analysis result; and

using the values associated with the chosen set of actual model input variables to produce the process model;

wherein the selected set of potential model input variables includes less than all of the multiplicity of process input variables.

22. The method of developing a process model according to claim 21, further including the steps of selecting a succession of different sets of potential model input variables and performing a regression analysis on each of the succession of different sets of potential model input variables to develop a regression analysis result associated with each of the succession of different sets of potential model input variables and wherein the step of choosing includes the step of choosing one of the succession of sets of potential model input variables as the set of actual model input variables based on the regression analysis results.

23. The method of developing a process model according to claim 22, wherein the step of selecting the succession of sets of potential model input variables selects a new set of potential model input variables based on the correlation measurements.

24. The method of developing a process model according to claim 21, wherein the step of determining includes the step of calculating a non-linear correlation coefficient between one of the process input variables and the process output variable.

25. The method of developing a process model according to claim 21, wherein the step of performing a regression analysis includes the step of implementing a partial least squares analysis.