Method and apparatus for improving the accuracy of linear program based models

• US 7,463,937 B2
• Filed: 11/10/2005
• Issued: 12/09/2008
• Est. Priority Date: 11/10/2005
• Status: Active Grant

• Current Assignee: Todd Matson
• Original Assignee: William Joseph Korchinski
• Inventors: Korchinski, William Joseph
• Primary Examiner(s): DeCady; Albert
• Assistant Examiner(s): CHANG, SUNRAY
• Application Number:
  US11/270,792
• Publication Number:
  US 20070106399A1
• Time in Patent Office: 1,125 Days
• Field of Search: 700/29, 700/31
• US Class Current: 700/31
• CPC Class Codes: G05B 13/042 in which a parameter or coefficient is automatically adjusted to optimise the performance
• CPC Class Codes: G05B 17/02 electric

Abstract

A method and apparatus are disclosed for evaluating the performance of a Linear Program model of an operation. The significant properties of the current invention include the use of a historical record or operating data as a means to improve the evaluation of the Linear Program model quality. The historical record of operating data also allows a more reliable means for calculating corrected Linear Program model coefficients. Related applications are disclosed for viewing graphically both the uncorrected and the corrected Linear Program model'"'"'s performance. Statistical methods are disclosed as further aids in understanding both the uncorrected and corrected Linear Program model'"'"'s performance.

Claims

1. A computer-implemented method for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process comprising the steps of:

a) providing said continuous industrial process, said continuous industrial process having

at least one measured input variable having a setting value for adjusting said continuous industrial process, and

at least one measured output variable having a measurable value;

b) providing a Linear Program model of said continuous industrial process,

the Linear Program model having

a first set of independent variables having a value for each of said measured input variables,

the Linear Program model further having

a first set of dependent variables having a value for each of said measured output variables;

the Linear Program model further having

a first set of Linear Program model coefficients comprising base coefficients;

c) providing a historical record of operating conditions for said continuous industrial process, said historical record of operating conditions having

at least one set of historical values for said first set of independent variables, and

at least one corresponding set of historical values for said first set of dependent variables;

d) screening said historical record according to at least one predetermined rule,

the predetermined rule identifying screened operating data, said screened operating data comprising any of said historical record having a quality selected from the group consisting of bad and noisy,

the predetermined rule further performing upon said screened operating data an action selected from the group of discard, correct, limit, and filter operating data so identified;

e) calculating the uncorrected Linear Program predictions of said Linear Program model,

said uncorrected Linear Program predictions comprising

a first Linear Program calculation of said first set of dependent variables, said first Linear Program calculation performed using

said first set of Linear Program model coefficients, and

said at least one set of historical values for said first set of independent variables;

f) using the information contained in said uncorrected Linear Program predictions together with the corresponding set of historical values for said first set of modeled dependent variables, in order to display verbal and numerical suggested actions to improve the accuracy of said Linear Program,

whereby said method for evaluating the accuracy of a Linear Program model relative to a corresponding continuous industrial process results in substantially improved accuracy for said Linear Program model. Said Linear program model now having improved accuracy, more closely matches said continuous industrial process, thereby enabling improved operation, economic or otherwise of said continuous industrial process.

2. The method of claim 1, further comprising the steps of:

g) calculating a prediction error

said prediction error being calculated by means of taking the difference between said uncorrected Linear Program predictions and said at least one set of historical values for said first set of dependent variables;

h) plotting said prediction error for each of said first set of dependent variables relative to each of said first set of independent variables;

whereby said plotting indicates the degree of cross correlation between each of said prediction errors and each of said first set of independent variables.

3. The method of claim 2, further comprising the steps of:

i) calculating a numerical average for each of said prediction errors for each of said first set of dependent variables;

j) comparing each of said numerical averages for each of said prediction errors for each of said first set of dependent variables with statistically valid limits.

4. The method of claim 3, wherein said comparison further comprises

k) displaying said numerical average for each of said prediction errors for each of said first set of dependent variables;

l) displaying a verbal suggested action based on each of said comparisons in step k,

whereby said numerical averages enable a rapid appraisal of the accuracy of said Linear Program model.

5. The method of claim 1 wherein said first set of Linear Program model coefficients further comprises shift coefficients.

6. The method of claim 5, further comprising the steps of:

m) calculating a numerical cross correlation value for each of said prediction errors for each of said first set of dependent variables relative to each of said first set of independent variables;

n) comparing each of said numerical cross correlation values for each of said prediction errors for each of said first set of dependent variables relative to each of said first set of independent variables with statistically valid limits.

7. The method of claim 6, further comprising the steps of:

o) displaying said numerical cross correlation value for each of said prediction errors for each of said first set of dependent variables relative to each of said first set of independent variables;

p) displaying a verbal suggested action based on each of said comparisons in step o,

whereby said numerical cross correlations enable a rapid appraisal of the accuracy of said Linear Program model.

8. A computer-implemented system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process comprising:

a computer, said computer having first software, said first software comprising an implementation of said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model;

said continuous industrial process having

at least one measured input variable having a setting value for adjusting said continuous industrial process, and

at least one measured output variable having a measurable value;

said Linear Program model having

a first set of independent variables having a value for each of said measured input variables,

said Linear Program model further having

a first set of dependent variables having a value for each of said measured output variables ,and

said Linear Program model further having

a first set of Linear Program model coefficients comprising base coefficients;

a historical record of operating conditions for said continuous industrial process, said historical record of operating conditions having,

at least one set of historical values for said first set of independent variables,

at least one set of historical values for said first set of dependent variables;

software for screening said historical record according to at least one predetermined rule,

the predetermined rule identifying screened operating data, comprising any of said historical record having a quality selected from the group consisting of bad and noisy,

the predetermined rule further performing upon said screened operating data an action selected from the group of discard, correct, limit, and filter operating data so identified;

a set of uncorrected Linear Program predictions of said Linear Program model,

said set of uncorrected Linear Program predictions comprising

a first Linear Program calculation of said first set of dependent variables, said first Linear Program calculation performed using

said first set of Linear Program model coefficients

said at least one set of historical values for said first set of independent variables;

a means of comparing said uncorrected Linear Program predictions with said at least one set of historical values for said first set of dependent variables, said comparison further comprising

a display showing verbal and numerical suggested actions to improve the accuracy of said Linear Program,

whereby said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process has substantially improved accuracy of said Linear Program model. Said Linear Program model mow having improved accuracy, more closely matches said continuous industrial process, thereby enabling improved operation, economic or otherwise of said continuous industrial process.

9. The apparatus of claim 8 wherein said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process further comprises

a prediction error,

said prediction error being calculated by means of taking the difference between said uncorrected Linear Program predictions and said at least one set of historical values for said first set of dependent variables;

a plot of said prediction error for each of said first set of dependent variables relative to each of said first set of independent variables,

whereby said plot indicates the degree of cross correlation between each of said prediction errors and each of said first set of independent variables.

10. The apparatus of claim 9 wherein said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process further comprises

the calculation of a numerical average for each of said prediction errors for each of said first set of dependent variables;

a means of comparing each of said numerical averages for each of said prediction errors for each of said first set of dependent variables with statistically valid limits,

whereby said numerical averages enable a rapid appraisal of the accuracy of said Linear Program model.

11. The apparatus of claim 10 wherein said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process further comprises

a display showing said numerical average for each of said prediction errors for each of said first set of dependent variables;

a display showing a verbal suggested action for each of said comparisons of said numerical averages,

whereby said numerical averages enable a rapid appraisal of the accuracy of said Linear Program model.

12. The apparatus of claim 8 wherein said first set of Linear Program model coefficients further comprises shift coefficients.

13. The apparatus of claim 12 wherein said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process further comprises

the calculation of a numerical cross correlation value for each of said prediction errors for each of said first set of dependent variables relative to each of said first set of independent variables;

a means of comparing each of said numerical cross correlation values for each of said prediction errors for each of said first set of dependent variables relative to each of said first set of independent variables with statistically valid limits,

whereby said numerical cross correlations enable a rapid appraisal of the accuracy of said Linear Program model.

14. The apparatus of claim 13 wherein said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process further comprises

a display showing said numerical cross correlation value for each of said prediction errors for each of said first set of dependent variables relative to each of said first set of independent variables;

a display showing a verbal suggested action for each of said comparisons of said numerical cross correlations,

whereby said numerical cross correlations enable a rapid appraisal of the accuracy of said Linear Program model.

15. A method for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process comprising the steps of:

a) providing said continuous industrial process, said continuous industrial process having

at least one measured input variable having a setting value for adjusting said continuous industrial process, and

at least one measured output variable having a measurable value;

b) providing a Linear Program model of said continuous industrial process,

the Linear Program model having

a first set of independent variables having a value for each of said measured input variables,

said Linear Program model further having

a first set of dependent variables having a value for each of said measured output variables

said Linear Program model further having

a first set of model coefficients comprising base coefficients;

c) providing a calibrated reference model of said continuous industrial process,

the calibrated reference model having

a second set of independent variables having a value for each of said measured input variables,

the calibrated reference model further having

a second set of dependent variables having a value for each of said measured output variables;

a set of model parameters;

d) defining a scope for said calibrated reference model,

said scope representing

at least a portion of said continuous industrial process,

said scope comprising

a third set of independent variables consisting of at least a portion of said first set of independent variables,

wherein at least one of said third set of independent variables corresponds to one of said at least one measured input variable, and

said scope further comprising

a third set of dependent variables consisting of at least a portion of said first set of dependent variables;

wherein at least one of said third set of dependent variables corresponds to one of said at least one measured output variable;

e) providing a historical record of operating conditions for said continuous industrial process, said historical record of operating conditions having,

at least one set of historical values for said first set of independent variables,

at least one set of historical values for said first set of dependent variables;

f) screening said historical record according to at least one predetermined rule,

the predetermined rule identifying screened operating data, comprising any of said historical record having a quality selected from the group consisting of bad and noisy,

the predetermined rule further performing upon said screened operating data an action selected from the group of discard, correct, limit, and filter operating data so identified;

g) using said calibrated reference model to calculate a calibrated reference model prediction

said calibrated reference model prediction further consisting of

a set of values for each of said third set of dependent variables calculated by said calibrated reference model

said calibrated reference model prediction further being calculated at

each point in said historical record;

h) calculating a calibrated reference model prediction error,

said calibrated reference model prediction error being equal to the sum of the squares of the differences between said calibrated reference model predictions and each corresponding of said at least one set of historical values for said first set of dependent variables;

i) producing a recalibrated reference model by selecting said recalibrated reference model parameters so as to minimize said recalibrated reference model prediction error;

j) calculating new Linear Program-model coefficients,

said calculation further comprising

use of said recalibrated reference model to evaluate the change in each of said third set of dependent variables relative to each of said third set of independent variables,

whereby said method for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to continuous industrial process provides the best match between said Linear Program model and said historical record.

16. The method of claim 15, further comprising the steps of:

replacing said calibrated reference model with a calibrated nonlinear model.

17. The method of claim 16, further comprising the steps of:

replacing said calibrated reference model with a Linear Program model.

18. The method of claim 17, further comprising the steps of:

k) Calculating corrected Linear Program predictions of said Linear Program model,

said corrected Linear Program predictions comprising

a second Linear Program calculation of said third set of dependent variables, said second Linear Program calculation performed using

said new Linear Program model coefficients

said at least one set of historical values for said third set of independent variables;

l) comparing said corrected Linear Program predictions with said at least one set of historical values for said third set of dependent variables,

whereby said display shows the accuracy of said corrected Linear Program predictions relative to said historical record.

19. The method of claim 18, further comprising the steps of:

m) calculating a prediction error

said prediction error being calculated by means of taking the difference between said corrected Linear Program predictions and said at least one set of historical values for said third set of dependent variables;

n) plotting said prediction error for each of said third set of dependent variables relative to each of said third set of independent variables,

whereby said plotting indicates the degree of cross correlation between each of said third set of prediction errors and each of said first set of independent variables.

20. The method of claim 19, further comprising the steps of:

o) calculating a numerical average for each of said prediction errors for each of said third set of dependent variables;

p) comparing each of said numerical averages for each of said prediction errors for each of said third set of dependent variables with statistically valid limits.

21. The method of claim 20, further comprising the steps of:

q) displaying said numerical average for each of said prediction errors for each of said third set of dependent variables;

r) displaying a verbal suggested action based on each of said comparisons in step s.

22. The method of claim 15 wherein said first set of Linear Program model coefficients further comprises shift coefficients.

23. The method of claim 22, further comprising the steps of:

s) calculating a numerical cross correlation value for each of said prediction errors for each of said third set of dependent variables relative to each of said third set of independent variables;

t) comparing each of said numerical cross correlation values for each of said prediction errors for each of said third set of dependent variables relative to each of said third set of independent variables with statistically valid limits.

24. The method of claim 23, further comprising the steps of:

u) displaying said numerical cross correlation value for each of said prediction errors for each of said third set of dependent variables relative to each of said third set of independent variables;

v) displaying a verbal suggested action based on each of said comparisons in step u.

25. A computer-implemented system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process comprising:

a computer, said computer having first software, said first software comprising an implementation of said system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to a corresponding continuous industrial process,

said computer further having second software, said second software comprising a calibrated reference model of said continuous industrial process;

said continuous industrial process having

at least one measured input variable having a setting value for adjusting said continuous industrial process, and

at least one measured output variable having a measurable value;

said Linear Program model having

a first set of independent variables having a value for each of said measured input variables,

said Linear Program model further having

a first set of dependent variables having a value for each of said measured output variables

said Linear Program model further having

a first set of Linear Program model coefficients comprising

base coefficients;

a) providing said continuous industrial process, said continuous industrial process having

at least one measured input variable having a setting value for adjusting said continuous industrial process, and

at least one measured output variable having a measurable value;

b) providing a Linear Program model of said continuous industrial process,

said Linear Program model having

a first set of independent variables having a value for each of said measured input variables

the Linear Program model further having

a first set of dependent variables having a value for each of said measured output variables

the Linear Program model further having

a first set of model coefficients comprising

base coefficients;

c) providing a calibrated reference model of said continuous industrial process,

the calibrated reference model having

a second set of independent variables having a value for each of said measured input variables,

the calibrated reference model further having

a second set of dependent variables having a value for each of said measured output variables;

a set of model parameters;

d) defining a scope for said calibrated reference model,

said scope representing

at least a portion of said continuous industrial process,

said scope comprising

a third set of independent variables consisting of at least a portion of said first set of independent variables,

wherein at least one of said third set of independent variables corresponds to one of said at least one measured input variable, and

said scope further comprising

a third set of dependent variables consisting of at least a portion of said first set of dependent variables;

wherein at least one of said third set of dependent variables corresponds to one of said at least one measured output variable;

e) providing a historical record of operating conditions for said continuous industrial process, said historical record of operating conditions having,

at least one set of historical values for said first set of independent variables,

at least one set of historical values for said first set of dependent variables;

f) screening said historical record according to at least on predetermined rule,

the predetermined rule identifying screened operating data, comprising any of said historical record having a quality selected from the group consisting of bad and noisy,

the predetermined rule further performing upon said screened operating data an action selected from the group of discard, correct, limit, and filter operating data so identified;

g) using said calibrated reference model to calculate a calibrated reference model prediction

said calibrated reference model prediction further consisting of

a set of values for each of said third set of dependent variables calculated by said calibrated reference model

said calibrated reference model prediction further being calculated at each point in said historical record;

h) calculating a calibrated reference model prediction error,

said calibrated reference model prediction error being equal to the sum of the squares of the differences between said calibrated reference model predictions and each corresponding of said at least one set of historical values for said first set of dependent variables;

i) producing a recalibrated reference model by selecting said recalibrated reference model parameters so as to minimize set recalibrated reference model prediction error;

j) calculating new Linear Program model coefficients,

said calculation further comprising

use of said recalibrated reference model to evaluate the change in each of said third set of dependent variables relative to each of said third set of independent variables,

whereby said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to continuous industrial process provides the best match between said Linear Program model and said historical record. Said Linear program model now having improved accuracy, more closely matches said continuous industrial process, thereby enabling improved operation, economic or otherwise of said continuous industrial process.

26. The apparatus of claim 25 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy of a Linear Program model relative to continuous industrial process further comprises

replacing said calibrated reference model with a calibrated nonlinear model.

27. The apparatus of claim 26 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

replacing said calibrated reference model with a Linear Program model.

28. The apparatus of claim 27 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

a set of corrected Linear Program predictions of said Linear Program model,

said corrected Linear Program predictions comprising

a second Linear Program calculation of said third set of dependent variables, said second Linear Program calculation performed using

said new Linear Program base and shift model coefficients

said at least one set of historical values for said third set of independent variables, and

a comparison of said corrected Linear Program predictions with said at least one set of historical values for said third set of dependent variables.

29. The apparatus of claim 28 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

a calculated prediction error

said prediction error being calculated by means of taking the difference between said corrected Linear Program predictions and said at least one set of historical values for said third set of dependent variables, and

a set of plots of said prediction error for each of said third set of dependent variables relative to each of said third set of independent variables.

30. The apparatus of claim 29 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

a calculation of a numerical average for each of said prediction errors for each of said third set of dependent variables,

a comparison of each of said numerical averages for each of said prediction errors for each of said third set of dependent variables with statistically valid limits.

31. The apparatus of claim 30 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

a display of said numerical average for each of said prediction errors for each of said third set of dependent variables,

a verbal suggested action based on each of said displays of said numerical averages.

32. The apparatus of claim 25 wherein said first set of Linear Program model coefficients further comprises shift coefficients.

33. The apparatus of claim 32 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

a calculation of a numerical cross correlation value for each of said prediction errors for each of said third set of dependent variables relative to each of said third set of independent variables;

a comparison of each of said numerical cross correlation values for each of said prediction errors for each of said third set of dependent variables relative to each of said third set of independent variables with statistically valid limits.

34. The apparatus of claim 33 wherein said computer-implemented system for enabling the rapid appraisal of and for improving the accuracy a Linear Program model relative to continuous industrial process further comprises

a display of said numerical cross correlation value for each of said prediction errors for each of said third set of dependent variables relative to each of said third set of independent variables;

a verbal suggested action based on each of said a display of said numerical cross correlation values.