Modeling a process using a composite model comprising a plurality of regression models

• US 8,032,341 B2
• Filed: 01/04/2007
• Issued: 10/04/2011
• Est. Priority Date: 01/04/2007
• Status: Active Grant

• Current Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Original Assignee: Fisher-Rosemount Systems Incorporated (Emerson Electric Company)
• Inventors: Miller, John P.
• Primary Examiner(s): Rodriguez; Paul
• Assistant Examiner(s): Chad; Aniss
• Application Number:
  US11/619,846
• Publication Number:
  US 20080167839A1
• Time in Patent Office: 1,734 Days
• Field of Search: 703/2
• US Class Current: 703/2
• CPC Class Codes: C10G 11/187 Controlling or regulating
• CPC Class Codes: G05B 23/021 adopting a different treatment of each operating region or a different mode of the monitored system, e.g. transient modes; different operating configurations of monitored system
• CPC Class Codes: G05B 23/024 Quantitative history assessment, e.g. mathematical relationships between available data; Functions therefor; Principal component analysis [PCA]; Partial least square [PLS]; Statistical classifiers, e.g. Bayesian networks, linear regression or correlation analysis; Neural networks

Abstract

A system for detecting abnormal operation of at least a portion of a process plant includes a composite model for modeling at least the portion of the process plant. The model may be configurable to include multiple regression models corresponding to multiple different operating regions of the portion of the process plant. A new model may be generated from two or more of the regression models, and the composite model may be revised to replace the two or more regression models with the new model. The system may also include a deviation detector configured to determine if the actual operation of the portion of the process plant deviates significantly from the operation predicted by the composite model. If there is a significant deviation, this may indicate an abnormal operation.

Claims

1. A method for generating a model of a process in a process plant, comprising:

collecting a plurality of groups of data sets for the process in the process plant, the data sets generated from process variables of the process, the process variables including a first process variable and a second process variable;

generating, using a computing device, a plurality of regression models of the process using the groups of data sets, wherein each regression model is generated using a corresponding group of data sets from the plurality of groups, wherein each regression model models at least the first process variable as a function of at least the second process variable over a respective range of the second process variable, and wherein each regression model comprises a plurality of corresponding parameters generated using the corresponding group of data sets;

generating, using the computing device, a composite model of the process to include the plurality of regression models simultaneously;

generating, using the computing device, a new model of the process using parameters of at least two of the regression models;

determining whether the composite model includes more than a maximum number of regression models; and

if the composite model includes more than the maximum number of regression models, revising, using the computing device, the composite model of the process to replace the at least two of the regression models with the new model.

2. A method according to claim 1, wherein each group of data sets in the plurality of groups of data sets corresponds to a different operating region of the process;

wherein each regression model of the plurality of regression models corresponds to a different operating region of the process;

wherein the at least two regression models comprise regression models from adjacent operating regions.

3. A method according to claim 2, wherein the at least two regression models comprises a first regression model and a second regression model;

wherein the method further comprises generating an interpolated model between the first regression model and the second regression model in the at least two regression models;

wherein revising the composite model of the process comprises revising the composite model of the process to replace the at least two of the regression models and the interpolated model with the new model.

4. A method according to claim 1, further comprising choosing the at least two regression models from the plurality of regression models.

5. A method according to claim 4, wherein choosing the at least two regression models comprises determining the at least two regression models that will result in a new model having a smallest associated error.

6. A method according to claim 1, further comprising:

generating a plurality of new models of the process corresponding to respective groups of regression models, wherein each new model is generated using parameters of the regression models in the corresponding group of regression models;

choosing one of the new models; and

wherein revising the composite model comprises revising the composite model of the process to replace with the selected one of the new models the corresponding group of regression models.

7. A method according to claim 6, further comprising determining for each new model a corresponding indicator of difference between the new model and the corresponding group of regression models;

wherein selecting one of the new models comprises selecting a new model having the smallest corresponding indicator of difference.

8. A method according to claim 7, wherein determining for each new model the corresponding indicator of difference between the new model and the corresponding group of regression models comprises determining a corresponding integral square error between the new model and the corresponding group of regression models;

wherein selecting the new model having the smallest indicator of difference comprises selecting the new model having the smallest corresponding integral square error.

9. A non-transitory tangible medium storing machine readable instructions, the machine readable instructions, when executed by the one or more machines, causing the one or more machines to:

collect a plurality of groups of data sets for the process in the process plant, the data sets generated from process variables of the process, the process variables including a first process variable and a second process variable;

generate a plurality of regression models of the process using the groups of data sets, wherein each regression model is generated using a corresponding group of data sets from the plurality of groups, wherein each regression model models at least the first process variable as a function of at least the second process variable over a respective range of the second process variable, and wherein each model comprises a plurality of corresponding parameters generated using the corresponding group of data sets;

generate a composite model of the process to include the plurality of regression models simultaneously;

generate a new model of the process using parameters of at least two of the regression models;

determine whether the composite model includes more than a maximum number of regression models; and

if the composite model includes more than the maximum number of regression models, revise the composite model of the process to replace the at least two of the regression models with the new model.

10. A method for generating a model of a process in a process plant, comprising:

collecting data sets for the process in the process plant, the data sets generated from process variables of the process;

generating, using a computing device, a new regression model of the process using the collected data sets;

revising, using the computing device, a composite model of the process to include the new regression model in a plurality of regression models of the composite model, wherein the composite model simultaneously includes the plurality of regression models, wherein each regression model models at least a first process variable as a function of at least a second process variable over a respective range of the second process variable;

incrementing, when the composite model is revised to include the new regression model, a number N representative of the number of regression models in the composite model;

wherein each regression model in the plurality of regression models comprises a plurality of corresponding parameters generated using a corresponding group of data sets from a plurality of groups of data sets;

comparing the number N to a maximum number N_MAX of regression models that the composite model may include at a given time;

if the number N of regression models exceeds the number N_MAX, generating a new model of the process using parameters of at least two of the regression models in the plurality of regression models; and

revising, using the computing device, the composite model of the process to replace the at least two of the regression models with the new model.

11. A method according to claim 10, wherein each group of data sets in the plurality of groups of data sets corresponds to a different operating region of the process;

wherein each regression model of the plurality of regression models corresponds to a different operating region of the process;

wherein the at least two regression models comprise regression models from adjacent operating regions.

12. A method according to claim 11, wherein the at least two regression models comprises a first regression model and a second regression model;

wherein the composite model comprises an interpolated model corresponding to a region between a first operating region corresponding to the first regression model and a second operating region corresponding to the second regression model in the at least two regression models;

wherein revising the composite model to replace the at least two of the regression models with the new model comprises revising the composite model to replace the first regression model, the second regression model, and the interpolated model with the new model.

13. A method according to claim 10, further comprising choosing the at least two regression models from the plurality of regression models.

14. A method according to claim 13, wherein choosing the at least two regression models comprises determining a subset of regression models from a plurality of subsets of regression models that will result in a new model having a smallest associated error.

15. A method according to claim 10, further comprising:

generating a plurality of new models of the process corresponding to respective groups of regression models, wherein each new model is generated using parameters of the regression models in the corresponding group of regression models;

selecting one of the new models; and

wherein revising the composite model of the process to replace the at least two regression models comprises revising the composite model of the process to replace with the selected one of the new models the corresponding group of regression models.

16. A method according to claim 15, further comprising determining for each new model a corresponding indicator of difference between the new model and the corresponding group of regression models;

wherein selecting one of the new models comprises selecting a new model having the smallest corresponding indicator of difference.

17. A method according to claim 16, wherein determining for each new model the corresponding indicator of difference between the new model and the corresponding group of regression models comprises determining a corresponding integral square error between the new model and the corresponding group of regression models;

wherein selecting the new model having the smallest indicator of difference comprises selecting the new model having the smallest corresponding integral square error.

18. A non-transitory tangible medium storing machine readable instructions, the machine readable instructions, when executed by the one or more machines, causing the one or more machines to:

collect data sets for the process in the process plant, the data sets generated from process variables of the process;

generate a new regression model of the process using the collected data sets;

revise a composite model of the process to include the new regression model in a plurality of regression models of the composite model;

increment, when the composite model is revised to include the new regression model, a number N representative of the number of regression models in the composite model;

wherein each regression model in the plurality of regression models comprises a plurality of corresponding parameters generated using a corresponding group of data sets from a plurality of groups of data sets;

comparing the number N to a maximum number N_MAX of regression models that the composite model may include at a given time;

if the number N of regression models exceeds the number N_MAX, generate a new model of the process using parameters of at least two of the regression models in the plurality of regression models; and

revise the composite model of the process to replace the at least two of the regression models with the new model.

19. A system for detecting an abnormal operation in a process plant, comprising:

at least one computing device configured to implement;

a configurable composite model of the process in the process plant simultaneously comprising a plurality of a number N of regression models, each regression model having been generated using a corresponding group of data sets from a plurality of groups of data sets, the data sets generated from process variables of the process, the process variables including a first process variable and a second process variable, wherein each regression model models at least the first process variable as a function of at least the second process variable over a respective range of the second process variable, wherein each regression model comprises a plurality of corresponding parameters generated using the corresponding group of data sets;

a module operable to compare the number N to a maximum number N_MAX of regression models the composite model may simultaneously comprise;

a module to subsequently replace, if N>

N_MAX,at least two of the regression models with a new model generated using parameters of the at least two of the regression models; and

a deviation detector coupled to the configurable composite model, the deviation detector configured to determine if the process significantly deviates from an output of the composite model.

20. A system according to claim 19, wherein the module is configured to include a further regression model in the plurality of regression models.

21. A system according to claim 19, wherein the module is configured to replace with the new model at least one interpolated model and the at least two regression models.

22. A system according to claim 21, wherein the configurable composite model is configured to generate a prediction of a Y value as a function of an X value, wherein the Y value is generated from a dependent process variable and the X value is generated from an independent process variable.

23. A system according to claim 22, wherein the deviation detector is configured to determine if the Y value significantly differs from the predicted Y value.

24. A system according to claim 23, wherein the deviation detector is configured to determine if the Y value significantly differs from the predicted Y value by comparing a difference between the Y value and the predicted Y value to a threshold.

25. A system according to claim 24, wherein the threshold varies as a function of the X value.

26. A system according to claim 23, further comprising:

a first statistical generator coupled to the configurable composite model, the first statistical generator to generate the X values as a mean of the independent process variable; and

a second statistical generator coupled to the configurable composite model, the second statistical generator to generate the Y values as a mean of the dependent process variable.

27. A system according to claim 26, wherein the second statistical generator additionally generates standard deviation values of the dependent process variable;

the system further comprising an additional configurable composite model coupled to the first statistical generator, the second statistical generator, and the deviation detector, the additional configurable composite model to generate a prediction of the standard deviation of the dependent process variable as a function of the mean of the independent process variable;

wherein the deviation detector is configured to determine if the Y value significantly differs from the predicted Y value by comparing a difference between the Y value and the predicted Y value to a threshold, wherein the threshold is a function of the prediction of the standard deviation.