Apparatus and method for analyzing model quality in a process control environment

• US 20070112442A1
• Filed: 11/17/2005
• Published: 05/17/2007
• Est. Priority Date: 11/17/2005
• Status: Active Grant

• Current Assignee: Honeywell International Inc.
• Original Assignee: Honeywell International Inc.
• Inventors: Lu, Joseph, Zhan, Charles
• Granted Patent: US 7,421,374 B2
• Time in Patent Office: Days
• US Class Current: 700/29
• CPC Class Codes: G05B 17/02 electric

Abstract

A method includes identifying a signal and a disturbance associated with a model. The signal and disturbance are identified using historical data associated with one or more process variables. The method also includes decomposing the signal and the disturbance at a plurality of resolution levels. The method further includes extracting a plurality of data segments from the signal using the decomposed signal and the decomposed disturbance. In addition, the method includes determining a quality of the model using the extracted data segments and at least a portion of the historical data.

Claims

1. A method, comprising:

identifying a signal and a disturbance associated with a model, the signal and disturbance identified using historical data associated with one or more process variables;

decomposing the signal and the disturbance at a plurality of resolution levels;

extracting a plurality of data segments from the signal using the decomposed signal and the decomposed disturbance; and

determining a quality of the model using the extracted data segments and at least a portion of the historical data.

2. The method of claim 1, wherein:

the model is associated with one or more controlled variables and one or more manipulated variables, the one or more controlled variables at least partially controllable via the one or more manipulated variables; and

the historical data associated with the one or more process variables comprises at least one of;

one or more measurements associated with the one or more controlled variables; and

one or more measurements associated with the one or more manipulated variables.

3. The method of claim 2, wherein:

the model comprises at least one submodel, each submodel associated with one controlled variable and one manipulated variable;

the signal comprises a signal for each submodel, the signal for a particular submodel comprising a prediction associated with that submodel; and

the disturbance comprises a disturbance for each submodel, the disturbance for a particular submodel comprising a sum of predictions associated with other submodels.

4. The method of claim 1, wherein extracting the plurality of data segments comprises:

identifying a plurality of points associated with the decomposed signal;

selecting at least some of the points; and

extracting data segments associated with the selected points.

5. The method of claim 4, wherein:

the points comprise local maximum and local minimum values; and

each selected point is associated with at least one of;

a signal-to-noise ratio above a first threshold and a noise-to-signal ratio below a second threshold.

6. The method of claim 1, wherein determining the quality of the model comprises:

determining a first overall predictability index for the model; and

determining the quality of the model based on the first overall predictability index.

7. The method of claim 6, wherein determining the first overall predictability index comprises:

determining a predictability for each extracted data segment;

determining a predictability index for each of at least one resolution level, the predictability index for a particular resolution level based on the predictabilities for the extracted data segments associated with that resolution level; and

determining the first overall predictability index based on the predictability indexes for at least one of the resolution levels.

8. The method of claim 7, further comprising determining a second overall predictability index for the model, the quality of the model based on the first and second overall predictability indexes, the second overall predictability index determined by:

determining a gain multiplier;

determining a model prediction using the gain multiplier; and

determining the second overall predictability index using the model prediction determined using the gain multiplier.

9. The method of claim 7, wherein:

the predictability index for the particular resolution level is based on a weighted sum of the predictabilities for the extracted data segments associated with that resolution level; and

the first overall predictability index is based on a weighted sum of the predictability indexes for at least two of the resolution levels.

10. An apparatus, comprising:

at least one memory capable of storing historical data associated with one or more process variables; and

at least one processor capable of;

identifying a signal and a disturbance associated with a model using the historical data;

decomposing the signal and the disturbance at a plurality of resolution levels;

extracting a plurality of data segments from the signal using the decomposed signal and the decomposed disturbance; and

determining a quality of the model using the extracted data segments and at least a portion of the historical data.

11. The apparatus of claim 10, wherein:

the model comprises at least one submodel, each submodel associated with one controlled variable and one manipulated variable, the controlled variable at least partially controllable via the manipulated variable;

the signal comprises a signal for each submodel, the signal for a particular submodel comprising a prediction associated with that submodel; and

the disturbance comprises a disturbance for each submodel, the disturbance for a particular submodel comprising a sum of predictions associated with other submodels.

12. The apparatus of claim 10, wherein the at least one processor is capable of extracting the plurality of data segments by:

identifying a plurality of points associated with the decomposed signal;

selecting at least some of the points; and

extracting data segments associated with the selected points.

13. The apparatus of claim 10, wherein the at least one processor is capable of determining the quality of the model by:

determining a first overall predictability index for the model; and

determining the quality of the model based on the first overall predictability index.

14. The apparatus of claim 13, wherein the at least one processor is capable of determining the first overall predictability index by:

determining a predictability for each extracted data segment;

determining a predictability index for each of at least one resolution level, the predictability index for a particular resolution level based on the predictabilities for the extracted data segments associated with that resolution level; and

determining the first overall predictability index based on the predictability indexes for at least one of the resolution levels.

15. The apparatus of claim 14, wherein the at least one processor is further capable of determining a second overall predictability index for the model, the quality of the model based on the first and second overall predictability indexes, the second overall predictability index determined by:

determining a gain multiplier;

determining a model prediction using the gain multiplier; and

determining the second overall predictability index using the model prediction determined using the gain multiplier.

16. The apparatus of claim 14, wherein:

the predictability index for the particular resolution level is based on a weighted sum of the predictabilities for the extracted data segments associated with that resolution level; and

the first overall predictability index is based on a weighted sum of the predictability indexes for at least two of the resolution levels.

17. A computer program embodied on a computer readable medium and operable to be executed by a processor, the computer program comprising computer readable program code for:

identifying a signal and a disturbance associated with a model, the signal and disturbance identified using historical data associated with one or more process variables;

decomposing the signal and the disturbance at a plurality of resolution levels;

extracting a plurality of data segments from the signal using the decomposed signal and the decomposed disturbance; and

determining a quality of the model using the extracted data segments and at least a portion of the historical data.

18. The computer program of claim 17, wherein the computer readable program code for determining the quality of the model comprises computer readable program code for:

determining a first overall predictability index for the model; and

determining the quality of the model based on the first overall predictability index.

19. The computer program of claim 18, wherein the computer readable program code for determining the first overall predictability index comprises computer readable program code for:

determining a predictability for each extracted data segment;

determining a predictability index for each of at least one resolution level, the predictability index for a particular resolution level based on a weighted sum of the predictabilities for the extracted data segments associated with that resolution level; and

determining the first overall predictability index based on a weighted sum of the predictability indexes for at least one of the resolution levels.

20. The computer program of claim 19, wherein the computer program further comprises computer readable program code for determining a second overall predictability index, the quality of the model based on the first and second overall predictability indexes, the second overall predictability index determined by:

determining a gain multiplier;

determining a model prediction using the gain multiplier; and

determining the second overall predictability index using the model prediction determined using the gain multiplier.