Adaptive model training system and method

• US 8,892,478 B1
• Filed: 03/14/2013
• Issued: 11/18/2014
• Est. Priority Date: 11/30/2007
• Status: Active Grant

• Current Assignee: Intellectual Assets LLC
• Original Assignee: Intellectual Assets LLC
• Inventors: Bickford, Randall L., Palnitkar, Rahul M.
• Primary Examiner(s): Gaffin, Jeffrey A
• Assistant Examiner(s): Sugent, James F
• Application Number:
  US13/803,436
• Time in Patent Office: 614 Days
• Field of Search: 706/12, 706/16, 706/22, 706/25, 700/28, 700/47, 702/85, 702/189, 702/190, 374 1- 3
• US Class Current: 706/12
• CPC Class Codes: G01K 15/005 Calibration
• CPC Class Codes: G01N 21/274 Calibration, base line adjustment, drift correction
• CPC Class Codes: G01R 35/005 Calibrating; Standards or reference devices, e.g. voltage or resistance standards, "golden" references G01R33/0035, G01R35/002 take precedence
• CPC Class Codes: G05B 13/02 electric
• CPC Class Codes: G05B 23/0235 based on a comparison with predetermined threshold or range, e.g. "classical methods", carried out during normal operation; threshold adaptation or choice; when or how to compare with the threshold
• CPC Class Codes: G06F 17/10 Complex mathematical operations function generation by table look-up G06F1/03; evaluation of elementary functions by calculation G06F7/544
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/004 Artificial life, i.e. computing arrangements simulating life
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 5/02 Knowledge representation; Symbolic representation
• CPC Class Codes: G06N 5/04 Inference or reasoning models
• CPC Class Codes: G06Q 50/06 Energy or water supply
• CPC Class Codes: H02J 3/00 Circuit arrangements for ac mains or ac distribution networks

Abstract

An adaptive model training system and method for filtering asset operating data values acquired from a monitored asset for selectively choosing asset operating data values that meet at least one predefined criterion of good data quality while rejecting asset operating data values that fail to meet at least the one predefined criterion of good data quality; and recalibrating a previously trained or calibrated model having a learned scope of normal operation of the asset by utilizing the asset operating data values that meet at least the one predefined criterion of good data quality for adjusting the learned scope of normal operation of the asset for defining a recalibrated model having the adjusted learned scope of normal operation of the asset.

Claims

1. A computer-implemented adaptive model training method, said method comprising the steps of:

providing a set of selectively chosen good data values acquired from operation of an asset wherein the provided set of selectively chosen good data values acquired from operation of the asset are selectively chosen based on at least one predefined criterion of good data quality;

separating the provided set of selectively chosen good data values acquired from operation of the asset based on at least one predefined criterion for assigning each provided selectively chosen good data value to at least one group of separated selectively chosen good data values;

selecting representative selectively chosen good data values from each of the at least one group of separated selectively chosen good data values based on at least one predefined criterion for sampling at least one representative selectively chosen good data value from each of the at least one group of separated selectively chosen good data values;

forming a reduced set of data values by combining the selected representative selectively chosen good values from each of the at least one group of separated selectively chosen good data values; and

storing the reduced set of data values for defining a learned scope of operation of the asset.

2. The computer-implemented method of claim 1 further comprising a step of calibrating a model by utilizing at least a portion of the stored reduced set of data values for defining a calibrated model having a learned scope of operation of the asset.

3. The computer-implemented method of claim 1 further comprising the steps of:

providing a previously calibrated model having a learned scope of operation of the asset; and

recalibrating the previously calibrated model by utilizing at least a portion of the stored reduced set of data values for defining a recalibrated model having an adjusted learned scope of operation of the asset.

4. The computer-implemented method of claim 1 wherein the step of separating the provided set of selectively chosen good data values into the at least one group of separated selectively chosen good data values is further comprised of separating the provided set of selectively chosen good data values into at least one cluster of selectively chosen good data values wherein each of the at least one cluster of selectively chosen good data values has an approximately Gaussian distribution about a cluster center.

5. The computer-implemented method of claim 1 wherein the step of selecting representative selectively chosen good data values from each of the at least one group of separated selectively chosen good data values based on the at least one predefined criterion for sampling at least one representative selectively chosen good data value from each of the at least one group of separated selectively chosen good data values is further comprised of:

providing a predefined total number of samples required in the reduced set of data values; and

selecting a number of samples from each of the at least one group of separated selectively chosen good data values proportionate to a percentile of a total number of the selectively chosen good data values included in each of the at least one group of separated selectively chosen good data values so that the predefined total number of samples required in the reduced set of data values are selected.

6. The computer-implemented method of claim 1 wherein the step of selecting representative selectively chosen good data values from each of the at least one group of separated selectively chosen good data values based on the at least one predefined criterion for sampling at least one representative selectively chosen good data value from each of the at least one group of separated selectively chosen good data values is further comprised of:

sorting the selectively chosen good data values in each of the at least one group of separated selectively chosen good data values in accordance with a criterion of similarity of each included selectively chosen good data value in each of the at least one group of separated selectively chosen good data values to a respective determined center of each of the at least one group of separated selectively chosen good data values; and

selecting representative selectively chosen good data values using equally spaced samples from the sorted selectively chosen good data values in each of the at least one group of separated selectively chosen good data values.

7. A computer-implemented adaptive model training method, said method comprising the steps of:

providing a set of selectively chosen good data values acquired from operation of an asset wherein the selectively chosen good data values acquired from operation of the asset are selectively chosen based on at least one predefined criterion of good data quality;

clustering the provided selectively chosen good data values based on a computed similarity between each of the provided selectively chosen good data values and at least one predefined criterion for assigning each of the provided selectively chosen good data values to at least one cluster of similar selectively chosen good data values;

selecting representative selectively chosen good data values from each of the at least one cluster of similar selectively chosen good data values;

forming a reduced set of data values approximating a distribution of the provided selectively chosen good data values by combining the selected representative selectively chosen good data values from each of the at least one cluster of similar selectively chosen good data values; and

storing the reduced set of data values approximating the distribution of the provided selectively chosen good data values for defining a learned scope of operation of the asset.

8. The computer-implemented method of claim 7 further comprising a step of calibrating a model by utilizing at least a portion of the stored reduced set of data values approximating the distribution of the provided selectively chosen good data values for defining a calibrated model having a learned scope of operation of the asset.

9. The computer-implemented method of claim 7 further comprising the steps of:

providing a previously calibrated model having a learned scope of operation of the asset; and

recalibrating the previously calibrated model by utilizing at least a portion of the stored reduced set of data values approximating the distribution of the provided selectively chosen good data values for defining a recalibrated model having an adjusted learned scope of operation of the asset.

10. A computer-implemented data reduction method, said method comprising:

providing a set of selectively chosen good data values acquired from operation of an asset wherein the selectively chosen good data values acquired from operation of the asset are selectively chosen based on at least one predefined criterion of good data quality;

recursively dividing the set of provided selectively chosen good data values into a plurality of clusters until each of the plurality of clusters satisfies at least one predefined test statistic so that each of the plurality of clusters comprises similar selectively chosen good data values;

selecting representative selectively chosen good data values from each of the plurality of clusters of similar selectively chosen good data values;

forming a reduced set of data values approximating a distribution of the provided selectively chosen good data values by combining the selected representative selectively chosen good data values from each of the plurality of clusters of similar selectively chosen good data values; and

storing the reduced set of data values approximating the distribution of the provided selectively chosen good data values for defining a learned scope of operation of the asset.

11. The computer-implemented method of claim 10 further comprising a step of calibrating a model by utilizing at least a portion of the stored reduced set of data values approximating the distribution of the provided selectively chosen good data values for defining a calibrated model having a learned scope of operation of the asset.

12. The computer-implemented method of claim 10 further comprising the steps of:

providing a previously calibrated model having a learned scope of operation of the asset; and

recalibrating the previously calibrated model by utilizing at least a portion of the stored reduced set of data values approximating the distribution of the provided selectively chosen good data values for defining a recalibrated model having an adjusted learned scope of operation of the asset.

13. A computer-implemented adaptive model training method, said method comprising the steps of:

obtaining a set of initial data values from operation of an asset;

sorting the obtained set of initial data values into at least one set of operating mode partitioned data values based on at least one predefined criterion for determining an operating mode of the asset;

filtering the at least one set of operating mode partitioned data values based on at least one predefined criterion of good data quality to obtain at least one set of good quality data values for at least one operating mode of the asset;

separating the at least one set of good quality data values for the at least one operating mode of the asset based on at least one predefined criterion for assigning each of the good quality data values to at least one set of separated good quality data values for the at least one operating mode of the asset;

selecting at least one set of representative good quality data values for the at least one operating mode of the asset from each of the at least one set of separated good quality data values for the at least one operating mode of the asset based on at least one predefined criterion for sampling representative good quality data values from each of the at least one set of separated good quality data values for the at least one operating mode of the asset; and

forming a reduced set of data values for the at least one operating mode of the asset by combining each of the at least one set of the selected representative good quality data values for the at least one operating mode of the asset.

14. The computer-implemented adaptive model training method of claim 13 further comprising:

providing a previously calibrated model for th at least one operating mode of the asset;

recalibrating the model for the at least one operating mode of the asset utilizing the reduced set of data values for the at least one operating mode of the asset; and

storing the recalibrated model for the at least one operating mode of the asset.