Machine learning for power grid

• US 8,751,421 B2
• Filed: 01/15/2013
• Issued: 06/10/2014
• Est. Priority Date: 07/16/2010
• Status: Expired due to Fees

• Current Assignee: Trustees Of Columbia University In The City Of New York (Columbia University), Consolidated Edison Company of New York Incorporated (Consolidated Edison Incorporated)
• Original Assignee: Trustees Of Columbia University In The City Of New York (Columbia University), Consolidated Edison Company of New York Incorporated (Consolidated Edison Incorporated)
• Inventors: Anderson, Roger N., Boulanger, Albert, Rudin, Cynthia, Waltz, David, Salleb-Aouissi, Ansaf, Chow, Maggie, Dutta, Haimonti, Gross, Phil, Bert, Huang, Ierome, Steve, Isaac, Delfina, Kressner, Arthur, Passonneau, Rebecca J., Radeva, Axinia, Wu, Leon L., Hofmann, Peter, Dougherty, Frank
• Primary Examiner(s): VINCENT, DAVID ROBERT
• Application Number:
  US13/742,124
• Publication Number:
  US 20130232094A1
• Time in Patent Office: 511 Days
• Field of Search: 706/12, 706/20, 706/45, 706/62
• US Class Current: 706/12
• CPC Class Codes: G05B 23/0229 knowledge based, e.g. expert systems; genetic algorithms
• CPC Class Codes: G05B 23/0281 Quantitative, e.g. mathematical distance; Clustering; Neural networks; Statistical analysis
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 20/10 using kernel methods, e.g. support vector machines [SVM]
• CPC Class Codes: G16Z 99/00 Subject matter not provided for in other main groups of this subclass
• CPC Class Codes: Y04S 10/52 Outage or fault management, e.g. fault detection or location

Abstract

A machine learning system for ranking a collection of filtered propensity to failure metrics of like components within an electrical grid that includes a raw data assembly to provide raw data representative of the like components within the electrical grid; (b) a data processor, operatively coupled to the raw data assembly, to convert the raw data to more uniform data via one or more data processing techniques; (c) a database, operatively coupled to the data processor, to store the more uniform data; (d) a machine learning engine, operatively coupled to the database, to provide a collection of propensity to failure metrics for the like components; (e) an evaluation engine, operatively coupled to the machine learning engine, to detect and remove non-complying metrics from the collection of propensity to failure metrics and to provide the collection of filtered propensity to failure metrics; and (f) a decision support application, operatively coupled to the evaluation engine, configured to display a ranking of the collection of filtered propensity to failure metrics of like components within the electrical grid.

Claims

1. A machine learning system for determining propensity to failure metrics of like components within an electrical grid comprising:

(a) a raw data assembly to provide raw data representative of the like components within the electrical grid;

(b) a data processor, operatively coupled to the raw data assembly, to convert the raw data to more uniform data via one or more data processing techniques;

(c) a database, operatively coupled to the data processor, to store the more uniform data;

(d) a machine learning engine, operatively coupled to the database, to provide;

(i) a ranking of the collection of propensity to failure metrics for the like components, and

(ii) absolute value of the propensity to failure metrics for the like components;

(e) an evaluation engine, operatively coupled to the machine learning engine, to detect and remove non-complying metrics from the collection of propensity to failure metrics and to provide the collection of filtered propensity to failure metrics; and

(f) a decision support application, operatively coupled to the evaluation engine, configured to display a ranking of the collection of filtered propensity to failure metrics of like components within the electrical grid.

2. The machine learning system of claim 1, wherein the raw data representative of the like components is obtained at least in part from a maintenance record or a maintenance request for at least one of the like components.

3. The machine learning system of claim 1, wherein the data processing techniques include one or more of an inferential join, pattern matching, information extraction, text normalization, querying overlapping data to find inconsistencies, and inference from related or duplicate records.

4. The machine learning system of claim 1, wherein the non-complying metrics are detected based on one or more of Area Under the Receive Operating Characteristic (ROC) Curve (AUC), weighted AUC, and a pre-defined percentage fail exclusion.

5. The machine learning system of claim 1, wherein the evaluation engine further comprises an outage derived database to store outage derived data sets that capture dynamic precursor to fail data representative of at least one of the like components.

6. The machine learning system of claim 5, wherein the dynamic precursor to fail data is obtained from a time-shifted time domain ending at just before the time of failure and beginning at a pre-selected time prior to the failure.

7. The machine learning system of claim 1, wherein the like components are secondary network components selected from cables, joints, terminators, hammerheads, manholes and transformers.

8. The machine learning system of claim 7, wherein the like components are manholes.

9. The machine learning system of claim 1, wherein the propensity to failure metric is selected from mean time between failure (MTBF) and mean time to failure (MTTF).

10. The machine learning system of claim 1, wherein the evaluation engine comprises a data historian to record a collection of filtered propensity to failure metrics over a first time period and a collection of filtered propensity to failure metrics over a second time period, and a data analyzer to determine whether the collection of filtered propensity to failure metrics at the first time period is statistically improved over the collection of filtered propensity to failure metric at the second time period.

21. The system of claim 1, wherein the absolute value of the propensity to failure metrics for the like components is determined using Cox proportional hazard model.

22. The system of claim 1, wherein absolute value of the propensity to failure metrics for the like components is determined by assessing interaction between one or more covariates that could affect one or more failure metrics for the like component.

11. A method for determining propensity to failure metrics of like components within an electrical grid via machine learning comprising:

(a) providing a raw data assembly to provide raw data representative of the like components within the electrical grid;

(b) processing the raw data to convert the raw data to more uniform data via one or more data processing techniques;

(c) storing the more uniform data in a database;

(d) transmitting the more uniform data to a machine learning engine to provide a collection of propensity to failure metrics for the like components;

(e) evaluating the collection of propensity to failure metrics in an evaluation engine to detect and remove non-complying metrics from the collection of propensity to failure metrics and to provide the collection of filtered propensity to failure metrics;

(f) ranking the collection of filtered propensity to failure metrics obtained from the evaluation engine and displaying the ranking on a decision support application; and

(g) determining an absolute value of the propensity to failure metrics for the like components.

12. The method of claim 11, wherein the raw data representative of the like components is obtained at least in part from a maintenance record or a maintenance request for at least one of the like components.

13. The method of claim 11, wherein the data processing techniques include one or more of an inferential join, pattern matching, information extraction, text normalization, querying overlapping data to find inconsistencies, and inference from related or duplicate records.

14. The method of claim 11, wherein the non-complying metrics are detected based on one or more of Area Under the Receive Operating Characteristic (ROC) Curve (AUC), weighted AUC, and a pre-defined percentage fail exclusion.

15. The method of claim 11, wherein the evaluating further comprises storing outage derived data sets that capture dynamic precursor to fail data representative of at least one of the like components.

16. The method of claim 15, wherein the dynamic precursor to fail data is obtained from a time-shifted time domain ending at or just before the time of failure and beginning at a pre-selected time prior to the failure.

17. The method of claim 11, wherein the like components are components selected from feeders, cables, joints, terminators, hammerheads, manholes and transformers.

18. The machine learning system of claim 17, wherein the like components are manholes.

19. The method of claim 11, wherein the propensity to failure metric is selected from mean time between failure (MTBF) and mean time to failure (MTTF).

20. The method of claim 11, wherein the evaluating further comprises recording a collection of filtered propensity to failure metrics over a first time period and a collection of filtered propensity to failure metrics over a second time period, and analyzing to determine whether the collection of filtered propensity to failure metrics at the first time period is statistically improved over the collection of filtered propensity to failure metrics at the second time period.