System and method for grading electricity distribution network feeders susceptible to impending failure

• US 7,945,524 B2
• Filed: 07/23/2008
• Issued: 05/17/2011
• Est. Priority Date: 01/23/2006
• 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., Long, Phil, Boulanger, Albert, Gross, Philip, Becker, Hila, Mastrocinque, Mark, Johnson, John A., Arias, Marta, Waltz, David L., Kressner, Arthur, Koenig, Matthew
• Primary Examiner(s): Holmes; Michael B.
• Assistant Examiner(s): COUGHLAN, PETER D
• Application Number:
  US12/178,553
• Publication Number:
  US 20090157573A1
• Time in Patent Office: 1,028 Days
• Field of Search: None
• US Class Current: 706/12
• CPC Class Codes: G06N 20/00 Machine learning

Abstract

A machine learning system creates failure-susceptibility rankings for feeder cables in a utility'"'"'s electrical distribution system. The machine learning system employs martingale boosting algorithms and Support Vector Machine (SVM) algorithms to generate a feeder failure prediction model, which is trained on static and dynamic feeder attribute data. Feeders are dynamically ranked by failure susceptibility and the rankings displayed to utility operators and engineers so that they can proactively service the distribution system to prevent local power outages. The feeder rankings may be used to redirect power flows and to prioritize repairs. A feedback loop is established to evaluate the responses of the electrical distribution system to field actions taken to optimize preventive maintenance programs.

Claims

1. A system for facilitating maintenance of feeders in an electricity distribution network, the system comprising:

values of a plurality of feeder attributes related to past performance history;

a particular dataset for training a model configured to determine the relative susceptibilities to failure of the feeders based on the values of the plurality of feeder attributes, wherein the particular dataset for training comprises data on feeders from a failure history database and other feeders that are selected based on their situational association with a specific failed feeder and/or selected based on a similarity measure used in statistical causal inference; and

a machine learning engine configured to train the model and to apply the trained model to the values of a plurality of feeder attributes so as to generate rankings of the feeders indicative of their relative susceptibilities to failure; and

a decision support application configured to provide the rankings of the feeders available to operators and engineers so as to facilitate maintenance of the feeders.

2. The system of claim 1, wherein the feeder attributes include static feeder attributes and dynamic feeder attributes.

3. The system of claim 1, wherein the feeder attributes include at least one of the feeder attributes listed in Appendix A.

4. The system of claim 1, wherein the machine learning engine includes one of a martingale boosting algorithm, a SVM algorithm and a combination thereof.

5. The system of claim 1, wherein the model comprises a ranking function with values determined by sorting a list of the feeders on a feeder attribute.

6. The system of claim 1, wherein the machine learning engine is configured to evaluate the model against test data using Area Under Receiver Operating Characteristic Curve metrics.

7. The system of claim 1, wherein the machine learning engine is configured to apply the model to dynamic feeder data based on field sensors to update the feeder rankings.

8. The system of claim 1, wherein the decision support application includes a web-based user interface.

9. The system of claim 1, further comprising a tracking application that determines the electrical system responses to maintenance actions taken on the electricity distribution network, and integrates the responses with the decision support application to optimize electricity distribution network performance.

10. The system of claim 1, further comprising an online model-selection engine that can monitor the performance of a pool of candidate models generated by varying different training mode parameters.

11. The system of claim 1, wherein the feeder attributes include a feeder attribute selected from cable sections, joints, transformers, terminators, load pocket weight, power quality, and past outage history.

12. A method for facilitating maintenance of feeders in an electricity distribution network, the method comprising:

providing a failure prediction model;

training the failure prediction model on values of a plurality of feeder attributes related to past performance history;

applying the trained failure prediction model to values of the plurality of feeder attributes so as to generate rankings of the feeders indicative of their relative susceptibilities to failure; and

providing the rankings of the feeders to operators and engineers so as to facilitate maintenance of the feeders.

13. The method of claim 12, wherein training the failure prediction model includes using a martingale boosting algorithm.

14. The method of claim 12, wherein training the failure prediction model on values of a plurality of feeder attributes related to past performance history, includes training on the values of at least one of the feeder attributes listed in Appendix A.

15. The method of claim 12, wherein applying the trained failure prediction model to values of the plurality of feeder attributes includes applying the model to dynamic values so as to dynamically update rankings of the feeders indicative of their relative susceptibilities to failure.

16. The method of claim 12, wherein providing the rankings of the feeders to operators and engineers includes providing the rankings over a web-based user interface.

17. The method of claim 12, wherein providing the rankings of the feeders to operators and engineers includes providing a tracking application that determines the electrical system responses to maintenance actions taken on the electricity distribution network.

18. The method of claim 17 further comprising providing a decision support application which integrates providing the rankings of the feeders to operators and engineers with the results of the tracking application to enable optimization of electricity distribution network performance and determining maintenance policies.

19. The method of claim 12, further comprising using an online model-selection engine to monitor the performance of a pool of candidate models generated by varying different training mode parameters.

20. The method of claim 12, wherein training the failure prediction model on values of a plurality of feeder attributes related to past performance history, includes training on the values of at least one feeder attribute selected from cable sections, joints, transformers, terminators, load pocket weight, power quality, and past outage history.

21. A computer-readable medium for facilitating maintenance of feeders in an electricity distribution network, the computer-readable medium comprising a set of instructions for:

training a failure prediction model on values of a plurality of feeder attributes related to past performance history;

applying the trained failure prediction model to values of the plurality of feeder attributes so as to generate rankings of the feeders indicative of their relative susceptibilities to failure; and

providing the rankings of the feeders to operators and engineers so as to facilitate maintenance of the feeders.

22. The computer-readable medium of claim 21, wherein the set of instructions includes a martingale boosting algorithm.

23. The computer-readable medium of claim 21, wherein the set of instructions includes instructions for training the failure prediction model on the values of at least one of the attributes listed in Appendix A.

24. The computer-readable medium of claim 21, wherein the set of instructions includes instructions for applying the trained failure prediction model to dynamic values of the plurality of feeder attributes so as to dynamically update the feeder rankings.

25. The computer-readable medium of claim 21, wherein the set of instructions for making the feeder rankings available to operators and engineers includes instructions for displaying the feeder rankings over a web-based interface.

26. The computer-readable medium of claim 21, wherein the set of instructions further comprises instructions for a feedback loop that measures electrical distribution network reactions to maintenance actions taken based on the feeder rankings made available to engineers an operators, so that electricity distribution network performance can be optimized.

27. The computer-readable medium of claim 21, wherein the set of instructions further comprises instructions to monitor the performance of a pool of candidate models generated by varying different training mode parameters.

28. The computer-readable medium of claim 21, wherein the set of instructions includes instructions for training the failure prediction model on the values of at least one attribute selected from cable sections, joints, transformers, terminators, load pocket weight, power quality, and past outage history.

29. A system for assessing likelihood of failure in an “

items”

distribution network, the distribution network having a plurality of networked or interlinked components involved in transmission of the items to end-users or recipients, the system comprising;

a failure-prediction model configured to determine the relative susceptibility-to-failures of individual network components based on an input data set of network component attribute values; and

a machine learning engine configured to train the model and to apply the trained model to the input data set so as to generate a list of the network components ranked by their relative susceptibility-to-failures.

30. The system of claim 29 wherein the items are one of electric power, water, communications, transportation, sewer, gas, and industrial goods and services.

31. A method for assessing likelihood of failure in an “

items”

distribution network, the distribution network having a plurality of networked or interlinked components involved in transmission of the items to end-users or recipients, the method comprising;

providing a failure-prediction model configured to determine the relative susceptibility-to-failures of individual network components based on an input data set of network component attribute values;

training the model on a training data set of network component attribute values; and

applying the trained model to the input data set so as to generate a list of the network components ranked by their relative susceptibility-to-failures.

32. The method of claim 31 wherein the items are one of electric power, water, communications, transportation, sewer, gas, and industrial goods and services.