Electrical transformer failure prediction

• US 9,652,723 B2
• Filed: 06/06/2016
• Issued: 05/16/2017
• Est. Priority Date: 06/05/2015
• Status: Active Grant

• Current Assignee: SAS Institute Incorporated
• Original Assignee: SAS Institute Incorporated
• Inventors: Anderson, Thomas Dale, Duarte, James Edward, Falahi, Milad
• Primary Examiner(s): CHANG, LI WU
• Application Number:
  US15/173,927
• Publication Number:
  US 20160358106A1
• Time in Patent Office: 344 Days
• Field of Search: None
• CPC Class Codes: G05B 23/0254 based on a quantitative model, e.g. mathematical relationships between inputs and outputs; functions: observer, Kalman filter, residual calculation, Neural Networks
• CPC Class Codes: G05B 23/0283 Predictive maintenance, e.g. involving the monitoring of a system and, based on the monitoring results, taking decisions on the maintenance schedule of the monitored system; Estimating remaining useful life [RUL] preventive maintenance, i.e. planning maintenance according to the available resources without monitoring the system G06Q10/06
• CPC Class Codes: G06F 11/00 Error detection; Error correction; Monitoring error detection, correction or monitoring in information storage based on relative movement between record carrier and transducer G11B20/18; monitoring, i.e. supervising the progress of recording or reproducing G11B27/36; in static stores G11C29/00
• CPC Class Codes: G06N 20/00 Machine learning
• CPC Class Codes: G06N 3/08 Learning methods
• CPC Class Codes: G06N 7/01 Probabilistic graphical models, e.g. probabilistic networks

Abstract

A computing device predicts a probability of a transformer failure. An analysis type indicator defined by a user is received. A worth value for each of a plurality of variables is computed. Highest worth variables from the plurality of variables are selected based on the computed worth values. A number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator. A first model and a second model are also selected based on the received analysis type indicator. Historical electrical system data is partitioned into a training dataset and a validation dataset that are used to train and validate, respectively, the first model and the second model. A probability of failure model is selected as the first model or the second model based on a comparison between a fit of each model.

Claims

1. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is greater than 200 based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator, wherein the selected first model is a decision tree model and the selected second model is a neural network model;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a probability of failure model as the validated first model or the validated second model based on the comparison;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

2. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a third model based on the received analysis type indicator, wherein the selected first model is a decision tree model, the selected second model is a neural network model, and the selected third model is a backward regression model;

train the selected third model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained third model using the validation dataset to statistically assess the fit by the trained third model to the historical electrical system data;

compare the fit by the trained third model to the fit by the trained second model and the fit by the trained first model;

select a probability of failure model as the validated first model, the validated second model, or the validated third model based on the comparisons;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

3. The non-transitory computer-readable medium of claim 2, wherein the predetermined number is greater than 200.

4. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a third model based on the received analysis type indicator;

train the selected third model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained third model using the validation dataset to statistically assess the fit by the trained third model to the historical electrical system data;

compare the fit by the trained third model to the fit by the trained second model and the fit by the trained first model;

create a fourth model as an ensemble model based on the received analysis type indicator using the trained third model, the trained second model, and the trained first model;

train the created fourth model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained fourth model using the validation dataset to statistically assess the fit by the trained fourth model to the historical electrical system data;

select a probability of failure model as the validated first model, the validated second model, or the validated third model based on the comparisons;

compare the fit by the trained fourth model to the fit by the selected probability of failure model;

update the probability of failure model selection as the validated fourth model or the selected probability of failure model based on the comparison between the trained fourth model and the selected probability of failure model;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

5. The non-transitory computer-readable medium of claim 4, wherein the selected first model is a decision tree model, the selected second model is a neural network model, and the selected third model is a backward regression model.

6. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator, wherein the selected first model is a decision tree model and the selected second model is a stepwise regression model;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a probability of failure model as the validated first model or the validated second model based on the comparison;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

7. The non-transitory computer-readable medium of claim 6, wherein the predetermined number is less than 200 and greater than 100.

8. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a third model based on the received analysis type indicator, wherein the selected first model is a decision tree model, the selected second model is a backward regression model, and the selected third model is a logistic regression model;

train the selected third model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained third model using the validation dataset to statistically assess the fit by the trained third model to the historical electrical system data;

compare the fit by the trained third model to the fit by the trained second model and the fit by the trained first model;

select a probability of failure model as the validated first model, the validated second model, or the validated third model based on the comparisons;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

9. The non-transitory computer-readable medium of claim 8, wherein the predetermined number is less than 200 and greater than 100.

10. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a probability of failure model as the validated first model or the validated second model based on the comparison;

select second highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the second highest worth variables is limited to a second predetermined number based on the received analysis type indicator;

select a third model based on the received analysis type indicator;

train the selected third model using values from the training dataset of the selected second highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained third model using the validation dataset to statistically assess the fit by the trained third model to the historical electrical system data;

compare the fit by the trained third model to the fit by the selected probability of failure model;

update the probability of failure model selection as the validated third model or the selected probability of failure model based on the comparison between the trained third model and the selected probability of failure model;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

11. The non-transitory computer-readable medium of claim 10, wherein the predetermined number is less than 200 and greater than 100, and the second predetermined number is greater than 200.

12. The non-transitory computer-readable medium of claim 11, wherein the selected first model is a first decision tree model, the selected second model is a backward regression model, and the selected third model is a second decision tree model.

13. The non-transitory computer-readable medium of claim 10, wherein the predetermined number is less than 200 and greater than 100, and the second predetermined number is less than 100.

14. The non-transitory computer-readable medium of claim 13, wherein the selected first model is a first decision tree model, the selected second model is a backward regression model, and the selected third model is a forward regression model.

15. The non-transitory computer-readable medium of claim 10, wherein, before receiving electrical system data, the computer-readable instructions further cause the computing device to:

select third highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the third highest worth variables is limited to a third predetermined number based on the received analysis type indicator;

select a fourth model based on the received analysis type indicator;

train the selected fourth model using values from the training dataset of the selected third highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained fourth model using the validation dataset to statistically assess the fit by the trained fourth model to the historical electrical system data; and

compare, after updating the probability of failure model selection, the fit by the trained fourth model to the fit by the selected probability of failure model; and

update the probability of failure model selection as the validated fourth model or the updated probability of failure model based on the comparison between the trained fourth model and the selected probability of failure model.

16. The non-transitory computer-readable medium of claim 15, wherein the predetermined number is less than 200 and greater than 100, the second predetermined number is greater than or equal to 200, and the third predetermined number is less than 100.

17. The non-transitory computer-readable medium of claim 15, wherein the selected first model is a first decision tree model, the selected second model is a backward regression model, the selected third model is a second decision tree model, and the selected fourth model is a forward regression model.

18. The non-transitory computer-readable medium of claim 15, wherein the selected highest worth variables include meter event information, meter interval usage data, and transformer attribute information.

19. The non-transitory computer-readable medium of claim 18, wherein the selected second highest worth variables further include customer information, location information, weather information, meter attribute information, and meter maintenance information.

20. The non-transitory computer-readable medium of claim 19, wherein the selected third highest worth variables include only the weather information and the meter interval usage data.

21. A non-transitory computer-readable medium having stored thereon computer-readable instructions that when executed by a computing device cause the computing device to:

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

supplement the received historical electrical system data with values computed from one or more of the plurality of data points defined for each observation, wherein the values computed from one or more of the plurality of data points are counts associated with a plurality of electric meters that are summed over a predefined time interval, wherein the counts are further summed for the transformer to which the plurality of electric meters is connected;

partition the supplemented historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a probability of failure model as the validated first model or the validated second model based on the comparison;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

22. The non-transitory computer-readable medium of claim 21, wherein the plurality of data points are associated with at least one of a phase error event, a time change event, a power outage event, a reverse rotation event, a tamper event, an unknown event, and an out of range event.

23. The non-transitory computer-readable medium of claim 21, wherein the counts are summed over a plurality of predefined time intervals, wherein the plurality of predefined time intervals include a last month, a last three months, and a last six months.

24. A computing device comprising:

a processor; and

a non-transitory computer-readable medium operably coupled to the processor, the computer-readable medium having computer-readable instructions stored thereon that, when executed by the processor, cause the computing device to

receive historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partition the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receive an analysis type indicator defined by a user;

compute a worth value for each of the plurality of variables;

select highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

select a first model based on the received analysis type indicator;

train the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validate the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

select a second model based on the received analysis type indicator, wherein the selected first model is a decision tree model and the selected second model is a stepwise regression model;

train the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

compare the fit by the trained first model to the fit by the trained second model;

select a probability of failure model as the validated first model or the validated second model based on the comparison;

receive electrical system data for a transformer;

execute the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

update a failure probability for the transformer based on the computed probability of failure.

25. The computing device of claim 24, wherein the predetermined number is less than 200 and greater than 100.

26. The computing device of claim 24, wherein, before receiving electrical system data, the computer-readable instructions further cause the computing device to:

select second highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the second highest worth variables is limited to a second predetermined number based on the received analysis type indicator;

select a third model based on the received analysis type indicator;

train the selected third model using values from the training dataset of the selected second highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validate the trained third model using the validation dataset to statistically assess the fit by the trained third model to the historical electrical system data; and

compare, after selecting the probability of failure model, the fit by the trained third model to the fit by the selected probability of failure model; and

update the probability of failure model selection as the validated third model or the selected probability of failure model based on the comparison between the trained third model and the selected probability of failure model.

27. A method of predicting a probability of a transformer failure, the method comprising:

receiving historical electrical system data that includes a plurality of observations with a plurality of data points defined for each observation, wherein each data point of the plurality of data points is associated with a variable to define a plurality of variables;

partitioning, by a computing device, the received historical electrical system data into a training dataset and a validation dataset, wherein the validation dataset is different from the training dataset;

receiving an analysis type indicator defined by a user;

computing, by the computing device, a worth value for each of the plurality of variables;

selecting, by the computing device, highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the highest worth variables is limited to a predetermined number based on the received analysis type indicator;

selecting, by the computing device, a first model based on the received analysis type indicator;

training, by the computing device, the selected first model using values from the training dataset of the selected highest worth variables to predict a probability of failure of a plurality of electrical transformers;

validating, by the computing device, the trained first model using the validation dataset to statistically assess a fit by the trained first model to the historical electrical system data;

selecting, by the computing device, a second model based on the received analysis type indicator, wherein the selected first model is a decision tree model and the selected second model is a stepwise regression model;

training, by the computing device, the selected second model using values from the training dataset of the selected highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validating, by the computing device, the trained second model using the validation dataset to statistically assess the fit by the trained second model to the historical electrical system data;

comparing, by the computing device, the fit by the trained first model to the fit by the trained second model;

selecting, by the computing device, a probability of failure model as the validated first model or the validated second model based on the comparison;

receiving electrical system data for a transformer;

executing, by the computing device, the selected probability of failure model with the received electrical system data to compute a probability of failure of the transformer; and

updating, by the computing device, a failure probability for the transformer based on the computed probability of failure.

28. The method of claim 27, wherein the predetermined number is less than 200 and greater than 100.

29. The method of claim 27, further comprising:

selecting second highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the second highest worth variables is limited to a second predetermined number based on the received analysis type indicator;

selecting a third model based on the received analysis type indicator;

training the selected third model using values from the training dataset of the selected second highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validating the trained third model using the validation dataset to statistically assess the fit by the trained third model to the historical electrical system data; and

comparing, after selecting the probability of failure model, the fit by the trained third model to the fit by the selected probability of failure model; and

updating the probability of failure model selection as the validated third model or the selected probability of failure model based on the comparison between the trained third model and the selected probability of failure model.

30. The method of claim 29, wherein before receiving electrical system data the method further comprises:

selecting third highest worth variables from the plurality of variables based on the computed worth values, wherein a number of variables of the third highest worth variables is limited to a third predetermined number based on the received analysis type indicator;

selecting a fourth model based on the received analysis type indicator;

training the selected fourth model using values from the training dataset of the selected third highest worth variables to predict the probability of failure of the plurality of electrical transformers;

validating the trained fourth model using the validation dataset to statistically assess the fit by the trained fourth model to the historical electrical system data; and

comparing, after updating the probability of failure model selection, the fit by the trained fourth model to the fit by the selected probability of failure model; and

updating the probability of failure model selection as the validated fourth model or the updated probability of failure model based on the comparison between the trained fourth model and the selected probability of failure model.