Data selection device, data selection method, and program

Abstract

To reduce generation time while suppressing a decrease in the accuracy of an estimation model.SOLUTION: A data selection device of an embodiment has a grouping unit and a data reduction unit. The grouping unit groups partial discharge data that includes a partial discharge signal measured in an apparatus in which partial discharge occurs and partial discharge cause data indicating a cause of the partial discharge into respective groups based on the partial discharge signal. The data reduction unit reduces the partial discharge data based on the number of partial discharge data divided in each group and a predetermined threshold value.SELECTED DRAWING: Figure 2

Description

Technical Field

[0001] Embodiments of the present invention relate to a data selection device, a data selection method, and a program.

Background Art

[0002] Due to the aging deterioration of power equipment, the insulation performance of the surface or internal insulators of the power equipment decreases. Partial discharge occurs from the location where the insulation performance has decreased. If the decrease in insulation performance progresses, there is a possibility of insulation breakdown in the power equipment. In addition, the insulation performance may also decrease due to a surge voltage that suddenly enters the power equipment.

[0003] There is a known method of measuring a signal emitted from equipment such as power equipment and diagnosing the state of the equipment, such as the cause of partial discharge, based on the signal. For example, there is a method of determining the type of partial discharge by inputting a φqn pattern into a neural network. For the learning of the model to be determined, teacher data combining the measured partial discharge signal and noise for each cause of partial discharge is used.

Prior Art Documents

Patent Documents

[0004]

Patent Document 1

Summary of the Invention

Problems to be Solved by the Invention

[0005] However, the teacher data may include similar data. When there is a large amount of similar data, the time required to generate a model using the teacher data becomes long, and the estimation accuracy of the generated model may not improve. In addition, when there is a large amount of similar data, overfitting may occur in the generation of the model. The problem to be solved by the present invention is to provide a data selection device, a data selection method, and a program that can shorten the generation time while suppressing a decrease in the accuracy of the estimation model.

Means for Solving the Problem

[0006] The data selection device according to the embodiment includes a grouping unit and a data reduction unit. The grouping unit groups partial discharge data including partial discharge signals measured in devices where partial discharge occurs and partial discharge cause data indicating the cause of the partial discharge into each group based on the partial discharge signals. The data reduction unit reduces the partial discharge data based on the number of partial discharge data grouped into each group and a predetermined threshold value.

Brief Description of the Drawings

[0007]

Figure 1

Figure 2

Figure 3

Figure 4

Figure 5

Figure 6

Figure 7

Figure 8

Figure 9

Figure 10

Figure 11

Best Mode for Carrying Out the Invention

[0008] Hereinafter, a data selection device, a data selection method, and a program according to an embodiment will be described with reference to the drawings. FIG. 1 is a diagram showing the configuration of an estimation model generation system 1 according to the present embodiment. The estimation model generation system 1 generates an estimation model for estimating the cause of partial discharge of a device based on the partial discharge signal of the device. The estimation model generation system 1 includes a partial discharge sensor 10, a data selection device 11, and an estimation model generation device 12.

[0009] The partial discharge sensor 10 is provided in the device T. The partial discharge sensor 10 measures a signal generated by the partial discharge of the device T. Examples of the partial discharge sensor 10 include a TEV sensor that measures a transient ground voltage passing through the housing due to partial discharge, a vibration sensor that detects vibration due to partial discharge, and a microphone that detects sound waves due to partial discharge. The partial discharge sensor 10 outputs the measured partial discharge signal to the data selection device 11.

[0010] The data selection device 11 selects the partial discharge signal input from the partial discharge sensor 10 and outputs the selected data to the estimation model generation device 12.

[0011] The estimation model generation device 12 generates an estimation model for estimating the cause of partial discharge based on the data input from the data selection device 11.

[0012] FIG. 2 is a diagram showing the configuration of the data selection device 11 according to the present embodiment. The data selection device 11 includes a partial discharge data storage unit 111, a grouping unit 112, a grouping result storage unit 113, a data reduction unit 114, a reduction method reception unit 115, and a data output unit 116.

[0013] The partial discharge data storage unit 111 stores by associating the input partial discharge signal with data indicating the cause of the partial discharge occurring in the device T (hereinafter referred to as partial discharge cause data). The partial discharge cause data is stored in the partial discharge data storage unit 111, for example, when the administrator of the device T inputs the cause of the partial discharge in the device T to the data selection device 11. FIG. 3 is a diagram showing an example of the partial discharge data stored in the partial discharge data storage unit 111. In the partial discharge data storage unit 111, data in which the partial discharge cause data indicating the discharge cause (for example, void, metal foreign matter) is associated with the partial discharge signal is stored. Hereinafter, the data stored in the partial discharge data storage unit 111 is referred to as partial discharge data.

[0014] In the estimation model generation system 1, the cause of the partial discharge is embedded in the device T in advance, and the partial discharge is generated by applying a voltage. The partial discharge signal generated at this time is measured by the partial discharge sensor 10 and stored in the partial discharge data storage unit 111. Also, the partial discharge cause data is stored by recording in the partial discharge data storage unit 111 the data indicating the cause of the partial discharge embedded in the device T in advance.

[0015] The grouping unit 112 groups the partial discharge data into a plurality of groups based on the partial discharge signal for each cause of the partial discharge. The grouping unit 112 extracts, for example, feature quantities from the partial discharge signal, and groups the partial discharge data into each group based on the extracted feature quantities. The feature quantity extraction method is, for example, a method of converting the partial discharge signal from the time domain to the frequency domain and performing dimensionality compression on the converted data. The conversion from the time domain to the frequency domain is, for example, a Fourier transform. The method of dimensionality compression is, for example, PCA (Principal Component Analysis), UMAP (Uniform Manifold Approximation and Projection), t-SNE (t-Distributed Stochastic Neighbor Embedding), or a method based on machine learning.

[0016] The method of grouping by the grouping unit 112 based on the partial discharge signal or the feature amount is not particularly limited. The grouping method may be clustering that divides the partial discharge data into unknown clusters, or classification that divides the partial discharge data into known classes. Examples of clustering algorithms include the K-Means method, the Mean-shift method, the DBSCAN method, etc. The clustering algorithm may use machine learning. The clustering algorithm may cluster the partial discharge data based on data indicating the number of feature amounts within a predetermined range.

[0017] FIG. 4 is a diagram showing an example of the grouped partial discharge data. FIG. 4 shows the partial discharge data and the groups to which each partial discharge data belongs. The partial discharge data may include the extracted feature amounts.

[0018] The grouping result storage unit 113 stores the grouping result for each cause of partial discharge by the grouping unit 112. The grouping result storage unit 113 stores, for example, the partial discharge data and the groups to which each partial discharge data belongs.

[0019] The data reduction unit 114 deletes some of the partial discharge data among the partial discharge data to reduce the partial discharge data. The data reduction unit 114 deletes some of the partial discharge data based on, for example, the number of partial discharge data belonging to each group and a predetermined threshold value. The data reduction unit 114 reduces the partial discharge data so that, for example, for a group in which the number of partial discharge data belonging thereto is equal to or greater than the predetermined threshold value, the number of partial discharge data belonging to the group becomes equal to or less than the predetermined threshold value. The data reduction unit 114 may randomly select and delete partial discharge data by, for example, the number of the difference between the number of data and the threshold value from the group.

[0020] The data reduction unit 114 may delete some partial discharge data based on a predetermined reduction rate. For example, the data reduction unit 114 sets a threshold value so that the number of partial discharge data reaches the predetermined reduction rate, and for a group in which the number of partial discharge data belonging thereto is equal to or greater than the set threshold value, the partial discharge data is reduced so that the number of partial discharge data in the group becomes equal to or less than the predetermined threshold value.

[0021] The data reduction unit 114 may reduce the partial discharge data so that the number of partial discharge data belonging to each group is the same. As a result, the number of partial discharge data belonging to each group is averaged.

[0022] The data reduction unit 114 reduces the number of partial discharge data by performing the above method on the partial discharge data for each cause of partial discharge.

[0023] When the number of partial discharge data with a common cause of partial discharge is equal to or greater than a predetermined threshold value, the data reduction unit 114 may delete the partial discharge data so that the number of the partial discharge data becomes equal to or less than the predetermined threshold value regardless of the group. At this time, for the partial discharge data whose number is equal to or greater than the predetermined threshold value, the data reduction unit 114 uses the ratio of the number of partial discharge data to the predetermined threshold value as the reduction rate, and for each cause of partial discharge, reduces the number of partial discharge data by a method similar to the above method so as to achieve the reduction rate.

[0024] The data reduction unit 114 may reduce the partial discharge data so that the number of partial discharge data for each cause of partial discharge is the same. At this time, for the partial discharge data for each cause of partial discharge, the data reduction unit 114 uses the ratio of the number of partial discharge data to the minimum value of the number of partial discharge data for each cause of partial discharge as the reduction rate, and for each cause of partial discharge, reduces the number of partial discharge data by a method similar to the above method so as to achieve the reduction rate.

[0025] The reduction method reception unit 115 receives the method by which the data reduction unit 114 reduces data. The method of reducing data is, for example, reduction based on a predetermined threshold value, reduction based on a predetermined reduction rate, or reduction by averaging. Further, the method of reducing data includes reduction based on the number of partial discharge data belonging to each group in a specific cause of partial discharge, and reduction based on the number for each cause of partial discharge.

[0026] The reduction method reception unit 115 may display, for example, the grouped results of the partial discharge data. The reduction method reception unit 115 may display, for example, the number of partial discharge data for each cause of partial discharge, and the number of partial discharge data for each group belonging to a specific cause of partial discharge. The number of partial discharge data may be shown by a graph (such as a bar graph or a pie chart). A person who inputs the data reduction method to the reduction method reception unit 115 may input the data reduction method based on the display of the number of partial discharge data.

[0027] FIG. 5 is a diagram showing the display by the reduction method reception unit 115 and the input to the reduction method reception unit 115. FIG. 5 shows an example in the case where the cause of partial discharge is a void as reduction based on the number of partial discharge data belonging to each group in a specific cause of partial discharge. Here, the cause of partial discharge may be selectable by input. In the example shown in FIG. 5, the display by the reduction method reception unit 115 shows the number for each group to which the partial discharge data having void as the cause of partial discharge belongs. In the example shown in FIG. 5, the number of partial discharge data is represented by a bar graph. The example shown in FIG. 5 shows the input of the reduction rate, threshold value, and averaging. In the example shown in FIG. 5, by inputting a value to the reduction rate and pressing the execute button, inputting a value to the threshold value and pressing the execute button, or pressing the averaging button, the partial discharge data is reduced. When a value is input to the reduction rate and the execute button is pressed, the data reduction unit 114 sets a threshold value so as to achieve the reduction rate in which the number of partial discharge data is input, and for a group in which the number of partial discharge data belonging thereto is equal to or greater than the set threshold value, reduces the partial discharge data so that the number of partial discharge data of the group becomes equal to or less than a predetermined threshold value.

[0028] When a value is input to the threshold and the execution button is pressed, the data reduction unit 114 reduces the partial discharge data for a group in which the number of partial discharge data belonging thereto is equal to or greater than the input threshold so that the number of partial discharge data belonging to the group becomes equal to or less than the input threshold. When the averaging button is pressed, the data reduction unit 114 reduces the partial discharge data so that the number of partial discharge data belonging to each group becomes the same number.

[0029] FIG. 6 is a diagram showing the display by the reduction method reception unit 115 and the input to the reduction method reception unit 115. FIG. 6 shows an example of reduction based on the number for each cause of partial discharge. Whether to perform reduction based on the number of partial discharge data belonging to each group for a specific cause of partial discharge or to perform reduction based on the number for each cause of partial discharge may be selectable by input. In the example shown in FIG. 6, the display by the reduction method reception unit 115 shows the number of partial discharge data for each cause of partial discharge. In the example shown in FIG. 6, the number of partial discharge data is represented by a bar graph. The example shown in FIG. 6 shows the input of the reduction rate, threshold, and averaging. In the example shown in FIG. 6, the partial discharge data is reduced by inputting a value to the reduction rate and pressing the execution button, inputting a value to the threshold and pressing the execution button, or pressing the averaging button. When a value is input to the reduction rate and the execution button is pressed, the data reduction unit 114 sets a threshold so as to achieve the input reduction rate of the number of partial discharge data, and for a cause of partial discharge in which the number of partial discharge data belonging thereto is equal to or greater than the set threshold, reduces the partial discharge data so that the number of partial discharge data for the cause of partial discharge becomes equal to or less than a predetermined threshold.

[0030] When a value is input to the threshold and the execution button is pressed, the data reduction unit 114 reduces the partial discharge data for a cause of partial discharge in which the number of partial discharge data belonging thereto is equal to or greater than the input threshold so that the number of partial discharge data belonging to the cause of partial discharge becomes equal to or less than the input threshold. When the averaging button is pressed, the data reduction unit 114 reduces the partial discharge data so that the number of partial discharge data belonging to each cause of partial discharge becomes the same number.

[0031] The reduction method reception unit 115 may receive multiple inputs, and the data reduction unit 114 may perform reduction of partial discharge data multiple times. At this time, the reduction methods input may be different or the same.

[0032] The data output unit 116 outputs the partial discharge data not deleted by the data reduction unit 114. The output partial discharge data is input to the estimation model generation device 12.

[0033] FIG. 7 is a flowchart showing the operation of the data selection device 11 according to the present embodiment. The grouping unit 112 groups the partial discharge data stored in the partial discharge data storage unit 111 based on the partial discharge signal (step S1). The reduction method reception unit 115 displays the grouping result of the partial discharge data by the grouping unit 112 (step S2). The person who inputs the reduction method checks the grouping result and inputs the reduction method. The reduction method reception unit 115 receives the input reduction method (step S3). The data reduction unit 114 reduces the partial discharge data according to the reduction method received by the reduction method reception unit 115 (step S4). The data output unit 116 outputs the partial discharge data not deleted to the estimation model generation device 12 (step S5).

[0034] The grouping unit 112 may group the partial discharge data into a plurality of groups based on the partial discharge signal regardless of the cause of the partial discharge. At this time, the data reduction unit 114 For a group in which the number of partial discharge data is equal to or greater than a threshold value set for each group, the partial discharge data may be reduced so that the number of partial discharge data belonging to the group becomes equal to or less than the threshold value set for each group. The threshold value set for each group may be set manually, such as by input to the data selection device 11, or may be set automatically.

[0035] The method for automatically setting the threshold value set for each group is, for example, a method of setting the threshold value based on the distance between groups or the number of types of causes included in the group so as to achieve a predetermined reduction rate. When the distance from other groups is far or the number of types of causes included in the group is small, it is easy to estimate the cause from the partial discharge signal. However, when the distance from other groups is close or the number of types of causes included in the group is large, it is difficult to estimate the cause from the partial discharge signal. Therefore, in the method for automatically setting the threshold value set for each group, for example, when the distance from other groups is far or the number of types of causes included in the group is small, the threshold value is set to a small value, and when the distance from other groups is close or the number of types of causes included in the group is large, the threshold value is set to a large value.

[0036] Note that the data reduction unit 114 may reduce the number of partial discharge data based on a predetermined threshold value and a predetermined reduction rate regardless of the cause of the partial discharge.

[0037] FIG. 8 is a diagram showing the configuration of the estimation model generation device 12 according to the present embodiment. The estimation model generation device 12 generates an estimation model for estimating the cause of partial discharge. The estimation model generation device 12 includes a teacher data generation unit 121, an estimation model generation unit 122, and an estimation model output unit 123.

[0038] The teacher data generation unit 121 generates teacher data based on the input partial discharge data. The teacher data generation unit 121 adds known noise data to the partial discharge signal of the partial discharge data, and generates teacher data in which the partial discharge cause data and the partial discharge signal to which the noise data is added are associated. Here, preprocessing before learning may be performed on the teacher data.

[0039] The estimation model generation unit 122 generates an estimation model based on the teacher data generated by the teacher data generation unit 121. The estimation model generation unit 122 generates an estimation model by inputting the teacher data into a machine learning algorithm such as a neural network. The estimation model is a model that takes a partial discharge signal as an input and outputs an estimation result of the cause of the partial discharge. The output estimation result indicates, for example, each cause of the partial discharge by probability. Taking a specific example, the output estimation result is, for example, that the probability that the cause of the partial discharge is a metal foreign object is 55%, the probability that the cause of the partial discharge is a void is 30%, and the probability that the cause of the partial discharge is a winding is 15%. The estimation result may present one cause of the estimated partial discharge.

[0040] The estimation model output unit 123 outputs the generated estimation model. The output estimation model is stored in the estimation device 21 described later.

[0041] FIG. 9 is a flowchart showing the operation of the estimation model generation device 12 according to the present embodiment. The teacher data generation unit 121 generates teacher data based on the partial discharge data input from the data selection device 11 (step S11). The estimation model generation unit 122 generates an estimation model trained to take the partial discharge data among the teacher data as an input and output the partial discharge cause data (step S12). The estimation model output unit 123 outputs the generated estimation model (step S13).

[0042] In the estimation model generation device 12, the time required to generate the estimation model increases as the number of teacher data increases, and usually is proportional to the number of teacher data. In the present embodiment, the partial discharge signals measured by the partial discharge sensor 10 are selected by the data selection device 11. Therefore, compared with the case of generating teacher data from all the partial discharge signals measured by the partial discharge sensor 10 and generating an estimation model, in the present embodiment, the number of teacher data can be reduced, and the time required to generate the estimation model can be shortened. Further, since the data reduced by the data selection device 11 are similar data, the accuracy of the estimation model generated by the estimation model generation device 12 does not decrease, but rather the occurrence of overfitting can be prevented, and the accuracy of the estimation model can be improved.

[0043] FIG. 10 is a diagram showing the configuration of the estimation system 2 according to the present embodiment. The estimation system 2 estimates the cause of partial discharge in the device to be estimated based on the partial discharge signal of the device E to be estimated. The estimation system 2 includes a partial discharge sensor 20 and an estimation device 21.

[0044] The partial discharge sensor 20 is provided in the device E to be estimated, and measures a signal generated by partial discharge of the device E to be estimated. The partial discharge sensor 20 may be the same sensor as the partial discharge sensor 10 in the estimation model generation system 1.

[0045] The estimation device 21 estimates the cause of partial discharge based on the partial discharge signal measured by the partial discharge sensor 20. The estimation device 21 includes a measurement data acquisition unit 211, an estimation unit 212, an output unit 213, and a storage unit 219. The storage unit 219 stores the estimation model input from the estimation model generation device 12.

[0046] The measurement data acquisition unit 211 acquires the partial discharge signal measured by the partial discharge sensor 20. The measurement data acquisition unit 211 stores the acquired signal in the storage unit 219.

[0047] The estimation unit 212 estimates the cause of the partial discharge based on the partial discharge signal acquired by the measurement data acquisition unit 211. The estimation unit 212 estimates the cause of the partial discharge by inputting the partial discharge signal into an estimation model stored in the storage unit 219.

[0048] The output unit 213 outputs the estimation result. The output from the output unit 213 is displayed, for example, by a display device. The output from the output unit 213 is stored, for example, in an external storage device.

[0049] FIG. 11 is a flowchart showing the operation of the estimation device 21 according to this embodiment. The measurement data acquisition unit 211 acquires a partial discharge signal measured by the partial discharge sensor 20 (step S21). The estimation unit 212 estimates the cause of the partial discharge based on the partial discharge signal acquired by the measurement data acquisition unit 211 (step S22). The output unit 213 outputs the estimation result by the estimation unit 212 (step S23).

[0050] As described above, an embodiment of the present invention has been described in detail with reference to the drawings. However, the specific configuration is not limited to the above, and various design changes and the like can be made without departing from the gist of the present invention.

[0051] Some or all of the data selection device 11, the estimation model generation device 12, and the estimation device 21 in the above-described embodiment may be implemented by a computer. In that case, a program for realizing this function may be recorded on a computer-readable recording medium, and the program recorded on this recording medium may be read into a computer system and executed to realize it. Here, the “computer system” is assumed to include the OS and the hardware of peripheral devices. Further, the “computer-readable recording medium” refers to a portable medium such as a flexible disk, a magneto-optical disk, a ROM, a CD-ROM, etc., and a recording device such as a hard disk built in a computer system. Furthermore, the “computer-readable recording medium” refers to a communication line such as when transmitting a program via a network such as the Internet or a communication line such as a telephone line, which dynamically holds the program for a short time, and in that case, a server or a client It may also include a volatile memory inside a computer system that holds a program for a certain period of time. Also, the above program may be for realizing a part of the above-described functions, and furthermore, it may be possible to realize the above-described functions in combination with a program already recorded in the computer system. Also, some or all of the data selection device 11, the estimation model generation device 12, and the estimation device 21 may be realized using a programmable logic device such as an FPGA (Field Programmable Gate Array).

[0052] Although some embodiments of the present invention have been described, these embodiments are presented by way of example and are not intended to limit the scope of the invention. These embodiments can be implemented in various other forms, and various omissions, replacements, and changes can be made without departing from the gist of the invention. These embodiments and their modifications are included in the scope and gist of the invention, and are also included in the invention described in the claims and the equivalent scope thereof.

Explanation of Signs

[0053] 1 Presumed model generation system, 10 Partial discharge sensor, 11 Data selection device, 111 Partial discharge data storage unit, 112 Grouping unit, 113 Grouping result storage unit, 114 Data reduction unit, 115 Reduction method reception unit, 116 Data output unit, 12 Presumed model generation device, 121 Teacher data generation unit, 122 Presumed model generation unit, 123 Presumed model output unit, 2 Presumption system, 21 Presumption device, 211 Measurement data acquisition unit, 212 Presumption unit, 213 Output unit, 219 Storage unit

Claims

1. A grouping unit that groups partial discharge data including a partial discharge signal measured in a device where partial discharge has occurred and partial discharge cause data indicating the cause of the partial discharge into each group based on the partial discharge signal, A data reduction unit that reduces the partial discharge data based on the number of pieces of partial discharge data grouped into each group and a predetermined threshold value, A data selection device comprising:

2. The grouping unit extracts a feature amount from the partial discharge signal and groups the partial discharge data based on the feature amount, The data selection device according to Claim 1.

3. The grouping unit groups the partial discharges for each cause of the partial discharge, The data reduction unit reduces the partial discharge data for each cause of the partial discharge, The data selection device according to Claim 1 or 2.

4. The threshold value is set to achieve a predetermined reduction rate, The data selection device according to Claim 1 or 2.

5. A reduction method reception unit that displays the grouping result by the grouping unit and receives an input of the threshold value, Further comprising: The data selection device according to Claim 1.

6. The reduction method reception unit displays the number of pieces of partial discharge data in a graph, The data selection device according to Claim 5.

7. A grouping step of grouping partial discharge data including a partial discharge signal measured in a device where partial discharge has occurred and partial discharge cause data indicating the cause of the partial discharge into each group based on the partial discharge signal, A data reduction step of reducing the partial discharge data based on the number of pieces of partial discharge data grouped into each group and a predetermined threshold value, A data selection method having:

8. A program for causing a computer to execute the data selection method according to Claim 7.

Images