JPH06213958A - Judgment of partial discharge - Google Patents

Abstract

(57) [Abstract] [Purpose] In a partial discharge identification system using a neural network, particularly to improve an input signal to be learned by the neural network. [Configuration] When determining the presence or absence of partial discharge using a neural network, only partial discharge signals generated from various high-voltage devices (including power cables), and the actual site where the high-voltage device is installed or its vicinity The noise measured in the above is recorded separately, and the signal obtained by combining the above partial discharge signal and the above noise is used as the input signal when there is partial discharge, and the above noise is input when there is no partial discharge. The neural network was made to learn as a signal, and the presence or absence of partial discharge was judged.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a partial discharge identifying system using a neural network, and more particularly to an input signal to be learned by the neural network.

[0002]

2. Description of the Related Art There is a method of performing insulation diagnosis of high-voltage equipment, power cables and connection parts by measuring partial discharge. However, it is often difficult to determine the presence / absence of a partial discharge signal due to external noise. By the way, in recent years, a system has been proposed which uses a neural network to determine whether or not partial discharge has occurred. (For example, JP-A-2-29612 and JP-A-2-2970.
(See Japanese Patent Laid-Open No. 74-74370 and Japanese Patent Laid-Open No. 2-238370) Here, an example of identifying partial discharge generated from a connecting portion of a power cable will be described.

FIG. 2 is an explanatory view for identifying a partial discharge of a power cable connecting portion. A power cable 2 is connected to the charging transformer 1, and in this example, an insulation connection portion 3 which is a partial discharge measurement target is connected in the middle. The partial discharge signal from the insulating connection portion 3 is generated in the detection impedance 5 via the detection electrode 4. Then, the detected partial discharge signal is input to the spectrum analyzer 6 and S /
The output signal component corresponding to the frequency with a good N ratio is detected by the detection circuit 6
After being detected by ', the signal is input to the partial discharge measuring instrument 7. Information on the applied voltage phase of the power cable is also input to the partial discharge measuring device 7, and the magnitude (amplitude) of the partial discharge signal, the partial discharge occurrence phase with respect to the applied voltage, and the partial discharge occurrence frequency are stored. These pieces of information are input to the neural network 8 and the display 9 displays the result of determining whether or not partial discharge has occurred.

Here, the applied voltage phase is taken out from the tertiary winding 10 of the applied transformer 1. A computer is used for the neural network 8 and the display 9, and the computer controls the partial discharge measuring instrument 7 and transmits a signal to the computer. In addition, there is also a method of forming a waveform pattern of the generated partial discharge and inputting this pattern to the neural network 8.

FIG. 3 is an explanatory diagram for explaining the structure and function of the neural network 8. Here, as an example, the input layer 11, the intermediate layer 12, and the output layer 13 have a three-layer structure, and the cells 111 to 113 of the input layer 11 are connected to the cells 121 to 123 of the intermediate layer 12, respectively. Each of the cells 121 to 123 of the intermediate layer 12 is the cell 1 of the output layer 13.
It is connected to 31 (with partial discharge) and 132 (without partial discharge). Then, the cells 111 to 11 of the input layer 11
3 corresponds to a specific phase range (for example, 0 ° to 18 °) and an amplitude range (for example, 10 to 20 pC) among the information stored in the partial discharge measuring instrument 7, and is a signal that satisfies the range. The number becomes the input value for each cell. When a value is input to each cell of the input layer 11, evaluation is performed in the intermediate layer 12 and the output layer 13, and evaluation values are obtained in the cells 131 and 132 of the output layer 13. The presence or absence of partial discharge is determined based on the magnitude relationship of these evaluation values.

Next, a learning method of the neural network will be described. First, when a sample having a simulated defect is charged to generate partial discharge and the partial discharge signal detected by this is input to the neural network 8, the output value of the cell 131 of the output layer 13 is "1". , So that the output value of the cell 132 becomes “0”, and when a partial discharge is not generated using a sound sample, the cell 131 of the output layer 13
The learning of the neural network is performed so that the output value of the cell is “0” and the output value of the cell 132 is “1”. Then, after the learning is performed a sufficient number of times, the presence or absence of partial discharge is determined for the new sample.

[0007]

By the way, noise in a local high-voltage equipment (including a power cable, etc.) under a hot line condition is extremely dependent on the laying condition, the surrounding environment, the use condition and the time. Diverse. However, in the prior art, a simulated partial discharge signal S1 as shown in FIG. 4 generated from a sample including a simulated defect and a noise-only signal N1 as shown in FIG. 5 which is measured from a sound sample and has little noise. Is only learned by the neural network.

Therefore, for example, when the noise signal N2 is generated from the local power cable as shown in FIG. 6, it is determined whether the partial discharge is generated by using the neural network learned by the conventional learning method. The identification is problematic in terms of reliability. Further, even if only one noise pattern is learned, the noise is diversified depending on the place of use, the time when it is detected, etc. Therefore, it is very difficult to actually determine the presence or absence of partial discharge.

The present invention has been made in view of the above circumstances, and in order to improve the reliability of the partial discharge identification technique,
The purpose is to provide an input signal for a neural network to learn.

[0010]

According to the present invention, when a neural network is applied to determine the presence or absence of partial discharge, only partial discharge signals generated from various high-voltage devices (including power cables) and actual high-voltage signals are detected. Noise measured at or near the site where the voltage equipment is installed,
Are recorded separately, and the neural network learns the signal obtained by combining the noise with the partial discharge signal as the input signal when there is partial discharge, and the noise as the input signal when there is no partial discharge. This is a partial discharge identification method for determining the presence or absence of discharge.

[0011]

By performing learning of the neural network using the signals corresponding to the site where various high voltage devices are actually installed, the accuracy of partial discharge determination is improved and high reliability can be obtained. .

[0012]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the partial discharge identifying method of the present invention will be described below with reference to the drawings. The system for detecting partial discharge is the same as that shown in FIGS. 2 and 3 described in the above-mentioned prior art. That is, FIG. 1 is a circuit diagram for determining partial discharge including a local signal synthesizing circuit. The signal from the partial discharge measuring instrument 7 is input to the local on-site signal synthesizing circuit 14, and the output signal of the on-site signal synthesizing circuit 14 is discriminated by the neural network 8 and displayed on the display 9. Next, a specific example of a signal to be learned by the neural network according to the embodiment of the present invention will be described.

C containing a void with a diameter of 5 mm in the laboratory
Create a V-cable. The partial discharge signal S2 detected from this CV cable is as shown in FIG. This partial discharge signal S2 is digitized by an A / D converter.
This digitized signal is recorded on the magneto-optical disk. At this time, when detecting a signal, noise mixed from the surroundings should be minimized. Therefore, this signal S2 can be regarded as a pure partial discharge signal.

At the site where the power cable is actually laid, the signal is measured by the measuring method shown in FIG. At this time, the power cable is not charged. Therefore, the detected signal is as shown in FIG.
As shown in (B), (C), and (D), noises that purely enter from the outside are shown. These signals N2 are also digitized and recorded on the magneto-optical disk as described above.

By the way, as described in the section of means for solving the above problems, noise changes with time, so that
The signal is detected and recorded at intervals of 0 minutes.
The signal (noise from the outside) measured from the power cable in the non-energized state is continuously measured for 24 hours.
There are a total of 144 signals N2. In order to distinguish these signals, N2 (1), N2 (2), ..., N2 (14
4) Describe as follows.

This is measured continuously for several days. Next, the averaging process is performed using the signals corresponding to the respective periods. That is, as shown in FIGS. 9A, 9B, 9C and 9D, N2 (1) measured on the first day, N2 (1) measured on the second day, and third day. Add the measured N2 (1) and divide this by the number of measurement days to obtain the average value. That is,
The total of N2 (1) measured by the Nth day is N2 (1) '
And is recorded on the magneto-optical disk as noise in this measured time zone. This noise N2 (1) 'is shown in FIG.
Shown in.

The neural network 8 is made to learn using the signals recorded on these magneto-optical disks.
When learning the signal when there is a partial discharge signal, the neural network 8 is made to learn the signal S3 (1) obtained by combining the signals S2 and N2 (1) 'as in the prior art. (See FIG. 1) The method for making the neural network 8 learn is the same as the prior art. Here, the noise N2 (1) averaged with the partial discharge signal S2 at the time of synthesis is used.
The ratio of 'is set to 1: 1 and is simply added.

At this time, the noise added to the signal S3 (1) is N
'Since the signal S3 (1) is _{N 2 (1)' 2 (} 1) becomes a signal corresponding to the time zone was measured. In this way
The signal obtained by synthesizing the signals S2 and N2 (3) 'is S3 (3)
As a result, the steps up to S3 (144) are created. And S
3 Neural network that learned (2), S3
A neural network in which each signal is learned, such as a neural network in which (3) is learned, is created. Therefore, these neural networks are neural networks corresponding to the time when the noise is measured.

Using the neural network learned by the above method, the partial discharge of the power cable under the live line condition is measured on site. Where the signal S3
(1) Neural network NN learned by
(1), signal S3 (2) is described as a neural network NN (2) trained. That is, there are NN (1) to NN (144).

Here, it is assumed that the time when the noise N2 (1) is measured is exactly 12:00. Then, NN (1) receives signal S3.
Since (1) is learned, it corresponds to the time when the noise N2 (1) was measured. Therefore, from 11:55 to 12:05, NN (1) is used for measurement.
Similarly, from 12:05 to 12:15 N
The measurement is performed using N (2), and the measurement is performed using the neural network corresponding to each time.

Here, a power cable line modeled on an actual on-site line was laid in a factory and measurements were performed. This power cable contains artificially created voids. FIG. 10 shows noise measured at 1 o'clock (here, N2 (6) is assumed). FIG. 11 shows noise measured at 8 o'clock (here, N2 (48) is assumed)
Is. FIG. 12 shows noise measured at 12:00 (here, N2 (72) is assumed). FIG. 13 shows the signal when there is almost no noise, and it can be confirmed that the partial discharge signal S is output.

By the way, when the partial discharge is identified by using the neural network created in the prior art, it is judged that there is partial discharge in FIG. 10, but FIG. 11 and FIG.
In No. 2, it was determined that there was no partial discharge. However, using the neural network created in this example, it was determined that there was partial discharge in any of FIGS. 10, 11 and 12. Thus, it was confirmed that the reliability of measurement was improved.

In the above example, attention is paid to the partial discharge signal generated from the power cable, but the identification of the partial discharge signal generated from other high voltage equipment (eg, GIS, rotating machine, transformer). Of course, it can be applied to. Further, when the partial discharge signal and the noise are combined, the present invention is
Although the ratio is set to 1: 1, it may be combined at a different ratio depending on the nature of the signal. Further, the synthesizing method is not limited to simply adding, and there are various methods such as multiplying by a certain coefficient. Further, although the present invention processes with a digital signal, this can also be performed with an analog signal.

In the above example, the data recording device uses the magneto-optical disk. However, it is also possible to use a magnetic tape or the like. Furthermore, in the present invention, the signal measured from the power cable in the non-power-supply state is used as noise, but this may be measured by another method. For example, noise may be measured by installing an antenna near the object to be measured.

[0025]

As described above, if the partial discharge identification method of the present invention is applied, the neural network is trained by using the noise actually generated in the field.
The presence / absence of partial discharge can be more reliably determined, and the reliability of the insulation diagnosis technique is significantly improved.

[Brief description of drawings]

FIG. 1 is a circuit diagram of a partial discharge determination circuit having a field signal synthesis circuit according to an embodiment of the present invention,

FIG. 2 is a circuit diagram of a prior art partial discharge determination circuit for a power cable connection,

FIG. 3 is an explanatory diagram for explaining a configuration and a function of the neural network in FIG.

FIG. 4 is a waveform diagram of a partial discharge signal measured from a sample including a simulated defect,

FIG. 5 is a waveform diagram of only noise when there is no partial discharge measured from a sound sample,

6] Waveform diagram of noise measured at the site,

FIG. 7 is a waveform diagram of a partial discharge signal measured from a sample including a simulated defect,

8 (A), (B), (C), and (D) are waveform diagrams of signals (external noise) measured from a power cable in a non-energized state,

9 (A), (B), (C), and (D) are the first day,
(E) is a waveform diagram of noise measured at the same time on the second day, the third day, and the Nth day, and (E) is a waveform diagram of a signal obtained by averaging the total measured on the Nth day by N.

FIG. 10 is a waveform diagram of a signal measured at 1 o'clock,

FIG. 11 is a waveform diagram of a signal measured at 8 o'clock,

FIG. 12 is a waveform diagram of a signal measured at 12:00,

FIG. 13 is a waveform diagram of a signal when there is almost no noise.

[Explanation of symbols]

 1 Power Supply Transformer 2 Power Cable 3 Connection Part 4 Detection Electrode 5 Detection Impedance 6 Spectrum Analyzer 6'Detection Circuit 7 Partial Discharge Measuring Instrument 8 Neural Network 9 Display 10 Tertiary Winding 11 Input Layer 12 Intermediate Layer 13 Output Layer 14 Local Use Signal synthesis circuit

Front page continuation (72) Inventor Akira Endo 5-1-1 Hidaka-cho, Hitachi-shi, Ibaraki "Inside the Hitachi Cable Power Systems Laboratory" (72) Inventor Hiroshi Suzuki 5-chome, Hidaka-cho, Ibaraki No. 1 "In Hitachi Cable Power Systems Laboratories"

Claims (1)

[Claims] 1. When determining the presence / absence of partial discharge using a neural network, only partial discharge signals generated from various high-voltage devices (including power cables) and a site where actual high-voltage devices are installed or When the noise measured in the vicinity is recorded separately, the signal obtained by combining the above partial discharge signal and the above noise is used as the input signal when there is partial discharge, and when the above noise is not partial discharge. A method for identifying partial discharge, characterized in that a neural network is used as an input signal for learning to determine the presence or absence of partial discharge.