JPH07181219A - Partial discharge detection method and device - Google Patents

Abstract

PURPOSE:To enable the detection of a partial discharge accurately and at a higher speed without being affected by any noise by a method wherein a frequency data with limited noises is taught to a neural network, features are extracted in a frequency characteristic data thereof when a partial discharge is detected and the features are inputted to the neural network to judge the generation of at least the partial discharge. CONSTITUTION:When a frequency characteristic data of a partial discharge to be outputted from a partial discharge detector 9 is detected, the extraction of features thereof and standardization thereof are performed. A frequency area with limited noises is selected from the frequency characteristic data subjected to the extraction of features and standardization to be turned to a pattern and the resulting data is inputted into a neural network device 20. The device 20 judges the generation of any partial discharge based on the frequency characteristic data patternized and the frequency characteristic data learned. This enables the detection of a partial discharge generated in an insulation equipment accurately and at a higher speed without being affected by noises.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a partial discharge detection method and apparatus for detecting partial discharge in a gas-insulated equipment or the like using an insulating medium such as SF ₆ gas as an insulating medium.

[0002]

2. Description of the Related Art In recent years, gas-insulated equipment as electric power equipment constituting substations and the like has a high-voltage conductor disposed in a metal container at ground potential, and a gas having excellent insulation performance, For example, it is formed by compressing and filling SF ₆ gas, and its adoption is steadily increasing.

This SF ₆ gas exhibits very excellent insulation characteristics under an equal electric field, but the insulation characteristics are extremely deteriorated under an unequal electric field. In gas-insulated equipment using such an insulating gas, various factors may be considered as disturbing the electric field distribution in the container. The main factors are defects such as dents on the surface of the high-voltage conductor and assembly. For example, metallic foreign matters mixed in at the time of transportation or during transportation. or,
It is considered that one of the factors is defective contact of the high-voltage conductor portion due to an assembly error and defects such as voids in the insulating spacer.

Due to the above defects, if an unequal electric field is formed inside the gas-insulated equipment, partial discharge may occur in the operating state, and eventually there is a possibility that the entire road may be destroyed. Therefore, it is necessary to surely detect the partial discharge before the all-road breakdown and prevent the all-road breakdown in advance.

From the above background, various partial discharge detection methods have been developed for detecting the partial discharge generated inside the gas-insulated equipment. One of the partial discharge detection methods is a method of detecting a partial discharge by extracting a signal of a specific frequency that is likely to cause a partial discharge and is less likely to contain noise.

FIG. 5 is a block diagram of an example to which such a partial discharge detection method is applied. That is, a general gas insulation device is housed in a metal container 1, and the metal container 1 is electrically divided by an insulating spacer 2 and connected in the longitudinal direction.

A high-voltage conductor 3, which is a high-voltage charging section electrically connected to a power transmission line (not shown), is supported in the metal container 1 by the insulating spacer 2. The metal container 1 is grounded by a ground wire (not shown).

A buried electrode 4 for voltage detection is provided on the insulating spacer 2, and a stray capacitance C2 exists between the electrode 4 and the metal container 1. The stray capacitance C2 serves as a capacitor 5. It is shown. And this capacitor 5
A partial discharge detector 7 is connected to both terminals via a signal lead-in wire 6.

In this partial discharge detector 7, an amplifier circuit 9 is connected to a filter 8 for taking out a signal of a specific frequency,
Further, the peak detector and the integrating circuit 10 are connected and configured. In addition, the power supply 11
Are connected.

With such a structure, when a high voltage is applied to the high voltage conductor 3, the voltage divider is formed by the stray capacitance C1 and the capacitor 5 existing between the high voltage conductor 3 and the electrode 4, and the capacitor 5 is formed. A shared voltage is generated at both ends of. When a partial discharge occurs in the insulating device, a high-frequency component (signal) resulting from the discharge is superimposed on this shared voltage and is input to the partial discharge detector 7.

The partial discharge detector 7 takes out a signal of a specific frequency corresponding to a partial discharge pulse by a filter 8, amplifies it by an amplifier circuit 9, and then outputs it to the outside through a peak detector and an integrating circuit 10. Then, the output signal detects that partial discharge has occurred inside the insulating device.

However, in the above partial discharge detection method,
A signal of a specific frequency, which is said to be prone to partial discharge and less likely to contain noise, is taken out, but from the analysis results of many data actually collected at multiple substations, the frequency of partial discharge is shown. It has been known that the characteristics and the frequency characteristics of noise differ depending on the configuration and environment of the substation, the connection position of the partial discharge detector, and the like.

That is, there is a possibility that noise may be included in the partial discharge signal detected by the conventional partial discharge detection method, and it is difficult to perform accurate partial discharge detection.

On the other hand, although the partial discharge waveform and the noise waveform differ depending on the substation and the measurement point, the noise frequency distribution at a specific measurement point can be patterned into several patterns, and frequency distributions that do not fit these patterns can be used. When the waveform is measured, it is considered to be a waveform due to partial discharge.

As described above, in order to pattern the frequency distribution of noise different for each measurement point and discriminate between noise and partial discharge from the input frequency distribution, a wealth of knowledge about partial discharge is required, and the present situation is concerned. It is done by knowledgeable experts.

[0016] However, the diagnosis by the expert may have an individual difference among the experts, or the knowledge, experience, and reproducibility are limited, and thus human error may occur. In particular, these problems are highlighted when diagnosing an unknown frequency distribution.

The above-mentioned problems are caused not only by the gas-insulated equipment but also by the partial discharge of the oil-filled insulation equipment in which the high voltage portion is arranged in the metal container of the ground potential in which the insulating oil is sealed. The same occurs when the detection is performed by the same method.

[0018]

As described above, in the detection of partial discharge, it is difficult to detect the partial discharge signal with high accuracy because the partial discharge signal may contain noise, and it is difficult for an expert to detect it. In the discrimination, human error may occur.

Therefore, according to the present invention, the partial discharge generated in the insulating device can be detected accurately and at high speed without being influenced by noise, and the unknown partial discharge waveform can also be accurately and quickly detected. An object of the present invention is to provide a highly accurate partial discharge detection method capable of detecting partial discharge and an apparatus therefor.

[0020]

According to a first aspect of the present invention, there is provided a partial discharge detection method for detecting a partial discharge generated in an insulating device in which a high voltage part is arranged in a ground potential metal container containing an insulating medium. , The frequency region with less noise is selected from the frequency characteristic data of the detection waveform of the partial discharge that may occur in the insulation device, this frequency data is learned to the neural network, and the partial discharge is detected. In this partial discharge detection method, at least the characteristics of the frequency characteristic data are extracted and input to a neural network, and at least partial discharge occurrence determination is performed by the neural network to achieve the above object.

According to the second aspect, the simulated partial discharge signal is injected into the insulating device, the partial discharge waveform by the simulated partial discharge signal is detected, and the frequency characteristic data is learned by the neural network.

According to the third aspect, the noise signal is injected into the insulating device, the waveform of the noise signal is detected, and the frequency characteristic data is learned by the neural network.

According to the fourth aspect, at least two frequency regions having less noise are selected from the frequency characteristic data. According to the fifth aspect, the frequency characteristic data is detected from two or more parts in the insulating device, and at least two or more frequency regions having less noise are selected from the frequency characteristic data and input to the neural network.

According to the sixth aspect, the partial discharge detecting device for detecting the partial discharge generated in the insulating device in which the high voltage part is arranged in the metal container of the ground potential in which the insulating medium is sealed corresponds to the partial discharge. Partial discharge detector that extracts a specific frequency signal and frequency characteristic data that selects at least a frequency region with less noise from the frequency characteristic data of the detection waveform of partial discharge that may occur in the insulating device are learned in advance, and The above-described object is achieved by including a frequency characteristic data of the partial discharge detected by the discharge detector, and a neural network means for determining the occurrence of the partial discharge based on the frequency characteristic data and the frequency characteristic data learned in advance. This is a partial discharge detection device to be tried.

According to the seventh aspect, the neural network means determines the occurrence of the partial discharge based on the frequency characteristic data of the partial discharge detected by the partial discharge detector and the frequency characteristic data learned in advance, and the partial discharge is generated. It has a function to analyze the factors of occurrence of.

[0026]

According to the first aspect of the present invention, the neural network is made to learn by selecting the frequency region having less noise from the frequency characteristic data of the detection waveform of the partial discharge that may occur in the insulating device. When partial discharge is detected in this state,
At least the characteristics of the frequency characteristic data of the partial discharge are extracted and input to the neural network, where at least the partial discharge occurrence determination is performed. This makes it possible to accurately and quickly determine the occurrence of partial discharge.

In this case, according to the second aspect, the learning to the neural network is performed by injecting the simulated partial discharge signal into the insulating device, detecting the partial discharge waveform by the simulated partial discharge signal, and obtaining the frequency characteristic data thereof. According to the present invention, the noise signal is injected into the insulating device, the waveform of the noise signal is detected, and the frequency characteristic data is learned.

According to the fourth aspect, at least two or more frequency regions having less noise are selected from the frequency characteristic data. This increases the detection accuracy. According to the fifth aspect, the frequency characteristic data is detected from two or more parts of the insulating device, and at least two or more frequency regions with less noise are selected and input to the neural network.

According to the sixth aspect, when the specific frequency signal corresponding to the partial discharge is taken out by the partial discharge detector, the detection output of the partial discharge detector is inputted to the neural network means, and based on the frequency characteristic data. At least a partial discharge occurrence determination is performed.

In this case, according to the seventh aspect, the neural network means determines the occurrence of the partial discharge based on the frequency characteristic data of the partial discharge and the frequency characteristic data learned in advance and analyzes the cause of the occurrence.

[0031]

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS An embodiment of the present invention will be described below with reference to the drawings. The same parts as those in FIG. 5 are designated by the same reference numerals, and detailed description thereof will be omitted. FIG. 1 is a configuration diagram of a partial discharge detection device applied to a gas insulation device. The neural network device 2 is connected to the output terminal of the partial discharge detector 7.
0 is connected.

The detection output of the partial discharge detector 7, that is, the frequency characteristic data is detected by means (not shown), and thereafter, characteristic extraction and standardization are performed on the frequency characteristic data, and the extracted frequency is used. A frequency region with less noise is selected from the characteristic data, and the selected frequency region data is patterned and input to the neural network device 20.

This neural network device 20 is
A frequency region with less noise is selected from the frequency characteristic data of the partial discharge waveform that may occur in the gas insulated device, the frequency characteristic data is learned in advance, and the partial discharge based on the detection output of the partial discharge detector 7 is performed. It has a function of determining the occurrence of partial discharge by comparing the characteristic of the frequency characteristic data of 1) with the characteristic of the frequency characteristic data learned in advance.

Further, the neural network device 20 is
The relationship between the frequency characteristic data of partial discharge and the cause of partial discharge has been learned. Specifically, the neural network device 20 has a function of comparing the characteristic of the frequency characteristic data of the patterned partial discharge with the characteristic of the frequency characteristic data learned in advance to determine the occurrence of the partial discharge,
It has a function of comparing the characteristic of the frequency characteristic data of the patterned partial discharge with the factor of the partial discharge based on the frequency characteristic data learned in advance and analyzing the factor of the partial discharge.

Next, the operation of the device configured as described above will be described. The neural network device 20 has learned in advance the frequency characteristic data of a large number of partial discharge waveforms and the frequency characteristic data of a noise waveform that may occur in the gas insulation device.

Therefore, a learning method for these partial discharges will be described with reference to FIG. The partial discharge detector 7 is connected to the embedded electrode 4a of the one insulating spacer 2a of the metal container 1 at the ground potential via the signal lead-in wire 6, and the embedded electrode 4b of the other insulating spacer 2b of the metal container 1 is connected. The simulated partial discharge signal generator 13 is connected via the signal injection line 12.

In this state, the simulated partial discharge signal generator 13 injects the simulated partial discharge signal through the signal injection line 12. In this case, the simulated partial discharge signal to be injected is several ns which is equivalent to the actual partial discharge waveform generated inside the gas-insulated equipment.
The signal has a rising edge of.

The simulated partial discharge signal thus injected into the buried electrode 4b is the stray capacitance C2 existing between the buried electrode 4b and the metal container 1, the buried electrode 4b and the high voltage conductor 3.
After entering the inside of the gas-insulated device via the voltage divider formed by the stray capacitance C1 existing between and, it propagates left and right in the figure at the same speed as the actual partial discharge signal.

The propagated signal is transmitted to the metal container 1
Via the voltage divider formed by the embedded electrode 4a of the other insulating spacer 2a, is superimposed on the shared voltage of the capacitor 5, is input to the partial discharge detector 7 via the signal lead-in line 6, and is detected as a partial discharge waveform. To be done.

Subsequently, the characteristics of the frequency characteristics of the partial discharge waveform are extracted by the simulated partial discharge signal which is the detection output of the partial discharge detector 7, and further standardized to form the frequency distribution pattern, and the frequency distribution pattern The data is obtained, and this data is learned by the neural network device 20 as the data of the frequency characteristic (distribution) pattern of the partial discharge waveform.

With such a method, the neural network device 20 similarly learns frequency characteristic data of many partial discharge waveforms by injecting simulated partial discharge signals from a plurality of embedded electrodes. The neural network device 20 also learns a large number of noise waveform frequency characteristic data in the same manner as for the noise waveform frequency characteristic data.

Next, referring to the processing procedure flow chart shown in FIG. 3, the processing procedure for determining the presence / absence of occurrence of partial discharge and the cause of occurrence of partial discharge by using the neural network device 20 which has learned many partial discharge waveforms. Explain.

When a high voltage is applied to the high-voltage conductor 3, a voltage divider formed by the stray capacitance C1 existing between the high-voltage conductor 3 and the electrode 4 and the capacitor 5 causes a shared voltage across the capacitor 5. Occurs. And step # 1
When a partial discharge occurs in the insulating device, a high-frequency component (signal) resulting from the discharge is superimposed on this shared voltage and is input to the partial discharge detector 7.

The partial discharge detector 7 takes out a signal of a specific frequency corresponding to the partial discharge pulse by the filter 8, amplifies it by the amplifier circuit 9, and then outputs it to the outside through the peak detector and the integrating circuit 10.

The detection output of the partial discharge detector 7 is the frequency characteristic data of the partial discharge waveform, an example of which is shown in FIG. Next, the following processing is performed. That is, when the frequency characteristic data of the partial discharge output from the partial discharge detector 7 is detected in step # 2, the characteristic extraction and standardization of the frequency characteristic data are performed in the next step # 3.

Next, in step # 4, the frequency region with less noise is selected from the frequency characteristic data obtained by the feature extraction and standardization, and the data of the frequency region selected in the next step # 5 is patterned. This data is input to the neural network device 20 in step # 6.

Next, the neural network device 20
In step # 7, the occurrence of partial discharge is determined based on the patterned frequency characteristic data and the frequency characteristic data learned in advance.

Next, the neural network device 20
In step # 8, the cause of partial discharge is analyzed based on the patterned frequency characteristic data and the frequency characteristic data learned in advance.

Finally, the neural network device 20
Indicates in step # 9 the partial discharge occurrence determination result,
The analysis result of the cause of partial discharge is displayed and output. As described above, in the above-described embodiment, the frequency region having less noise is selected from the frequency characteristic data of the detection waveform of the partial discharge that may occur in the gas insulation device, and the frequency characteristic data is sent to the neural network device 20. Since learning is performed to determine the occurrence of partial discharge, even if noise may be included in the partial discharge signal, the determination of partial discharge occurrence can be performed quickly and accurately, and in addition, partial discharge occurrence can be determined. There is no human error in the judgment.

By causing the neural network device 20 to learn the cause of partial discharge, the cause of partial discharge can also be determined. Furthermore, not only the known partial discharge waveform but also the unknown partial discharge waveform can be detected at high speed and with high reliability.

Further, since a large amount of data is not required for data determination, it is easy to systemize using a processing device such as a personal computer, and it is possible to carry it to the site or the like, which is excellent in practicality. .

Further, if at least two frequency regions having less noise are selected from the frequency characteristic data of partial discharge, the partial discharge can be detected with higher accuracy. On the other hand, even if the neural network device 20 detects frequency characteristic data from two or more locations in the insulating device and selects at least two frequency regions with less noise from these frequency characteristic data and learns, the higher frequency is obtained. Accurate partial discharge detection is possible.

The present invention is not limited to the above-mentioned embodiment but may be modified as follows. For example, the specific contents of the processing procedure, the specific device to be used, and the like can be appropriately selected.

Further, the present invention is not limited to gas insulation equipment, but can be applied as a partial discharge detection method for oil-filled insulation equipment in which a high-voltage part is arranged in a grounded metal container in which insulating oil is sealed. As in the case of the one embodiment described above, excellent effects can be obtained.

[0055]

As described above in detail, according to the present invention, it is possible to accurately and quickly detect a partial discharge generated in an insulating device without being influenced by noise, and to detect an unknown partial discharge waveform. It is also possible to provide a highly accurate partial discharge detection method and its device capable of detecting partial discharge accurately and at high speed.

[Brief description of drawings]

FIG. 1 is a configuration diagram showing an embodiment of a gas insulation device to which a partial discharge detection method according to the present invention is applied.

FIG. 2 is a diagram for explaining a learning method for a neural network device.

FIG. 3 is a flowchart showing a processing sequence of a partial discharge detection method.

FIG. 4 is a diagram showing frequency characteristic data of a partial discharge waveform.

FIG. 5 is a diagram for explaining a conventional partial discharge detection method.

[Explanation of symbols]

1 ... Metal container, 2 ... Insulating spacer, 3 ... High-voltage conductor, 4
... Buried electrode, 5 ... Capacitor, 6 ... Signal lead-in wire, 7
... partial discharge detector, 12 ... signal injection line, 13 ... simulated partial discharge signal generator, 20 ... neural network device.

 ─────────────────────────────────────────────────── ─── Continuation of the front page (72) Inventor Isao Oki 1-10-10 Irifune, Chuo-ku, Tokyo Tokyo Electric Power Co., Inc. System Research Laboratory (72) Taku Saito 2 Ukishima-cho, Kawasaki-ku, Kawasaki-shi, Kanagawa No. 1 Incorporated company Toshiba Hamakawasaki factory (72) Inventor Seiji Wakabayashi No. 2 Ukishimacho, Kawasaki-ku, Kawasaki-shi, Kanagawa Prefecture No. 2 Incorporated company Toshiba Hamakawasaki factory (72) Inventor ▲ Takaaki Sakaki Kawasaki, Kawasaki, Kanagawa 2-1-1 Ukishima-cho, Ukishima-cho, Ltd. Inside the Toshiba Hamakawasaki Plant (72) Inventor Shiro Maruyama 2-1-1, Ukishima-cho, Kawasaki-ku, Kawasaki-shi, Kanagawa (72) Incorporated Toshiba Satoshi Nagaoka Tokyo Metropolitan Port 1-1-1 Shibaura, Ward Toshiba Head Office Office

Claims (7)

[Claims] 1. A partial discharge detection method for detecting a partial discharge generated in an insulating device in which a high-voltage part is arranged in a grounded metal container having an insulating medium sealed therein, the method comprising: A frequency region with less noise is selected from the frequency characteristic data of a certain partial discharge waveform, this frequency characteristic data is learned in a neural network, and when the partial discharge is detected, at least the characteristic extraction of the frequency characteristic data of this partial discharge is performed. A method for detecting partial discharge, characterized by inputting to the neural network and determining at least the occurrence of the partial discharge by the neural network. 2. The method according to claim 1, wherein a simulated partial discharge signal is injected into the insulating device, a partial discharge waveform due to the simulated partial discharge signal is detected, and the frequency characteristic data is learned by a neural network. Partial discharge detection method. 3. The partial discharge detection method according to claim 1, wherein a noise signal is injected into the insulating device, a waveform due to the noise signal is detected, and the frequency characteristic data is learned by a neural network. 4. The partial discharge detection method according to claim 1, wherein at least two frequency regions having less noise are selected from the frequency characteristic data. 5. The frequency characteristic data is stored in the insulation device 2
2. The partial discharge detection method according to claim 1, wherein at least two or more frequency regions having less noise are selected from these frequency characteristic data and input to the neural network. 6. A partial discharge detection device for detecting a partial discharge generated in an insulating device in which a high voltage part is arranged in a grounded metal container enclosing an insulating medium, wherein a specific frequency signal corresponding to the partial discharge is detected. Partial discharge detector to be taken out, pre-learning frequency characteristic data selected at least a frequency region with less noise from the frequency characteristic data of the partial discharge waveform that may occur in the insulating device, and by the partial discharge detector The neural network means for inputting the detected frequency characteristic data of the partial discharge and at least determining the occurrence of the partial discharge based on the frequency characteristic data learned in advance and the frequency characteristic data learned in advance. Partial discharge detection device. 7. The neural network means determines the occurrence of the partial discharge based on the frequency characteristic data of the partial discharge detected by the partial discharge detector and the frequency characteristic data learned in advance, and the cause of the partial discharge. 7. The partial discharge detection device according to claim 6, which has a function of analyzing.