JPH0735811A - Discriminating method for partial discharge signal and locating method for partial discharge occurrence position using same - Google Patents

Abstract

PURPOSE:To prevent erroneous discrimination of taking noise signal for partial discharge signal. CONSTITUTION:Two output signals detected at both ends of a cable 1 to be measured are stored into a data memory section 10 and rising positions of the output signals are detected with a data extracting section 20 from the two output signal data stored to match while a data with a fixed time range is extracted. Then, waveform data thus extracted are supplied to neutral networks 31A and 31B to judge whether the output signals are partial discharge signals or not. When both the signals are determined to be partial discharge signals, the extracted waveform data are converted to waveforms in a frequency area with a Fourier transform section 42 and the conversion data are supplied to neutral networks 41A and 41B to whether the output signals are partial discharge signals. As a result. when both the output signals are discriminated to be partial discharge signals again, the position where a partial discharge occurs is located by the use of a position locating section 32.

Description

Detailed Description of the Invention

[0001]

BACKGROUND OF THE INVENTION 1. Field of the Invention The present invention relates to a method of discriminating a partial discharge signal generated in a high voltage device such as a transformer or a high voltage cable, and using this method to locate a partial discharge occurrence position in the high voltage cable. Regarding the method.

[0002]

2. Description of the Related Art In a high voltage cable, for example, a CV cable, a high voltage is applied to the end of the cable to detect the partial discharge in order to investigate and examine the AC breakdown characteristics. Further, partial discharge is measured in order to diagnose the degree of insulation deterioration of the already installed high-voltage cable.
When a partial discharge is detected in a cable with a short strip length, the cable is divided into several sections and the partial discharge is measured again to find out where the insulation breakdown occurs. However, for long cables wound on drums, such a method is not efficient. In addition, such an isolation method is not suitable for the insulation characteristic test of the installed cable. Therefore, the partial discharge occurrence position is determined by utilizing the propagation time difference of the partial discharge signals.

The applicant of the present invention has proposed a method of locating a partial discharge occurrence position using a neural network in order to accurately and quickly locate the partial discharge occurrence position (Japanese Patent Application No. 4-93642). That is, this orientation method detects output signals from both ends of the cable to be measured and analyzes the waveform of each output signal by a neural network. A large number of partial discharge signals and typical external noise waveforms are learned in advance in this neural network, and it is determined whether the output signal is the partial discharge signal or the noise signal based on the learning result. Then, when the input waveforms obtained from both ends of the cable to be measured are both determined to be partial discharge signals, the position location calculation processing is performed from the difference in their propagation times and the propagation velocity, and the generation position of the partial discharge signal is located. .

[0004]

However, even in the position locating method using the above-mentioned neural network, the neural network may erroneously discriminate the noise signal from the partial discharge signal. The location of the partial discharge occurrence position is determined based on the difference in the propagation time of the signals detected at both ends. As a result, even if partial discharge does not actually occur, it is determined that partial discharge has occurred and the position is located, so the reliability of the orientation accuracy is not necessarily high, and this point is improved. There was a need to do so.

The present invention has been made in order to solve such a point, and prevents the noise signal from being erroneously discriminated as a partial discharge signal and improves the discrimination accuracy of the partial discharge signal generated in the DUT. It is an object of the present invention to provide a method of discriminating a partial discharge signal capable of performing the above.

Further, according to the present invention, the discrimination accuracy of the partial discharge signal can be enhanced by reducing the possibility of erroneously discriminating the noise signal from the partial discharge signal, and thereby the reliability of the position location result can be enhanced. An object of the present invention is to provide a method of locating a partial discharge occurrence position.

[0007]

According to the method for discriminating a partial discharge signal of the present invention, the waveform of the output signal from the partial discharge signal detecting means provided in the object to be measured is supplied to the first neural network to output the waveform. Whether or not the signal is a partial discharge signal is determined, and when the output signal is determined to be a partial discharge signal, the output signal is converted into a frequency domain waveform, and the frequency domain waveform is applied to a second neural network. It is characterized in that it is supplied to determine whether or not the output signal is a partial discharge signal.

Further, in the method for locating the partial discharge occurrence position of the present invention, the partial discharge signal detecting means is provided at both ends of the cable to be measured, and the waveform of the output signal from one partial discharge signal detecting means is determined by the first neural network. The output signal is supplied to a network to determine whether or not the output signal is a partial discharge signal, and when the output signal is determined to be a partial discharge signal, the output signal is converted into a frequency domain waveform, and the frequency domain waveform is obtained. The second
Of the output signal from the other partial discharge signal detecting means and the waveform of the output signal from the other partial discharge signal detecting means to the third neural network. Is determined to be a partial discharge signal, and when the output signal is determined to be a partial discharge signal, the output signal is converted into a frequency domain waveform, and the frequency domain waveform is supplied to the fourth neural network. Whether the output signal is a partial discharge signal or not, and when the output signals of both the partial discharge signal detecting means are not noise signals, the partial discharge is generated based on the propagation time difference between the both output signals. It is characterized by locating the position.

[0009]

In the method of discriminating the partial discharge signal of the present invention, when the output signal waveform detected from the object to be measured is supplied to the first neural network, it is possible to discriminate whether the output signal is the partial discharge signal in this neural network. Done. On the other hand, even when the output signal is determined to be the partial discharge signal, the noise signal may be erroneously determined to be the partial discharge signal. Therefore, the output signal is converted into a waveform in the frequency domain, but the converted waveform is significantly different between the noise signal and the partial discharge signal. Therefore, when the waveform in the frequency domain is supplied to the second neural network, the output signal determined as the partial discharge signal can be again determined by the second neural network as to whether it is the partial discharge signal or not. Further, in the method for locating the partial discharge occurrence position of the present invention, the waveforms of the two output signals detected from both ends of the cable to be measured are supplied to the first and third neural networks in the preceding stage, respectively, and first, in the preceding stage. It is determined whether or not the two output signals are partial discharge signals. Next, in order to prevent the noise signal from being erroneously distinguished from the partial discharge signal,
The two output signals, which are discriminated as partial discharge signals in the former stage, are converted into waveforms in the frequency domain, these converted waveforms are supplied to the second and fourth neural networks in the latter stage, and the two output signals are again partially discharged. Whether or not it is a signal is determined. Therefore, it is possible to reliably remove the erroneously determined noise signal and locate the position where the partial discharge occurs.

[0010]

Embodiments of the present invention will be described below with reference to the drawings. FIG. 1 is a block diagram of a partial discharge position locator for implementing the method of the present invention. In the figure, the cable to be measured 1 is connected to the cable heads 2 at both ends thereof in series via lead wires 3 with coupling capacitors 4A and 4B and detection impedances 5A and 5B. The partial discharge signal generated inside the cable to be measured 1 is taken out from the detection impedances 5A and 5B.

A sine wave noise removing circuit 6A is connected to the detection impedance 5A, and its output is wired so as to be input to the data storage section 10 via an amplifier 7A.
The output of the detection impedance 5B is connected to the sine wave noise removing circuit 6B, and the output thereof is connected to the data storage unit 10 via the amplifier 7B. The data storage unit 10 is provided with a digital memory for storing the two types of output signals output from both ends of the cable under test 1. That is, these two kinds of output signals are respectively stored in the area for channel 1 and the area for channel 2 of the digital memory.

FIG. 3 shows a concrete block diagram of such a data storage unit 10. As shown in the figure, the output signal output from one end of the cable under test 1 is the input signal SA of the data storage unit 10, and the output signal output from the other end is the input signal of the data storage unit 10. SB. Here, these signals SA and SB are input to the sampling circuits 11A and 11B, and the A / D conversion units 12A and 1B are input.
After 2B, it is stored in the storage units 13A and 13B (steps S1 and S2 in FIG. 6).

The sampling circuits 11A and 11B are circuits for sampling the input signals SA and SB at appropriate intervals and loading them into the data storage section 10. Also, A / D
The conversion units 12A and 12B include sampling circuits 11A and 1B.
It is a circuit for converting the output of 1B into a digital signal. The storage units 13A (CH1) and 13B (CH2) are memories for converting the waveforms of the input signals SA and SB into digital signals and storing them.

Returning to FIG. 1 again, the digital signals of the storage units 13A and 13B of the data storage unit 10 are the data extraction unit 20.
Is supplied to. FIG. 2A is a block diagram of the data extraction unit 20. The data extraction unit 20 includes a data comparison unit 21 and
Data acquisition units 22A, 22 connected to the output side
It consists of B and. The data comparison unit 21 includes storage units 13A and 13A.
Each digital signal (data) is read from B, the change in the signal waveform is calculated by calculation, and the rising (or falling) point of the signal is detected. That is, for example, as shown in FIG. 2B, a predetermined trigger level T _{r is} preset.
Is set, and when the output signal (digital signal) on the channel 1 side becomes larger than this trigger level T _{r, an} approximate straight line connecting the trigger level T _r point on the output signal and two points before and after is obtained, and this approximate straight line The intersection with the signal level “0” is calculated as the rising point T ₁ of the output signal,
Detect. Further, as shown in FIG. 2C, the same calculation is performed on the output signal on the channel 2 side, and the rising point T ₂ is detected.

When the rising points T ₁ and T ₂ are detected in this way, the data comparison unit 21 receives the data acquisition units 22A and 22A.
Output a capture signal to B. When the data acquisition section 22A receives the acquisition signal, the data acquisition section 22A receives the signal 1 before the rising point T ₁ as a reference.
Data of 50 points, 0 points and 40 points behind, are stored in the storage unit 13A.
Further capture (step S3 in FIG. 6). In addition, when the other data capturing section 22B receives the capture signal, the rising point T
Data of 50 points in total, that is, 10 points before and 40 points after ₂ are taken in from the storage unit 13B (step S4 in FIG. 6). Therefore, since the rising positions of the output signals output from the cable to be measured 1 are made to coincide with each other and only the data having a constant time width is extracted, only the most characteristic data of the output signal waveform is described later in the neural network 31A. , 31B.

The waveform data captured by both data capturing units 22A and 22B is supplied to the pre-stage determination unit 30. The pre-stage determination unit 30 includes a pre-stage neural network 31.
A and 31B are provided. The neural network 31A is a part that analyzes the extracted waveform data output from one end of the cable under test 1 and captured by the data capturing part 22A of the data extracting part 20. The neural network 31B is a part for analyzing the extracted waveform data output from the other end of the cable under test 1 and captured by the data capturing part 22B of the data extracting part 20.

FIG. 4 is a structural explanatory view of the above-mentioned pre-stage neural networks 31A and 31B. The neural network itself uses software that is already commercially available. This neural network
A perceptron type network including three layers, that is, an input layer 33, an intermediate layer 34, and an output layer 35 is configured.
In the case of this embodiment, for example, there are 50 input layers 33, 10 intermediate layers 34, and 2 output layers 35. Then, 19 kinds of partial discharge signal waveforms and 23 kinds of noise waveforms were given in advance, and these 42 pieces of data were each learned 1000 times by the back propagation learning rule.

As a result, the neural networks 31A and 31B at the preceding stage analyze the waveform by the intermediate layer 34 when the input waveform is given to the input layer 33, and empirically have a characteristic peculiar to the partial discharge signal waveform. The signal waveform is recognized using the above learning result, and it is determined whether or not the input waveform is the partial discharge signal (steps S5 and S6 in FIG. 6). When the input signal waveform is the partial discharge signal, the output layer 3
One output of 5 is 0.8 or more and the other output is 0.2 or less. In the case of a noise waveform, one output is 0.2 and vice versa.
Thereafter, the other output operates so as to be 0.8 or more.
The other output results were not evaluated.

Neural networks 31A, 3 in the previous stage
The position locator 32 is connected to the output of 1B. The position locator 32 performs a position locating operation described later,
Furthermore, the above-mentioned neural networks 31A, 3 in the previous stage
When receiving the output of 1B and both outputs show the partial discharge signal,
That is, when one output of each output layer 35 of both neural networks 31A and 31B is 0.8 or more and the other output is 0.2 or less, the detection signal is output to the Fourier transform unit 42 of the arithmetic processing unit 40.

As shown in FIG. 1, the arithmetic processing section 40 is provided with neural networks 41A and 41B in the subsequent stage connected to the output of the Fourier transform section 42, and a decision detecting section 43. The Fourier transform unit 42 determines that both output signals are partial discharge signals by the preceding neural networks 31A and 31B (step S7 in FIG. 6), and when the detection signal is sent from the determination detection unit 43, the data acquisition is performed. Including part 2
Each extracted waveform data is read from 2A and 22B, and Fourier transform calculation processing is executed (steps S8 and S in FIG. 6).
9). As a result, the extracted waveform data is converted into waveform data represented by frequency-output level as shown in FIG. The waveform of the waveform data represented by the frequency-output level differs greatly between the partial discharge signal and the noise signal.

On the other hand, the neural network 41 in the latter stage
A and 41B are 20 input layers 4 as shown in FIG.
Four or five intermediate layers 45 and two output layers 46 are provided, and the conversion data obtained by converting a large number of partial discharge signals and noise signals into frequency-output level data in advance is 1000 according to the back propagation learning rule. Trained once. As a result, the neural networks 41A and 41
B is the case where the converted waveform data is given to the input layer 44,
The intermediate layer 45 analyzes the waveform, and empirically recognizes the waveform having the characteristic peculiar to the noise signal by using the previous learning result, and determines whether or not the waveform of the converted data is the noise signal (see FIG. 6). Steps S10 and S11). When the converted data waveform is a noise signal, one output of the output layer 46 is 0.8 or more and the other output is 0.2 or less, and one output of the partial discharge signal waveform is 0.2 or less, It operates so that the other output becomes 0.8 or more. The other output results were not evaluated.

The judgment detecting section 43 outputs the output results of the neural networks 41A and 41B in the subsequent stage to the position locating section 32. When the position locator 32 receives the determination result that at least one is a noise signal from the determination detector 43 (step S12 in FIG. 6), the both output waveforms are determined to be partial discharge signals by the preceding neural networks 31A and 31B. Even if a signal is received, position location calculation processing is not executed. On the other hand, when both of them receive the determination result of the partial discharge signal from the determination detection unit 43 and when the determination result of the evaluation cannot be performed as described later, the calculation of the partial discharge occurrence position is executed (step S13 of FIG. 6). ). Therefore, even if the noise signal is erroneously determined to be a partial discharge signal by the neural networks 31A and 31B in the former stage, the neural network 41 in the latter stage
Since the noise signal erroneously discriminated as the partial discharge signal by A and 41B can be removed, the position localization due to the erroneous discrimination of the noise signal as the partial discharge signal is not executed.

Next, the method for locating the generation position of the partial discharge signal according to the method of the present invention will be specifically described. That is, the length is 3
As a 4.6 m CV cable for measurement, a partial discharge is generated at a position of 14.4 m from the power supply, the partial discharge signal is discriminated, and an experiment of whether or not the position of the partial discharge generation position is accurate is conducted. I did. In this experiment, first,
The cable to be measured was disconnected from the device, and the partial discharge signal previously obtained from this cable was input to the device. In this case, 8 partial discharge signals and 60 noise signals were sequentially input, and the output results of the neural networks 31A and 31B at the preceding stage were examined. As a result, the neural network 31A on the side of the channel 1 discriminates all the eight input partial discharge signals as partial discharge signals, and of the 60 noise signals input, 13.4% of them is the partial discharge signal, 5
It was determined that 3.3% was a noise signal and 33.3% was unevaluable.

On the other hand, in the neural network 31B on the channel 2 side, of the eight input partial discharge signals, 7 were judged to be partial discharge signals, and the remaining 1 was judged to be unevaluable. Also, this neural network 31B
In the 60 noise signals, it was determined that 16.7% was a partial discharge signal, 70% was a noise signal, and 13.3% was unevaluable.

As described above, when the rising positions of the signal waveforms supplied to the preceding neural networks 31A and 31B are made to coincide with each other and only data having a constant time width is supplied, whether the output signal is a partial discharge signal or not. It is possible to discriminate whether or not it is highly accurate, and it is extremely rare to erroneously discriminate a noise signal from a partial discharge signal. This is because the data extraction unit 20 detects the rising edge of the signal waveform, matches the rising positions of the respective signal waveforms, and only the data of a certain time width is used in the preceding neural networks 31A, 31.
Since it has been supplied to B, the neural network 31A, 3
This is because the discrimination accuracy of 1B is improved.

Next, the preceding neural network 31
The signal waveforms of the noise signal erroneously discriminated as the partial discharge signal in A and the eight partial discharge signals accurately discriminated as the partial discharge signal in the neural network 31A are Fourier-transformed, and then the waveforms thereof are processed by the neural network 41.
I typed in A. Further, after the Fourier transform is performed on the signal waveforms of the noise signal erroneously discriminated as the partial discharge signal by the neural network 31B and the seven partial discharge signals accurately discriminated by the neural network 31B as the partial discharge signal, the waveform Was input to the neural network 41B. Then, the output results of the neural networks 41A and 41B in the latter stage of the arithmetic processing unit 40 were examined. As a result, the neural network 41A on the channel 1 side determined that the eight partial discharge signals could not be evaluated, but determined that the noise signal was a 100% noise signal. Similarly, the neural network 41B on the channel 2 side determined that seven partial discharge signals could not be evaluated, but the noise signal was determined to be a 100% noise signal. Thus, the noise signal erroneously discriminated as the partial discharge signal could be completely removed by the neural networks 41A and 41B in the subsequent stage.

Then, from the eight signals judged to be the partial discharge signal on the channel 1 side and the seven signals judged to be the partial discharge signal on the channel 2 side, seven sets of signals under the AND condition are extracted, and each of them is extracted. The location of the partial discharge generation location was determined based on the difference in signal propagation time. As a result, in any case, the position about 15 m from the power source was determined to be the position where the partial discharge occurred. In the above embodiment, 10 points before and 40 points after the rise (or fall) of the signal waveform were extracted, but the extraction points are not limited to this, and the point is to extract the characteristic points of the waveform. Alternatively, all rising waveforms may be extracted.

The present invention is not limited to the above embodiments. In the above embodiment, an example is shown in which the partial discharge of the actually laid cable is determined and the position thereof is determined, but the partial discharge signal is taken out from both ends of the cable wound around the drum or the like to measure. The same can be done in such cases and in the case of determining partial discharge of a high voltage device such as a transformer. In addition, the measurement circuit and wiring may be freely selected. Furthermore, the neural network may have any configuration as long as it can perform a certain learning based on the waveforms of the actual partial discharge signal and the noise signal and can distinguish the external noise from the partial discharge signal.

[0029]

As described above, according to the method of discriminating the partial discharge signal of the present invention, the output signal waveform detected from the object to be measured is supplied to the first neural network to determine whether the output signal is the partial discharge signal. Since it is determined whether or not the output signal is determined to be a partial discharge signal and is converted into a waveform in the frequency domain and supplied to the second neural network, a noise signal is erroneously determined as a partial discharge signal. Can be reliably prevented, and the determination accuracy of the partial discharge signal can be significantly improved. Further, according to the method for locating the partial discharge generation position of the present invention, the waveforms of the two output signals detected from both ends of the cable to be measured are supplied to the first and third neural networks in the preceding stages, respectively, and the two output signals are output. Is determined to be a partial discharge signal, and the two output signals determined to be partial discharge signals are further converted into waveforms in the frequency domain and supplied to the second and fourth neural networks in the subsequent stage. It is possible to provide a highly reliable method of locating a partial discharge occurrence position without misidentifying a noise signal as a partial discharge signal to perform position locating.

[Brief description of drawings]

FIG. 1 is a block diagram of a partial discharge occurrence position locating device for carrying out the method of the present invention.

FIG. 2A is a block diagram of a data extraction unit according to the present invention, and FIGS. 2B and 2C are waveform diagrams illustrating a data extraction method.

FIG. 3 is a block diagram of a data storage unit according to the present invention.

FIG. 4 is an explanatory diagram of the structure of the neural network in the first stage according to the present invention.

FIG. 5 is an explanatory diagram of a neural network structure of a latter stage according to the present invention.

FIG. 6 is a flow chart for explaining the method of the present invention.

FIG. 7 is a diagram showing a frequency-output level conversion waveform according to the present invention.

[Explanation of symbols]

 DESCRIPTION OF SYMBOLS 1 Cable to be measured 10 Data storage section 20 Data extraction section 30 Pre-stage determination section 31A, 31B Pre-stage neural network 32 Position locating section 40 Arithmetic processing section 42 Fourier transform section 41A, 41B Post-stage neural network

Claims (2)

[Claims] 1. A waveform of an output signal from a partial discharge signal detecting means provided on an object to be measured is supplied to a first neural network to determine whether or not the output signal is a partial discharge signal. When the signal is determined to be a partial discharge signal, the output signal is converted into a frequency domain waveform, and the frequency domain waveform is supplied to a second neural network to determine whether the output signal is a partial discharge signal. A method of discriminating a partial discharge signal, characterized in that: 2. The partial discharge signal detecting means is provided at both ends of the cable to be measured, and the waveform of the output signal from one partial discharge signal detecting means is supplied to the first neural network so that the output signal is a partial discharge signal. Whether or not the output signal is a partial discharge signal, the output signal is converted into a frequency domain waveform, and the frequency domain waveform is supplied to a second neural network to output the output. It is determined whether the signal is a partial discharge signal, and the waveform of the output signal from the other partial discharge signal detecting means is supplied to the third neural network to determine whether the output signal is a partial discharge signal. When the output signal is determined to be a partial discharge signal, the output signal is converted to a frequency domain waveform, and the frequency domain waveform is converted to a fourth neural network. To determine whether or not the output signal is a partial discharge signal, when the output signal of both the partial discharge signal detection means is not a noise signal, based on the propagation time difference between both output signals A method for locating a partial discharge generation position, which is characterized by locating a discharge generation position.