KR100729107B1 - Methods of Input Vector formation for Auto-identification of partial discharge source using neural networks - Google Patents

Abstract

Since the present invention does not use the applied voltage phase information of a high voltage device by using a neural network using a phase-integrated phase, the partial discharge is measured without receiving a phase signal from a PT or a voltage divider when measuring the partial discharge in the field. In addition to reducing the time required to prepare for the process, the measurement cost can be reduced, as well as reliable inference results for the discharge cause. Since the input vector of the neural network is extracted from the discharge type that does not include the magnitude information of the partial discharge signal, the same neural network can be used in other partial discharge measuring devices having different characteristics of amplifiers and signal attenuators, and power of multiple phases can exist simultaneously. The phase-integration and phase-correlation in the device can be used to identify the power phase of the site where the partial discharge occurs, so that the location of the partial discharge can be easily tracked. The method for generating an input vector,

Multi-layered perceptron structure or self organization map that automatically infers the cause of partial discharge signal generated in high voltage power equipment such as GIS, transformer, power cable, rotating equipment In the input vector used for the neural network circuit as described above, using the discharge signal measured from the partial discharge measuring device, the discharge voltage which follows the applied voltage phase Φ n-1 of the preceding discharge signal as the x-axis in the first continuous discharge signal A point is plotted on the (Φn-1, Φn) coordinate of the two-dimensional graph with the applied voltage phase Φn of the y-axis, and (Φn, Φn + corresponding to the phase of the applied voltage for two successive discharge signals). 1) generating a Φn: Φn-1: N graph by dividing the dots, and dividing the above Φn: Φn-1: N graph into two zones consisting of an upper left and a lower right with respect to a diagonal line. Afterwards, the region is moved to the Φn: Φn-1: N graph by moving the region above the upper left region, extracting the phase correlation to be used as the input vector of the neural network, and using it as the input vector of the neural network. Extracting the phase unmatched, and inputting the phase unmatched as an input vector of the neural network circuit.

Perceptron, Neural Network, Discharge Signal, Phase-correlation, Phase-integration

Description

Methods of input vector formation for auto-identification of partial discharge source using neural networks

1 is a configuration diagram of a general microwave partial discharge measurement device.

2 is a block diagram of a partial discharge signal input from a partial discharge sensor of a general microwave partial discharge measurement device.

FIG. 3 is an exemplary diagram of a partial discharge signal expressed by a Φ n: Φ n-1: N visualization method for explaining an input vector generation method of a neural network for partial discharge cause automatic reasoning according to an embodiment of the present invention.

4 is an exemplary diagram of a partial discharge signal expressed by a modified Φn: Φn-1: N visualization method for explaining an input vector generation method of a neural network for partial discharge cause automatic inference according to an embodiment of the present invention.

5 is an exemplary diagram of a general multilayer neural network.

6 is an exemplary view illustrating a power phase determination method of a discharge generation position using phase correlation according to an embodiment of the present invention.

Explanation of symbols on the main parts of the drawings

100: high voltage power device 110: power

120: partial discharge sensor 130: partial discharge measuring instrument

140: PT / potentiometer for power phase measurement

200: voltage applied to the high voltage power equipment

210: partial discharge signal 220: generation time of the partial discharge signal

230: magnitude of partial discharge signal 240: phase of applied voltage when partial discharge occurs

320: occurrence frequency of the partial discharge signal

410: phase correlation 420: phase correlation

500: input vector 510: primary layer

520: N-th layer 530: output layer

540: output vector 550: synapse

The present invention relates to the electric power field. More specifically, since the neural network circuit using phase integrator does not use the applied voltage phase information of a high voltage device, a phase signal from a PT or a voltage divider is measured during partial discharge measurement in the field. Partial discharges are measured without input, reducing preparation time for measurement, reducing measurement costs, and obtaining reliable inferences about the cause of discharge. Since the input vector of the neural network is extracted from the discharge type that does not include the magnitude information of the partial discharge signal, the same neural network can be used in other partial discharge measuring devices having different characteristics of the amplifier and the signal attenuator. It is possible to know the power phase of the part where partial discharge occurs by using phase uncorrelation and phase correlation in power equipment, so it is easy to track the location of partial discharge, which is advantageous for the follow-up measures for abnormal power equipment. It relates to an input vector generation method of.

In high voltage power equipment such as gas insulated switchgear (GIS), gas insulated transformer, inflow transformer, rotating equipment, gas insulated transmission line, power cable, partial discharge occurs as a precursor of failure. In this way, the discharge measuring device measures the microwaves generated together with the partial discharge when the partial discharge occurs in the high voltage power device, and analyzes the signal types and patterns of the microwaves thus measured to diagnose the partial discharge and the deterioration state of the power device. This is a device that can measure or recognize the failure sign of the power equipment in advance.

A general microwave partial discharge measuring apparatus will be described with reference to FIGS. 1 and 2 as follows.

When there is an internal abnormality in the high voltage power device 100 such as a gas insulated switchgear (GIS), a transformer, a motor, or a power cable, when the power supply 110 is supplied to the high voltage power device 100, As a symptom, partial discharge occurs.

In this case, the partial discharge sensor 120 mounted on the high voltage power device 100 detects the partial discharge signal and inputs the microwave discharge to the microwave partial discharge measuring instrument 130 to analyze the partial discharge signal.

Since the phase information of the power source 110 is essential in the existing analysis method, the applied voltage phase information is provided to the partial discharge meter 130 using the power transformer PT and the voltage divider 140.

The microwave partial discharge measuring unit 130 receives the partial discharge signal generation time 220, the partial discharge signal magnitude 230, and the partial discharge signal 210 from the partial discharge signal 210 as shown in FIG. 2 input from the partial discharge sensor 120. When the discharge occurs, basic information related to discharge, such as the phase 240 of the applied voltage 200, may be acquired.

The signal acquisition circuit in the microwave partial discharge measuring instrument 130 is provided with various types of amplifiers, signal attenuators and detection devices, and neural network circuits for signal classification, and the partial discharge signal input from the partial discharge sensor 120. (210) can be analyzed.

By analyzing the shape of the partial discharge signal 210 as described above, it is possible to diagnose whether or not the partial discharge and the deterioration state of the high voltage power device 100, and by using this to measure the failure signs of the high voltage power device 100 in advance Or perceivable.

The neural network circuit of the microwave partial discharge measuring instrument 130 uses an unprocessed measured partial discharge signal as an input vector, or in a phase resolved pulse sequence (PRPS) or a phase resolved partial discharge (PRPD) converted from the partial discharge signal. Use various extracted variables. Examples of these variables are the number of phase resolved discharge pulses or the magnitude of the discharge signal, and various statistical variables such as skew or kurtosis. Such an input vector directly or indirectly uses the applied voltage phase information of the discharge signal to the power device and the magnitude of the discharge signal.

However, when the discharge signal is measured using the mobile microwave partial discharge meter with respect to the high voltage power device 100, it may be difficult to simply receive phase information applied to the high voltage power device 100. That is, when the phase information of the applied voltage 200 is not known or the power phase of the position where the partial discharge occurs is not known, the microwave partial discharge measuring instrument 130 generally uses the phase 240 of the applied voltage 200 of the high voltage power device 100. It is assumed that the phase of the power supply of the partial discharge measuring instrument 130 is a phase of the voltage applied to the high voltage power device 100, and the existing neural network circuit can infer an erroneous discharge cause. In this case, the neural network circuit according to the conventional technology has a problem in that it is difficult for the user to take appropriate measures because the neural network circuit delivers misleading inferred discharge causes to the user.

In the case of a partial discharge measurement system installed on-line, a high-voltage power device 100 in which several phases exist at the same time, such as a three-phase integrated GIS, a transformer, or a three-phase electric motor, is not accurate. There is a problem that it is difficult to derive inference results.

In addition, when the magnitude of the discharge signal is used as the input vector of the neural network, even when the same discharge source is obtained, the discharge signal of various sizes can be obtained according to the distance between the sensor and the discharge source and the characteristics of the signal acquisition device. There is a need to use a discharge signal of a magnitude. In particular, when the gains of various amplifiers existing in the discharge signal acquisition apparatus are changed, the magnitude of the measured signal may be changed for the same discharge signal. Therefore, the neural network must be retrained in consideration of this.

An object of the present invention is to solve the conventional problems as described above, to automatically infer the cause of the partial discharge signal generated in high-voltage power equipment such as gas insulated switchgear (GIS), transformer, motor, power cable The main purpose of the present invention is to provide an input vector generation method of neural network circuits for automatic reasoning of partial discharges, which can generate various types of input vectors for neural network circuits and infer images using partial discharges.

Another object of the present invention is to measure the partial discharge generated in the precursor of failure in high voltage power equipment such as gas insulated switchgear (GIS), transformer, motor, power cable, to prevent the failure of the power equipment in advance The present invention provides a method of generating an input vector of a neural network circuit for partial discharge cause automatic inference, which indicates a cause of automatic discharge from the partial discharge signal.

Another object of the present invention is to use the input vector irrelevant to the phase information, it is possible to infer the cause of the partial discharge even in the situation where the phase information of the power applied to the high-voltage power equipment can not be obtained, as well as to the place where the discharge occurred The present invention provides a method for generating an input vector of a neural network circuit for partial discharge cause automatic inference, which can infer a phase of an applied power source.

It is still another object of the present invention to generate an input vector of a neural network circuit that does not use information on the magnitude of a discharge signal, thereby analyzing signals measured by a partial discharge measuring device having different characteristics of an amplifier and a signal attenuator of a data acquisition circuit. The present invention also provides a method of generating an input vector of a neural network circuit for automatic inference of partial discharge, which can use the same neural network circuit.

DETAILED DESCRIPTION OF THE PREFERRED EMBODIMENTS Hereinafter, the present invention will be described in detail with reference to the accompanying drawings in order to describe in detail enough to enable those skilled in the art to easily carry out the present invention. . Other objects, features, and operational advantages, including the object, operation, and effect of the present invention will become more apparent from the description of the preferred embodiment.

For reference, the embodiments disclosed herein are only presented by selecting the most preferred embodiment in order to help those skilled in the art from the various possible examples, the technical spirit of the present invention is not necessarily limited or limited only by this embodiment Rather, various changes, additions, and changes are possible within the scope without departing from the spirit of the present invention, as well as other equivalent embodiments.

3 is an exemplary diagram of a partial discharge signal expressed by Φn: Φn-1: N visualization method for explaining an input vector generation method of a neural network for partial discharge cause automatic inference according to an embodiment of the present invention, and FIG. Is an exemplary diagram of a partial discharge signal expressed by a modified Φn: Φn-1: N visualization method for explaining an input vector generation method of a neural network for partial discharge cause automatic inference according to an embodiment of the present invention.

3 and 4, the configuration of the input vector generation method of the neural network circuit for partial discharge cause automatic reasoning according to an embodiment of the present invention includes a gas insulated switchgear (GIS), a transformer, a power cable, Partial discharge in input vector used in neural network such as multi-layer perceptron structure or self organization map that automatically infers the cause of partial discharge signal generated in high voltage power equipment such as rotating equipment Using the discharge signal measured from the measuring device, a two-dimensional graph in which the applied voltage phase Φ n-1 of the preceding discharge signal is the x-axis and the applied voltage phase Φ n of the subsequent discharge signal is the y-axis in the first continuous discharge signal. Φn by marking a point on the (Φn-1, Φn) coordinates and then placing a point on (Φn, Φn + 1) corresponding to the phase of the applied voltage for two consecutive discharge signals. generating an n-1: N graph and dividing the above Φn: Φn-1: N graph into two zones consisting of an upper left and a lower right with respect to a diagonal line, and then the upper right area above the upper left area. Converting to a Φn: Φn-1: N graph by shifting, extracting a phase correlation to be used as an input vector of the neural network circuit, extracting a phase non-union to be used as an input vector of the neural network circuit, And inputting a phase unintegration as an input vector of the neural network circuit.

It is further preferable that the configuration of the present invention includes the step of inputting the above-described phase correlation and phase non-integration as an input vector of the neural network circuit.

The configuration of the present invention compares the phase correlation obtained from the discharge signal and the phase correlation obtained by shifting the phases by 120 degrees and 240 degrees, respectively, to the shape of the reference phase non-association so that the phase of the power applied to the discharge signal is generated. It is desirable to include further steps to find out.

The configuration of the present invention refers to the phase correlation obtained from the discharge signal and the phase correlation obtained by integrating the product of the phase correlation with the reference phase unmatched by 120 and 240 degrees respectively. It is preferable to further comprise the step of selecting the phase-unmatched most similar to the phase-unmatched.

According to the present invention, the cross-correlation of the phase correlation obtained from the discharge signal and the phase correlation obtained by shifting the phase by 120 degrees and 240 degrees, respectively, is obtained by obtaining a cross correlation with the reference phase non-association. It is preferable to further comprise the step of identifying the phase of the power applied to the place.

The operation of the input vector generation method of the neural network for partial discharge cause automatic inference according to the embodiment of the present invention by the above configuration is as follows.

The present invention extracts an input vector from which the phase information is not included in the magnitude and applied voltage of the discharge signal from the partial discharge signal, and then, even in a situation where there is no phase information of voltage applied to the high voltage power device through the neural network circuit using the same. It is possible to automatically infer the cause of the two input vectors, and since the information on the magnitude of the discharge signal is not used, the neural network training is not necessary considering the signal size. When the power of the phase is applied, an input vector including the inferred cause of partial discharge and phase information may be generated to infer the phase where the partial discharge occurs.

The present invention is to generate an input vector of a neural network circuit for inferring the exact cause of the discharge, the discharge time 220 of the information about the basic discharge signal as shown in Figure 2 from very small value to very large value Since it may be difficult to standardize, since the discharge signal size 230 has many points to consider when training the neural network, an input vector of the neural network is generated using only the phase 240 of the applied voltage.

In order to make an input vector, first, a discharge type according to "Φ n: Φ n-1: N visualization method" is created as shown in FIG. 3 by using the applied voltage phase 240 Φ at the time of discharge.

The above "Φn: Φn-1: N visualization method" represents the applied voltage phase relationship of two consecutive discharge signals, and when there is a continuous discharge signal 210 as shown in FIG. As shown in the two-dimensional graph of the first continuous discharge signal, the applied voltage phase Φ n-1 of the preceding discharge signal is the x axis, and the applied voltage phase Φ n of the subsequent discharge signal is the y axis. Φn) the point 300 is indicated. A dot 310 is then drawn at (Φn, Φn + 1) corresponding to the phase of the applied voltage for two successive discharge signals.

In this way, after the points are displayed at the coordinates corresponding to the applied voltage phases for the partial discharge occurring for a predetermined time T, the number of points displayed for each coordinate is shown by the color coordinate 320. The discharge type can be made by the "Φn: Φn-1: N visualization method" as shown in FIG. At this time, the number of points can be expressed by changing the density of the displayed points for a certain time.

The discharge type by the “Φn: Φn-1: N visualization method” is greatly different in shape for each discharge source, so that it is relatively easy to generate distinguishable input vectors for each discharge source. It is standardized to a size within 360 degrees.

In order to obtain an input vector without phase information, it is necessary to modify FIG. 3, which is divided into two zones consisting of an upper left surface 340 and a lower right surface 330 based on a diagonal line, and then a region of the lower right surface 330. By moving up the area of the upper left surface 340, it can be transformed into the graph 400 as shown in FIG. 4 and this visualization method is referred to as "strain Φn: Φn-1: N visualization method".

In the "strain Φ n: Φ n-1: N visualization method" graph shown in FIG. 4, the sum of the discharge counts of the Y 'axis with respect to the X' axis can be obtained. Since the summation has phase information, it is called a phase independent sum 410 in the present invention.

In FIG. 4, when the Y 'axis is added and the sum is performed in the X' axis direction, each sum point is a sum of discharge times corresponding to different phases rather than a specific phase on the Φ n axis and the Φ n-1 axis. A graph of the missing discharge counts can be obtained. In the present invention, this is called phase independent independent (420).

The phase relation 410 and the phase non-integration 420 are always positive as variables related to the number of discharges.

FIG. 5 is an example of a general multilayer neural network, and as shown in FIG. 5, the configuration of a general multilayer neural network includes neurons for determination such as an input vector 500 and a perceptron extracted from a measured discharge signal. A primary layer 510 and an N-order layer 520 composed of (operators), an output layer 530 that calculates an output vector 540 of a neural network circuit indicating an abnormal cause of a power device, and a synapse 550 connecting each layer )

When the phase unmatched 420 is used as the input vector of the multilayer neural network as described above, since the information on the magnitude of the discharge signal and the phase of the applied voltage is not used, the applied voltage phase information is not measured by the partial discharge signal measuring device. You can reason about this reasonably precisely.

In addition, when the phase unintegration 420 is used as the input vector of the multi-layer neural network circuit, the amplitude information is not used. Therefore, the signal amplification characteristics of the neural network circuit trained with the discharge signal measured by the specific signal acquisition apparatus are not retrained. It can also be used for other partial discharge signal acquisition devices.

Also, in the case of the neural network using the phase correlation 410 and the phase unintegration 420 as the input vector, the neural network uses the phase information of the applied voltage, which is incorrect even in the situation where the phase information cannot be obtained. Inference results can be seen, but when the phase information can be obtained, since the discharge information is used more than the neural network using only the phase unmatched 420, the recognition rate is high.

In addition, by using the phase correlation 410 and the phase non-integration 420, it is possible to infer the phase of the voltage applied to the discharge generation point without additional information on the phase in which the discharge occurs in the high voltage power device to which the power of several phases is applied. have.

For example, in the three-phase collective GIS in which A, B, and C phases exist simultaneously, partial discharge occurs on B and a discharge signal is measured by using a partial discharge measurement device synchronized with A. .

First, the phase correlation 600 and the phase non-integration are extracted from the discharge signal, and then the cause of discharge is inferred through a neural network using the phase non-integration of the discharge signal.

Next, the phase correlation 600 obtained from the discharge signal of the reference phase unmatched 630 used in the neural network training for the inferred discharge cause, and the phase shifted phase by 120 degrees and 240 degrees, respectively. Compare to the shape of correlation sum 610, 620.

In this case, the shape most similar to the reference phase unintegration 630 is phase correlation 610 shifted by 120 degrees. Therefore, partial discharge occurs in phase B where the 120-degree phase is lower than the phase A in which the partial discharge measuring device is synchronized. It can be seen. In this way, the applied voltage phase of the partial discharge occurrence position can be known.

In this case, the similarity to the phase correlation 600 of the measurement signal and the reference phase unmatching 630 of phase correlations 610 and 620 shifted by 120 degrees and 240 degrees, respectively, can be numerically indicated. There are several ways, the simplest of which is to multiply two phase correlations. According to this method, the more similar the shape of the two phase correlations, the larger the value of the product integration (the area of the curve of the phase correlation products).

In the above example, a phase shifted phase of 120 degrees with the largest area (640, 650, 660) obtained by integrating the product of the phase unmatched 600 and the phase shifted phase correlation (610, 620) with the reference phase unmatched (630) The correlation sum 650 is the phase unmatched most similar to the reference phase unmatched 630.

In addition, the phase can be inferred by a method such as cross correlation.

As described in the above embodiment, since the present invention does not use the applied voltage phase information of a high voltage device by using a neural network using a phase-integrated phase, a phase signal is input from a PT or a voltage divider when measuring partial discharge in the field. Partial discharges are measured without receiving them, which reduces the time to prepare for measurement, reduces the cost of measurement, and provides reliable inferences about the cause of discharge. Since the input vector of the neural network is extracted from the discharge type that does not include the magnitude information of the partial discharge signal, the same neural network can be used in other partial discharge measuring devices having different characteristics of amplifiers and signal attenuators, and power of multiple phases can exist simultaneously. The phase-integration and phase-correlation in the device can be used to identify the power phase of the site where the partial discharge occurs, so that the location of the partial discharge can be easily traced, which is advantageous for the follow-up action on the abnormality of the power device.

Claims (5)

Multi-layered perceptron structure or self organization map that automatically infers the cause of partial discharge signal generated in high voltage power equipment such as GIS, transformer, power cable, rotating equipment In the input vector used in the neural network, such as By using the discharge signal measured from the partial discharge measuring device, 2 is the x-axis of the applied voltage phase Φ n-1 of the preceding discharge signal in the first continuous discharge signal and the y-axis of the applied voltage phase Φ n of the subsequent discharge signal. Φn: Φn-1 by marking a point on the (Φn-1, Φn) coordinates of the dimensional graph and then placing a point on (Φn, Φn + 1) corresponding to the phase of the applied voltage for two consecutive discharge signals. Generating an N graph; The Φn: Φn-1: N graph is divided into two zones consisting of the upper left side and the lower right side of the diagonal, and then the upper right side is deformed by moving the upper left side. Converting to; Extracting a phase correlation to be used as an input vector of the neural network; Extracting a phase unintegration to be used as an input vector of the neural network; And And inputting said phase unintegration as an input vector of said neural network circuit. The method of claim 1, And inputting the phase correlation and phase unintegration as an input vector of the neural network circuit. The method according to claim 1 or 2, The phase correlation obtained from the discharge signal and the phase correlation obtained by shifting the phase by 120 degrees and 240 degrees, respectively, are compared with the shape of the reference phase unmatched to further determine the phase of the power applied to the discharge signal. An input vector generation method of a neural network for partial discharge cause automatic inference, characterized in that it comprises a. The method according to claim 1 or 2, The phase correlation that is obtained from the discharge signal and the phase correlation that shifts the phase by 120 degrees and 240 degrees, respectively, is the most similar to the reference phase mismatch. A method for generating an input vector of a neural network circuit for partial discharge cause automatic inference, characterized in that the method further comprises the step of selecting by phase unintegration. The method according to claim 1 or 2, The phase correlation obtained from the discharge signal and the phase correlation obtained by shifting the phase by 120 degrees and 240 degrees, respectively, are obtained. Method for generating an input vector of the neural network circuit for partial discharge cause automatic reasoning, characterized in that further comprising the step of finding.

Images