A method for monitoring partial discharge of a high-voltage switch cabinet

Abstract

The application relates to a partial discharge monitoring method for a high-voltage switch cabinet, and particularly relates to a partial discharge monitoring method for a high-voltage switch cabinet based on combined detection of transient earth voltage (TEV) and non-contact ultrasonic waves (AA). The application utilizes a double extreme value average threshold method for denoising pretreatment, and realizes high resonance and low resonance decomposition through a resonance sparse algorithm to form a recombined signal of denoised oscillation, periodic signals and impact signals rich in characteristic information. The recombined signal is input into a BP neural network data processing terminal together with an overcomplete atom library constructed based on an overcomplete redundant function to perform comparative analysis, interference caused by errors or noises can be avoided, and effective partial discharge pulse waveforms can be extracted. By adjusting response speed, quality factor and center frequency parameters, the partial discharge online monitoring and data analysis under different environments can be satisfied.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of on-line monitoring of partial discharge of high-voltage switch cabinet, and particularly relates to a partial discharge monitoring method for high-voltage switch cabinet based on transient earth voltage (TEV) and non-contact ultrasonic wave (AA) combined detection. BACKGROUND

[0002] The partial discharge is mainly the discharge of internal insulation of transformers, switches, power cables and other high-voltage power equipment under the action of high voltage. It is one of the main reasons for the failure of high-voltage power equipment.

[0003] The partial discharge detection is a common method for evaluating the insulation performance of high-voltage power equipment. The extraction of partial discharge pulse signal is a key step in the signal preprocessing stage of partial discharge measurement, and is the basis for in-depth analysis of the partial discharge signal. At present, the main method for extracting the partial discharge pulse is the wavelet threshold denoising method, which can achieve certain effect in processing the partial discharge pulse signal. However, there are still some problems.

[0004] The signal collected by the on-line detection method of partial discharge often contains a large amount of noise, which will make the edge of a single partial discharge pulse signal not obvious. The existing technology has the problems of large amount of calculation, obvious signal distortion, unclear pulse edge and the like, which makes the analysis and identification of the characteristics of the partial discharge pulse signal in the later stage difficult. The partial discharge monitoring method based on the double extreme value average threshold method and the resonance sparse decomposition algorithm has remarkable effect in extracting the partial discharge signal by using the data processing terminal of the BP neural network, and has the ability of anti-interference and anti-noise. SUMMARY

[0005] The present application aims to provide a partial discharge monitoring method for high-voltage switch cabinet based on transient earth voltage (TEV) and non-contact ultrasonic wave (AA) combined detection.

[0006] In order to achieve the above-mentioned purpose, the present application provides a partial discharge monitoring method based on the double extreme value average threshold method and the resonance sparse decomposition algorithm, which comprises the following steps:

[0007] Step 1: placing a sensor collecting end of a partial discharge monitoring device in the interior of a switch cabinet, connecting the collecting end to a data processing terminal, and collecting the discharge signal x(t) of the switch cabinet through the sensor collecting end of the monitoring device;

[0008] Step 2: processing the sampling signal x(t) according to the double extreme value average threshold method:

[0009] Step 3: adopting a threshold sliding window method, and acquiring the starting point A and the ending point B of a complete waveform of the partial discharge signal according to the threshold Th and the window width M;

[0010] Step 4: the signal between point A and point B is taken as a complete partial discharge signal waveform and is saved;

[0011] Step 5: step 3 is repeated until the end of the partial discharge signal x(t) is reached, and then the moving is stopped, to obtain a denoised preprocessed partial discharge signal x1(t);

[0012] Step 6: the partial discharge signal x1(t) is transformed into a frequency domain by using FFT on Matlab, to obtain a partial discharge signal x2(t) in the frequency domain;

[0013] Step 7: the overcomplete redundant function is used to process the signal x2(t), and an overcomplete atom library D={g γ} is constructed, wherein α is a coefficient set; is an M*N order base function matrix composed of base function vectors, g γ is an atom defined by a parameter group γ;

[0014] Step 8: the signal x2(t) is resonant sparse decomposed by a high-pass filter and a low-pass filter respectively, to obtain a high resonant decomposed signal x G (t) and a low resonant decomposed signal x D (t);

[0015] Step 9: the high resonant decomposed signal x G (t) and the low resonant decomposed signal x D (t) are input to a BP neural network data processing terminal; different partial discharge test standard maps are processed by using a sparse decomposition formula, to establish a sparse overcomplete dictionary library, which is input to the BP neural network data processing terminal; the high and low resonant decomposed signals and the sparse overcomplete dictionary library are compared and analyzed in the BP neural network data processing terminal, to extract effective partial discharge pulse signals.

[0016] As a preferred scheme, the step 2 specifically comprises the following steps:

[0017] Step 2-1: extreme values are taken for the sampling signal, to obtain envelope first maximum value distribution and envelope first minimum value distribution;

[0018] Step 2-2: second extreme values are taken for the envelope first extreme value distribution, to obtain second extreme values containing local extreme points;

[0019] Step 2-3: local extreme values representing noise disturbance and interference are filtered out, and effective extreme value ranges are reserved;

[0020] Step 2-4: according to the second maximum value distribution and the second minimum value distribution, a threshold Th of a current time period is obtained by taking a mean value of extreme values of a certain length.

[0021] As another preferred solution, the step 3 of the present application specifically comprises the following steps:

[0022] Step 3-1: Starting from the first point of the partial discharge signal x(t), move x(t) successively, when the absolute values of signal amplitudes in the window at point A are all greater than the threshold value Th, take the point A as the starting point of the signal waveform, and record the point A;

[0023] Step 3-2: Continue to move the partial discharge signal x(t) until the absolute values of signal amplitudes in the window at point B are all less than the threshold value Th, that is, take the point B as the ending point of the signal waveform, and record the point B.

[0024] Secondly, the step 7 of the present application specifically comprises the following steps:

[0025] Step 7-1: Select the atom most relevant to the signal residual from the over-complete atom library, and decompose the signal into the component on the best matching atom and the residual component:

[0026] x2(t) = <x2(t), g γ0 >g γ0 + R1x2(t)

[0027] Step 7-2: Repeat the decomposition for the residual signal after the best matching for multiple times:

[0028] R n x2(t) = <R n x2(t), g γn >g γn + R n+1 x2(t)

[0029] Step 7-3: After k times of decomposition, the residual satisfies the requirement, and the sparse decomposition formula of the signal is obtained:

[0030]

[0031] Step 7-4: Process the partial discharge standard atlas of different discharge tests by using the sparse decomposition formula, establish a sparse over-complete dictionary library, and input to the BP neural network data processing terminal.

[0032] In addition, the step 8 of the present application specifically comprises the following steps:

[0033] Step 8-1: Adjust the quality factor Q, α and β are the ratio factors of low pass and high pass, and r is the redundancy;

[0034] Step 8-2: When Q = 1, it is low resonance decomposition; when Q = 3, it is high resonance decomposition;

[0035] Step 8-3: In the resonance sparse decomposition process, different decomposition layers L correspond to different center frequencies f c The transient ground voltage detection sensor detects electromagnetic wave signals of 1MHz-100MHz, and the non-contact ultrasonic sensor detects ultrasonic signals with a center frequency of 20kHz-200kHz;

[0036] Step 8-4: f s is the frequency of the original sampling signal x(t);

[0037] Step 8-5: Obtain the high resonance decomposed oscillation and periodic signal x G (t) after denoising processing, and the low resonance decomposed impact signal x D (t) rich in characteristic information.

[0038] Advantages of the present application.

[0039] The partial discharge pulse monitoring method of the present application uses the double extreme value average threshold method for denoising preprocessing, and realizes high resonance and low resonance decomposition through the resonance sparse algorithm to form the recombined signal of the denoised oscillation, periodic signal and impact signal rich in characteristic information. Together with the overcomplete atom library constructed based on the overcomplete redundant function, it is input into the BP neural network data processing terminal for comparative analysis, which can avoid interference caused by errors or noise, extract effective partial discharge pulse waveform, and adjust the response speed (corresponding to step 3 in the specific embodiment), quality factor (corresponding to step 8 in the specific embodiment) and center frequency parameter (corresponding to step 8 in the specific embodiment) to meet the requirements of partial discharge online monitoring and data analysis in different environments. BRIEF DESCRIPTION OF DRAWINGS

[0040] The present application will be further described below in conjunction with the drawings and specific embodiments. The protection scope of the present application is not limited to the following descriptions.

[0041] Figure 1 is a flow chart of the method of the present application;

[0042] Figure 2 is a waveform diagram of the original partial discharge sampling signal;

[0043] Figure 3 is a partial discharge waveform diagram. DETAILED DESCRIPTION

[0044] The specific embodiments of the present application will be further described in detail below in conjunction with the drawings and examples.

[0045] The partial discharge monitoring method based on the double extreme value average threshold method and the resonance sparse decomposition algorithm comprises the following steps:

[0046] Step 1: Place the partial discharge monitoring device sensor collection end inside the switch cabinet, connect the collection end to the data processing terminal, and collect the switch cabinet partial discharge signal x(t) through the monitoring device sensor collection end;

[0047] Step 2: As shown in Figure 2 , process the partial discharge signal x(t) according to the double-pole average threshold method to obtain the threshold Th;

[0048] Step 2-1: Take the extreme value of the sampled signal to obtain the envelope first maximum value distribution and the envelope first minimum value distribution;

[0049] Step 2-2: Continue to take the second extreme value of the envelope first extreme value distribution to obtain the second extreme value containing local extreme points;

[0050] Step 2-3: Filter out the local extreme values representing noise disturbance and interference, and retain the effective extreme value range;

[0051] Step 2-4: According to the second maximum value distribution and the second minimum value distribution, take the average of the extreme values of a certain length to obtain the threshold Th of the current period (this step considers different lengths, and the average will be different. The length can be adjusted according to the demand to adjust the threshold. Therefore, compared with the existing threshold method, this step can dynamically adjust the threshold and dynamically adapt to various disturbances such as spikes, square waves, etc.);

[0052] Step 3: Use the threshold sliding window method and obtain the starting point A and the ending point B of the complete waveform of the partial discharge signal according to the threshold Th and the window width M;

[0053] Step 3-1: Start from the first point of the partial discharge signal x(t), move x(t) one by one, and when the absolute value of the signal amplitude in the window at point A is greater than the threshold Th, take the A point as the starting point of the signal waveform and record the A point;

[0054] Step 3-2: Continue to move the partial discharge signal x(t) until the absolute value of the signal amplitude in the window at point B is less than the threshold Th, that is, take the B point as the end point of the signal waveform and record the B point.

[0055] Step 4: Take the signal between A point and B point as a complete partial discharge signal waveform and save it;

[0056] Step 5: Repeat step 3 until the end of the partial discharge signal x(t) is reached, then stop moving, and obtain the denoising preprocessed partial discharge signal x1(t);

[0057] Step 6: Transform the partial discharge signal x1(t) to the frequency domain by using FFT (Fast Fourier Analysis) on Matlab to obtain the partial discharge signal x2(t) in the frequency domain;

[0058] Step 7: Use the overcomplete redundant function to process the signal x2(t) and construct the overcomplete atom library D = {g γ}; α is the coefficient set; is the M × N order basis function matrix composed of the basis function vector g γ defined by the parameter group γ;

[0059] Step 7-1: Select the atom most relevant to the residual signal x2(t) from the overcomplete atom library, and decompose the signal into two parts: the component on the best matching atom and the residual component:

[0060] x2(t) = <x2(t), g γ0 >g γ0 + R1x2(t)

[0061] Step 7-2: Repeat the decomposition of the residual signal after the best matching:

[0062] R n x2(t) = <R n x2(t), g γn >g γn + R n+1 x2(t)

[0063] Step 7-3: After k times of decomposition, the residual satisfies the requirement, and the sparse decomposition of the signal is obtained:

[0064]

[0065] Step 7-4: Process the partial discharge standard atlas of different discharge tests using the sparse decomposition, establish a sparse overcomplete dictionary library, and input it to the BP neural network data processing terminal;

[0066] Step 8: Perform resonance sparse decomposition on the signal x2(t) through high-pass and low-pass filters to obtain high resonance decomposition signal x G (t) and low resonance decomposition signal x D (t);

[0067] Step 8-1: Adjust the quality factor Q, α and β are the proportion factors of low-pass and high-pass, and r is the redundancy;

[0068] Step 8-2: When Q = 1, it is low resonance decomposition; when Q = 3, it is high resonance decomposition;

[0069] Step 8-3: In the process of resonance sparse decomposition, different decomposition layers L correspond to different center frequencies f c The transient voltage-to-ground detection sensor detects electromagnetic wave signals of 1MHz-100MHz, and the non-contact ultrasonic sensor detects ultrasonic signals with a center frequency of 20kHz-200kHz.

[0070] Step 8-4: (f s is the frequency of the original sampling signal x(t));

[0071] Step 8-5: Obtain the high resonance decomposition oscillation and periodic signal x G (t) and the low resonance decomposition impact signal x D (t) after denoising processing;

[0072] Step 9: Input the high resonance decomposition signal x G (t) and the low resonance decomposition signal x D (t) into the BP neural network data processing terminal; process the partial discharge standard atlas of different discharge tests by using the sparse decomposition formula, establish a sparse over-complete dictionary library, and input it into the BP neural network data processing terminal; compare and analyze the high and low resonance decomposition signals and the sparse over-complete dictionary library in the BP neural network data processing terminal, and extract effective partial discharge pulse signals, as shown in Fig. Figure 3

[0073] Compared with the partial discharge signal extraction method, the processing method of steps 7, 8 and 9 of the application can effectively reduce the interference of noise on the extraction effect before sparse decomposition; the sparse over-complete atom library based on different discharge test standard atlas is constructed and compared with the high and low resonance decomposition signals, which can more effectively identify and extract accurate partial discharge signals.

[0074] The partial discharge monitoring method of the application uses the double extreme value average threshold method for denoising preprocessing, and realizes high resonance and low resonance decomposition through the resonance sparse algorithm to form the denoised oscillation, periodic signal and impact signal rich in characteristic information. Together with the over-complete atom library constructed based on the over-complete redundant function, input into the BP neural network data processing terminal for comparison and analysis, which can avoid interference caused by errors or noise, extract effective partial discharge pulse waveforms, and adjust the response speed, quality factor and center frequency and other parameters to meet the requirements of partial discharge online monitoring and data analysis in different environments.

[0075] ​It can be understood that the above specific description of the present application is only for illustrating the present application and is not limited to the technical solutions described in the embodiments of the present application. Those skilled in the art should understand that the present application can still be modified or replaced equivalently to achieve the same technical effects. As long as the use needs are met, it is within the protection scope of the present application.

Claims

1. A method for partial discharge monitoring of a high voltage switchgear, characterized in that The method comprises the following steps: Step 1: a partial discharge monitoring device sensor acquisition end is arranged in a switch cabinet, the acquisition end is connected to a data processing terminal, and a partial discharge signal x(t) of the switch cabinet is collected through the monitoring device sensor acquisition end; Step 2: the partial discharge signal x(t) is processed according to a double-pole average threshold method to obtain a threshold Th; Step 3: a threshold sliding window method is adopted, and a starting point A and an ending point B of a complete waveform of the partial discharge signal are obtained according to the threshold Th and a window width M; Step 4: a signal between the A point and the B point is taken as a complete waveform of the partial discharge signal and is saved; Step 5: step 3 is repeated until a terminal point of the partial discharge signal x(t) is reached, and then the moving is stopped, so that a denoising preprocessed partial discharge signal x1(t) is obtained; Step 6: the partial discharge signal x1(t) is transformed into a frequency domain by using FFT on Matlab, and a partial discharge signal x2(t) in the frequency domain is obtained; Step 7: Using overcomplete redundant function The signal x2(t) is processed to construct an overcomplete atom library D = {g γ} ; α is a coefficient set; is an M x N order basis function matrix composed of basis function vectors, g γ is an atom defined by a parameter set γ; Step 8: The signal x2(t) is resonant-sparse decomposed by a high-pass filter and a low-pass filter respectively, to obtain a high resonant decomposition signal x G (t) and a low resonant decomposition signal x D (t); Step 9: high resonance decomposition signal x G (t) and low resonance decomposition signal x D (t) are input to the BP neural network data processing terminal; The partial discharge standard atlas of different discharge tests is processed by using the sparse decomposition formula, a sparse overcomplete dictionary library is established, and is input into a BP neural network data processing terminal; The high and low resonance decomposition signals and the sparse overcomplete dictionary library are compared and analyzed in the BP neural network data processing terminal, and effective partial discharge pulse signals are extracted; The step 2 specifically comprises the following steps: Step 2-1: extreme values of the sampling signal are taken to obtain envelope first maximum value distribution and envelope first minimum value distribution; Step 2-2: second extreme values of the envelope first extreme value distribution are taken to obtain second extreme values containing local extreme points; Step 2-3: local extreme values representing noise disturbance and interference are filtered out, and effective extreme value ranges are reserved; Step 2-4: according to the second maximum value distribution and the second minimum value distribution, the threshold Th of the current period is obtained by taking the mean value of the extreme values of a certain length; The step 3 specifically comprises the following steps: Step 3-1: starting from a first point of the partial discharge signal x(t), the x(t) is moved successively, when the absolute values of signal amplitudes in a window at the A point are all greater than the threshold Th, the A point is taken as a starting point of a signal waveform, and the A point is recorded; Step 3-2: the partial discharge signal x(t) is continuously moved until the absolute values of signal amplitudes in a window at the B point are all less than the threshold Th, that is, the B point is taken as an ending point of the signal waveform, and the B point is recorded.

2. The partial discharge monitoring method of a high-voltage switchgear according to claim 1, characterized in that The step 7 specifically comprises the following steps: Step 7-1: the atom most related to the signal x2(t) residual is selected from the overcomplete atom library, and the signal is decomposed into two parts of a component on the best matching atom and a residual component: x2(t) = <x2(t), g γ0 >g γ0 + R1x2(t) Step 7-2: the residual signal after the best matching is repeatedly decomposed for multiple times: R n x2(t) = <R n x2(t), g γn >g γn +R n+1 x2(t) Step 7-3: after k times of decomposition, the residual satisfies the requirement, and a sparse decomposition formula of the signal is obtained: Step 7-4: the partial discharge standard atlas of different discharge tests is processed by using the sparse decomposition formula, a sparse overcomplete dictionary library is established, and is input into a BP neural network data processing terminal.

3. The partial discharge monitoring method of a high-voltage switchgear according to claim 1, characterized in that The step 8 specifically comprises the following steps: Step 8-1: Adjusting the quality factor Q, a and β are the low-pass and high-pass scaling factors, and r is the redundancy. Step 8-2: when Q=1, it is low resonance decomposition; when Q=3, it is high resonance decomposition; Step 8-3: In the process of resonance sparse decomposition, different decomposition layers L correspond to different center frequencies f c The transient voltage to ground detection sensor detects electromagnetic wave signals of 1 MHz-100 MHz, and the non-contact ultrasonic sensor detects ultrasonic signals with a center frequency of 20 kHz-200 kHz. Step 8-4: f s is the frequency of the original sampled signal x(t). Step 8-5: Obtain the de-noised high-resolved oscillation and periodic signal x G (t), and the low-resolved shock signal x D (t) rich in feature information.

Images