A partial discharge live detection method for distribution network equipment in strong noise condition

Abstract

The application discloses a kind of strong noise condition distribution network equipment partial discharge live detection method, applied to live detection and operation and maintenance technical field, including the following specific steps: obtaining the partial discharge signal of multiple-end detection unit;The partial discharge signal is continuously suppressed, and pulse signal extraction is carried out;According to the extracted pulse signal, the position statistical scatter diagram of pulse source is drawn;The number and position of pulse source are estimated by Gaussian mixture model fitting the position statistical scatter diagram;Select the pulse signal in fixed confidence interval to separate multiple-source pulse, and draw the power frequency phase spectrum of different pulse sources to identify partial discharge.The application only relies on the position difference of pulse source to realize the separation of partial discharge source and noise source, and does not use the intensity information of signal, can remove pulse noise of any intensity, so as to ensure the effective detection and identification of partial discharge under high noise condition.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of live detection and operation and maintenance, and more particularly to a partial discharge live detection method for distribution network equipment under strong noise condition. BACKGROUND

[0002] Power equipment is the carrier of electric energy transmission, and its integrity and stability directly affect the reliability of the power supply system. Once the power equipment fails, it may cause the power supply system to fail to work, resulting in large-scale power outages, causing huge economic losses, and even causing personal injury and death. According to statistics, more than 70% of power equipment failures are caused by insulation defects. Therefore, carrying out insulation state evaluation of power equipment is an important measure to reduce its failure rate.

[0003] At present, partial discharge detection is one of the most common and effective ways for power equipment insulation state evaluation. Partial discharge detection mainly includes two categories: offline detection and online (live) detection. Offline detection requires the power equipment to be taken out of operation, and the insulation of the power equipment is tested by another high-voltage power supply, and at the same time, partial discharge detection is carried out. The advantage of offline detection is low noise interference level and high sensitivity, but its disadvantage is that the power equipment needs to be taken out of operation. However, in the power system, most of the power equipment is difficult to be taken out of operation, so the offline detection method can only be carried out before the power equipment is put into operation, and the coverage is very small. Online detection is to carry out partial discharge detection during the operation of the power equipment, which has the advantages of not needing the power equipment to be taken out of operation and convenient field implementation, but the disadvantage is that it is easily affected by the noise interference on the site, especially the pulse noise generated by power electronic devices, which has a waveform and spectrum similar to that of partial discharge signals, and seriously affects the sensitivity of online partial discharge detection.

[0004] In recent years, many partial discharge noise reduction methods have been proposed by domestic and foreign scholars, mainly including filtering technology, statistical technology, waveform analysis and pattern recognition. Filtering technology extracts the pulse component by analyzing the frequency domain or time-frequency domain of the signal, and realizes partial discharge signal noise reduction. However, it is almost ineffective for pulse noise with similar waveform to partial discharge. Statistical technology uses mean, variance, kurtosis and other statistical parameters to distinguish partial discharge signals and pulse noise. Waveform analysis defines time domain and frequency domain waveform parameters to represent the difference between partial discharge signals and pulse noise. However, once the pulse noise has similar waveform characteristics to the partial discharge pulse, the accuracy and reliability of these statistical techniques and waveform analysis methods will be greatly reduced. Pattern recognition method extracts partial discharge signals by identifying the characteristic differences between local discharge and noise signals in the signal, mainly including genetic algorithm, support vector machine, artificial neural network and deep learning algorithm. However, the classifier of the trained algorithm needs a large number of samples, which is actually unrealistic for on-site partial discharge measurement.

[0005] Therefore, further improving and enhancing the anti-interference ability of the partial discharge on-line (live) detection technology, realizing accurate, stable and efficient partial discharge on-line (live) detection is a problem to be solved by those skilled in the art. SUMMARY

[0006] Therefore, the application provides a partial discharge live detection method for distribution network equipment in a strong noise condition.

[0007] In order to achieve the above-mentioned purpose, the application adopts the following technical scheme:

[0008] A partial discharge live detection method for distribution network equipment in a strong noise condition comprises the following specific steps:

[0009] Obtaining partial discharge signals of a multi-terminal detection unit;

[0010] Performing continuous noise suppression on the partial discharge signals and extracting pulse signals;

[0011] Drawing a position statistical scatter plot of pulse sources according to the extracted pulse signals;

[0012] Fitting the position statistical scatter plot by a Gaussian mixture model to estimate the number and position of the pulse sources;

[0013] Selecting pulse signals in a fixed confidence interval to separate multi-source pulses, and drawing power frequency phase spectrum graphs of different pulse sources to identify partial discharge.

[0014] Preferably, in the partial discharge live detection method for distribution network equipment in a strong noise condition, the multi-terminal detection unit comprises a partial discharge sensor, a high-voltage isolation device, a high-speed sampling device, a synchronization device and a power frequency phase detection device.

[0015] Preferably, in the partial discharge live detection method for distribution network equipment in a strong noise condition, the specific steps of obtaining partial discharge signals of a multi-terminal detection unit are as follows:

[0016] Defining signals collected by K detection units, wherein each detection signal comprises a sum of signals of M partial discharge sources, one continuous interference signal and N pulse sources.

[0017] Preferably, in the partial discharge live detection method for distribution network equipment in a strong noise condition, the specific steps of performing continuous noise suppression on the partial discharge signals are as follows:

[0018] The second generation wavelet transform is realized, a second generation wavelet transform denoising function is defined, the second generation wavelet transform denoising function is applied to each detected signal, and a denoised signal is obtained, so that the continuous noise interference signal is removed.

[0019] Preferably, in the partial discharge live-line detection method for power distribution equipment in the above strong noise condition, the pulse signal extraction step is as follows:

[0020] The pulse signal is extracted by the sliding time window method and the kurtosis criterion.

[0021] A sliding time window is defined, and the kurtosis value of the signal in the time window is calculated.

[0022] If the kurtosis value is greater than a preset value, the denoised signal contains a pulse signal, and the pulse signal is rewritten; the above process is repeated to complete the pulse signal extraction in all detected sequence signals.

[0023] Preferably, in the partial discharge live-line detection method for power distribution equipment in the above strong noise condition, the specific steps for drawing the position statistical scatter plot of the pulse source are as follows:

[0024] Pulse matching: correlation analysis is performed on the pulse signals of each detection unit in a specified time window;

[0025] Pulse arrival time difference estimation: the cross-spectrum energy spectrum is calculated for the two pulse signals that have completed matching, the generalized cross-correlation function is calculated through the cross-spectrum energy spectrum, and the arrival time difference of the two pulses is determined according to the maximum value of the coefficients of the generalized cross-correlation function;

[0026] Pulse source position estimation: the position of the pulse source is determined by solving the positioning equation; the above process is repeated for all matched pulses to obtain a pulse source position matrix;

[0027] A scatter plot of the pulse source position matrix is drawn, which is the position statistical scatter plot of the pulse source.

[0028] Preferably, in the partial discharge live-line detection method for power distribution equipment in the above strong noise condition, the specific steps for estimating the number and position of the pulse source are as follows:

[0029] A Gaussian mixture model with E Gaussian distributions is defined;

[0030] The number of Gaussian distributions is determined according to the number of probability density peak points in the position statistical scatter plot of the pulse source, which is the number of discharge sources;

[0031] The expectation maximization algorithm is used to calculate the parameter matrix in the Gaussian mixture model fitted by the statistical scatter plot, and the mean vector is the position of the discharge source.

[0032] Preferably, in the partial discharge live detection method of the power distribution equipment in the strong noise condition, the specific steps of the multi-source pulse separation are as follows:

[0033] The pulses in the 95% confidence interval of each Gaussian distribution in the Gaussian mixture model are defined as the same pulse source, otherwise, they are positioned as noise or other pulse sources.

[0034] Preferably, in the partial discharge live detection method of the power distribution equipment in the strong noise condition, the partial discharge source identification is:

[0035] First, the phase distribution spectrum of all pulse signals in the same pulse source is drawn;

[0036] Then, it is judged whether the spectrum has power frequency phase concentration or not, if not, it is a noise signal, if yes, it is a partial discharge signal.

[0037] According to the technical scheme, compared with the prior art, the partial discharge live detection method of the power distribution equipment in the strong noise condition is provided, the partial discharge source and the noise source are separated only by the position difference of the pulse source, the intensity information of the signal is not used, the pulse noise of any intensity can be removed, so that the effective detection and identification of the partial discharge in the strong noise condition are ensured; in addition, the method can not only eliminate the interference of the strong noise signal, but also realize the identification, separation and positioning of the multi-source partial discharge; and it is suitable for the partial discharge live (online) detection of all power equipment. BRIEF DESCRIPTION OF DRAWINGS

[0038] In order to more clearly illustrate the technical scheme in the embodiments of the present application or the prior art, the drawings needed to be used in the embodiments or the prior art description will be briefly introduced below, and obviously, the drawings in the following description are only embodiments of the present application, and those skilled in the art can obtain other drawings according to the provided drawings without creating any creative labor.

[0039] Figure 1 is the method flowchart of the present application.

[0040] Figure 2 is the hardware framework schematic diagram of the present application.

[0041] Figure 3 is the partial discharge noise reduction and pulse extraction schematic diagram of the present application.

[0042] Figure 4 is the pulse source position statistical scatter diagram of the present application.

[0043] Figure 5 is the Gaussian mixture model fitting result schematic diagram of the present application.

[0044] Figure 6 is a flow chart of the algorithm of the present application.

[0045] Fig. 7(a) is a schematic diagram of the field implementation of the present application on a three-dimensional structure power equipment.

[0046] Fig. 7(b) is a scatter plot of the position statistics of the pulse source, a Gaussian mixture model fitting result, a pulse source separation result and a respective PRPD spectrum diagram of the present application.

[0047] Fig. 8(a) is a schematic diagram of the field implementation of the present application on a one-dimensional structure power equipment.

[0048] Fig. 8(b) is a scatter plot of the position statistics of the pulse source, a Gaussian mixture model fitting result, a pulse source separation result and a respective PRPD spectrum diagram of the present application.

[0049] In the figure: 21 partial discharge sensor, 22 high-voltage isolation device, 23 high-speed sampling device, 24 synchronization device, 25 power frequency phase detection device, 71 high-voltage switch cabinet, 72 first partial discharge source, 73 second partial discharge source, 74 pulse noise source, 75 first detection unit, 76 second detection unit, 77 third detection unit, 78 fourth detection unit, 81 power cable. DETAILED DESCRIPTION

[0050] The technical solutions in the embodiments of the present application will be clearly and completely described below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the scope of protection of the present application.

[0051] The embodiments of the present application disclose a partial discharge on-line detection method for distribution network equipment under strong noise condition, as shown in Figure 1 The specific steps include:

[0052] S101 acquiring partial discharge signals of a multi-terminal detection unit;

[0053] S102 performing continuous noise suppression on the partial discharge signals and extracting pulse signals;

[0054] S103 drawing a position statistics scatter plot of the pulse source according to the extracted pulse signals;

[0055] S104 fitting the position statistics scatter plot by a Gaussian mixture model to estimate the number and position of the pulse source;

[0056] S105 selects the pulse signals in the fixed confidence interval for multi-source pulse separation, and draws a power frequency phase spectrum of different pulse sources to identify partial discharge.

[0057] To further optimize the above technical solution, in S101, as shown in the figure, the multi-end detection unit includes a partial discharge sensor, a high-voltage isolation device, a high-speed sampling device, a synchronization device, and a power frequency phase detection device. Figure 2

[0058] The partial discharge signal is collected by multiple partial discharge detection units installed at different spatial positions. The type and bandwidth of the partial discharge sensor depend on the spatial size of the power equipment to be detected.

[0059] For example, for a cable or overhead line with a large spatial scale, a high-frequency current transformer with a bandwidth of 0.1-50MHz can be selected;

[0060] For power equipment such as high-voltage switch cabinet, transformer or GIS in the station with small spatial scale, a UHF sensor with a bandwidth of 0.3-3GHz can be selected.

[0061] The high-voltage isolation device uses optical fiber isolation method to transmit the partial discharge detection signal through optical fiber.

[0062] The sampling rate of the high-speed sampling device depends on the spatial size of the power equipment to be detected, for example, for a 500m cable, a sampling rate of 250MS / s is used, and the corresponding spatial resolution is about 1m; for a 1m×2m×5m switch cabinet, a sampling rate of 5GS / s is used, and the corresponding spatial resolution is about 7cm.

[0063] The synchronization device includes GPS synchronization and optical fiber synchronization functions, and the purpose is to realize data synchronization acquisition between multiple detection units, which is the premise of accurate pulse source position analysis, for example, for a cable with a large spatial scale, a GPS synchronization method is used, and for power equipment such as high-voltage switch cabinet, transformer or GIS in the station with small spatial scale, an optical fiber synchronization method is selected.

[0064] The function of the power frequency phase detection device is to measure the power frequency high-voltage signal of the power equipment to be detected and extract the power frequency phase information for partial discharge identification and diagnosis.

[0065] To further optimize the above technical solution, the specific steps of acquiring the partial discharge signal of the multi-end detection unit are as follows:

[0066] Define the signals collected by K detection units, where each detection signal contains the sum of M partial discharge source signals, 1 continuous interference signal, and N pulse source signals.

[0067] The expression is as follows:

[0068] ​S1(t) = PD 1，1 (t) + PD 1，2 (t) … + PD 1，M (t) + CN1(t)

[0069] + IN 1，1 (t) + IN 1，2 (t) … + IN 1，N (t)

[0070] S2(t) = PD 2，1 (t) + PD 2，2 (t) … + PD 2，M (t) + CN2(t)

[0071] + IN 2，1 (t) + IN 2，2 (t) … + IN 2，N (t)

[0072]

[0073] S K (t) = PD K，1 (t) + PD k，2 (t) … + PD k，M (t) + CN K (t)

[0074] + IN K，1 (t) + IN K，2 (t) … + IN K，N (t)

[0075] where S k (t) represents the signal collected by the kth detection unit, PD k，m (t) represents the signal generated by the mth local discharge source collected by the kth detection unit, CN k represents the continuous noise signal collected by the kth detection unit, IN k，n (t) represents the signal generated by the nth pulse interference source collected by the kth detection unit.

[0076] Figure 3 is a schematic diagram of local discharge noise reduction and pulse extraction. Figure 3 (a) shows the original signal waveform collected by the sensor; Figure 3 (b) shows the waveform after noise reduction by the second generation wavelet transform; Figure 3 (c) shows the pulse signal extracted by the sliding time window method and kurtosis criterion.

[0077] In order to further optimize the above technical solutions, the continuous noise suppression of the partial discharge signal is specifically as follows:

[0078] The continuous noise interference signal is removed by the second generation wavelet transform, defining a second generation wavelet transform denoising function, applying the second generation wavelet transform denoising function to each detected signal to obtain a denoised signal.

[0079] Specifically, the continuous noise suppression is realized by the second generation wavelet transform. The second generation wavelet transform denoising function is defined as SGWT(·), which includes wavelet decomposition, thresholding processing and signal reconstruction, wherein the threshold function adopts a soft threshold function. The signal S k (t) is applied to SGWT(·) to obtain a denoised signal S' k (t) as follows:

[0080] S' k (t) = SGWT(S k (t)) ≈ PD k，1 (t) + PD k，2 (t)… + PD K，M (t)

[0081] + IN k，1 (t) + IN k，2 (t)… + IN k，N (t)

[0082] The continuous noise interference signal CN k (t) in the signal S k (t) is removed.

[0083] In order to further optimize the above technical solutions, the steps of the pulse signal extraction are as follows:

[0084] The pulse signal extraction is realized by the sliding time window method and the kurtosis criterion.

[0085] A sliding time window is defined, and the kurtosis value of the signal in the time window is calculated.

[0086] If the kurtosis value is greater than a preset value, the denoised signal contains a pulse signal, and the pulse signal is rewritten. The above process is repeated to complete the pulse signal extraction in all detected sequence signals.

[0087] Specifically, the pulse signal extraction is realized by the sliding time window method and the kurtosis criterion. A sliding time window [t slide , t slide +W] is defined, wherein t slideis the start point of the time window, W is the width of the time window, and the recommended value range is 2-20 μs, the kurtosis value of the signal in the time window is calculated, as shown in the following formula:

[0088]

[0089] where S' k [t slide , t slide +W] is the signal of S' k (t) in the time window [t slide , t slide +W], μ is the mean of S' k [t slide , t slide +W], and σ is the standard deviation of S' k [t slide , t slide +W]. If the value of kurtosis is greater than 4, it is considered that S' k [t slide , t slide +W] contains a pulse signal, and the pulse signal is rewritten as where is the start point of the time window containing the i k th pulse in S' k (t).

[0090] Let t slide =t slide +W, and repeat the above process to complete the extraction of the pulse signal in all detected sequence signals.

[0091] In order to further optimize the above technical solution, the specific steps for drawing the position statistical scatter plot of the pulse source are as follows:

[0092] Pulse matching: correlation analysis is performed on the pulse signals of each detection unit in the specified time window;

[0093] Pulse arrival time difference estimation: the cross-spectral energy spectrum is calculated for the two pulse signals that have been matched, the generalized cross-correlation function is calculated through the cross-spectral energy spectrum, and the arrival time difference of the two pulses is determined according to the maximum value of the coefficients of the generalized cross-correlation function;

[0094] Pulse source position estimation: the position of the pulse source is determined by solving the positioning equation; the above process is repeated for all matched pulses to obtain a pulse source position matrix;

[0095] A scatter plot of the pulse source position matrix is drawn, which is the position statistical scatter plot of the pulse source.

[0096] Specifically, the detailed process is as follows:

[0097] (1) Pulse matching is realized by correlation analysis of pulse signals of each detection unit within a specified time window. Two pulses in any two detection signals are matched if and the criterion is as follows:

[0098]

[0099] where cov(·,·) is the cross-correlation function, max{·} is the maximum calculation function, ΔT is the maximum time difference allowed, which depends on the physical size of the power equipment to be detected, and can be approximately estimated by ΔT = 2D / v, where D is the farthest distance of all detection units.

[0100] (2) Pulse time difference of arrival estimation

[0101] Pulse time difference of arrival estimation is realized by solving the maximum of the generalized cross-correlation coefficient. For two pulses that have been matched and First, calculate their cross-spectral energy spectrum, as follows:

[0102]

[0103] where is the Fourier transform of pulse , is the complex conjugate of the Fourier transform of pulse .

[0104] Then, calculate the generalized cross-correlation function R(τ) by Ψ(ω), as follows:

[0105]

[0106] Finally, determine the time difference of arrival of two pulses by the maximum of the generalized cross-correlation coefficient as follows:

[0107]

[0108] where is the absolute time of arrival of pulse compared to the pulse occurrence time, is the absolute time of arrival of pulse compared to the pulse occurrence time

[0109] (3) Pulse source position estimation is realized by solving the (non-)linear positioning equation. Assuming the position of the pulse source is (x0, y0, z0), it should satisfy the following relationship with the sensors of each detection unit:

[0110]

[0111] where (x k ,y k ,z k ) is the physical position of the kth detection unit sensor, v is the propagation speed of the partial discharge signal in the power equipment to be tested (or the surrounding medium). For all k and k', is known.

[0112] The above positioning equation is a nonlinear equation set, the number of unknowns of the equation is k+3, and the number of equations is Therefore, the equation is an over-determined nonlinear equation set, which can be solved by Newton-Raphson method, incremental method and the like, and then the pulse source position (x0, y0, z0) is obtained.

[0113] Repeat the above process for all matched pulses, and a pulse source position matrix L will be obtained, where L = [l1, l2, …, l H ] T , where H is the total number of calculated pulse positions, where l h = (x 0,h ,y 0,h ,z 0,h ).

[0114] (4) Draw a scatter plot of the vector L, which is a statistical scatter plot of the position of the pulse source (see Figure 2). Figure 4 ).

[0115] In order to further optimize the above technical solution, the number and position of the pulse source are estimated as follows:

[0116] Define a Gaussian mixture model with E Gaussian distributions;

[0117] According to the number of probability density peak points in the statistical scatter plot of the position of the pulse source, the number of Gaussian distributions is determined, that is, the number of discharge sources;

[0118] Using the expectation maximization algorithm, the parameter matrix in the Gaussian mixture model fitted by the statistical scatter plot is calculated, and the mean vector is the position of the discharge source.

[0119] The specific implementation method is as follows:

[0120] (1) Model definition: first define a Gaussian mixture model (GMM) with E Gaussian distributions, as shown in the following formula:

[0121]

[0122] where U = [μ1, μ2, …, μ E ] T is the mean matrix, is the mean vector, and e is the covariance matrix. Since the fluctuations of the partial discharge location in various directions are uncorrelated, let e be the off-diagonal elements of

[0123] (2) Model solution: mainly divided into the following two steps:

[0124] First, according to the position of the pulse source, the number of probability density peak points in the scatter plot is determined to determine the number of Gaussian distribution E, that is, the number of discharge source.

[0125] Second, the expectation maximization algorithm is used to calculate the parameter matrix U and of the Gaussian mixture model fitted by the scatter plot, wherein

[0126] In order to further optimize the above technical scheme, the specific steps of multi-source pulse separation are as follows:

[0127] Define the pulses in the 95% confidence interval of each Gaussian distribution in the Gaussian mixture model as the same pulse source, otherwise, it is positioned as noise or other pulse source.

[0128] In order to further optimize the above technical scheme, the partial discharge source identification:

[0129] First, draw the phase distribution spectrum of all pulse signals in the same pulse source;

[0130] Then, judge whether the spectrum has power frequency phase concentration or not, if not, it is noise signal, if yes, it is partial discharge signal.

[0131] Figure 4 is a schematic diagram of the pulse source position scatter plot of the universal partial discharge live detection and identification method of power equipment under high noise condition of the present application. From the figure, it can be reflected that there are three discharge sources in the space, and each discharge source has obvious position "concentration".

[0132] Figure 5 is a schematic diagram of the Gaussian mixture model fitting result of the universal partial discharge live detection and identification method of power equipment under high noise condition of the present application. Figure 5 is the fitting result of Figure 4 .

[0133] Fig. 7 is a schematic diagram of the field implementation of the universal partial discharge live detection and identification method of power equipment under high noise condition of the present application on a three-dimensional structure power equipment. Fig. 7 shows the implementation case of the method on a high-voltage switch cabinet.

[0134] Figure 7(a) depicts the spatial relationship of two partial discharge sources, a noise source, four partial discharge detection units and a high voltage switchgear: the partial discharge sources are located inside the high voltage switchgear, the noise source is from outside the switchgear, and the four partial discharge detection units are located around the high voltage switchgear.

[0135] Figure 7(b) respectively shows the scatter plot of pulse source location statistics, the Gaussian mixture model fitting result, the pulse source separation result and their respective PRPD spectrum. From the last PRPD spectrum, it can be seen that the first two pulse sources have obvious phase concentration, thus are diagnosed as partial discharge sources, while the third pulse source does not have phase concentration and is diagnosed as a noise source. It is worth noting that from the PRPD spectrum, it can be seen that although the pulse amplitude of the noise source is higher than that of the partial discharge source, it will not cause any impact on the identification of partial discharge, which shows the unique advantage of the method proposed in the present application under high noise level.

[0136] Figure 8 is a schematic diagram of the field implementation of a general type of partial discharge live detection and identification method of power equipment under high noise condition on a one-dimensional structure of power equipment. Figure 7 shows the implementation case of the method on power cable. Figure 8(a) depicts the spatial relationship of two partial discharge sources, a noise source, two partial discharge detection units and a power cable: the partial discharge sources are located in the middle of the power cable, the noise source is from outside the cable, and the two partial discharge detection units are located at both ends of the cable. Figure 8(b) respectively shows the scatter plot of pulse source location statistics, the Gaussian mixture model fitting result, the pulse source separation result and their respective PRPD spectrum. From the last PRPD spectrum, it can be seen that the first two pulse sources have obvious phase concentration, thus are diagnosed as partial discharge sources, while the third pulse source does not have phase concentration and is diagnosed as a noise source. Similar to Figure 7, from the PRPD spectrum, it can be seen that although the pulse amplitude of the noise source is higher than that of the partial discharge source, it will not cause any impact on the identification of partial discharge, which shows the unique advantage of the method proposed in the present application under high noise level.

[0137] The various embodiments in the specification are described in a progressive manner, and each embodiment focuses on the difference from other embodiments, and the same or similar parts between various embodiments can be referred to each other. For the device disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple, and the relevant parts can be referred to the method part.

[0138] The foregoing description of the disclosed embodiments enables a person skilled in the art to make or use the application. Modifications of these embodiments will occur to persons of skill in the art, and that the appended claims are intended to cover all such modifications that do not depart from the true spirit and scope of the application. Therefore, the application is not limited to the embodiments shown but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for detecting partial discharge in distribution network equipment under strong noise conditions, characterized in that, The specific steps include the following: Acquire partial discharge signals from the multi-terminal detection unit; The partial discharge signal is subjected to continuous noise suppression and pulse signal extraction. Draw a scatter plot of the pulse source locations based on the extracted pulse signals; The number and location of pulse sources are estimated by fitting the positional statistical scatter plot using a Gaussian mixture model. Pulse signals within a fixed confidence interval are selected for multi-source pulse separation, and power frequency phase spectra of different pulse sources are plotted to identify partial discharge. The specific steps for plotting a scatter plot of pulse source location statistics are as follows: Pulse matching: Correlation analysis is performed on the pulse signals of each detection unit within a specified time window; Pulse arrival time difference estimation: For two matched pulse signals, calculate the cross-spectral energy spectrum, and calculate the generalized cross-correlation function using the cross-spectral energy spectrum; determine the arrival time difference of the two pulses based on the maximum value of the coefficient of the generalized cross-correlation function; Pulse source position estimation: The pulse source position is determined by solving the positioning equation; the above process is repeated for all matching pulses to obtain the pulse source position matrix; Plotting a scatter plot of the pulse source position matrix yields a statistical scatter plot of the pulse source positions. The specific steps for estimating the number and location of pulse sources are as follows: Define a Gaussian mixture model with E Gaussian distributions; The number of Gaussian distributions is determined by the number of probability density peaks in the scatter plot of the pulse source location, which is the number of discharge sources. The expectation-maximization algorithm is used to calculate the parameter matrix in the Gaussian mixture model fitted by the statistical scatter plot, where the mean vector is the location of the discharge source.

2. The method for detecting partial discharge in distribution network equipment under strong noise conditions according to claim 1, characterized in that, The multi-terminal detection unit includes: a partial discharge sensor, a high-voltage isolation device, a high-speed sampling device, a synchronization device, and a power frequency phase detection device.

3. The method for detecting partial discharge in distribution network equipment under strong noise conditions according to claim 1, characterized in that, The specific steps for obtaining the partial discharge signal of the multi-terminal detection unit are as follows: definition K Each detection unit collects signals, and each detection signal contains... M A partial discharge power supply signal, a continuous interference signal, and N The sum of signals from pulse sources.

4. The method for detecting partial discharge in distribution network equipment under strong noise conditions according to claim 1, characterized in that, The specific steps for continuous noise suppression of the partial discharge signal are as follows: This is achieved through second-generation wavelet transform. A second-generation wavelet transform denoising function is defined, and the second-generation wavelet transform denoising function is applied to each detected signal to obtain the denoised signal, thereby removing continuous noise interference signals.

5. The method for detecting partial discharge in distribution network equipment under strong noise conditions according to claim 1, characterized in that, The steps for pulse signal extraction are as follows: Pulse signal extraction was performed using the sliding time window method and kurtosis criterion. Define a sliding time window and calculate the kurtosis of the signal within the time window; If the kurtosis value is greater than the preset value, the noise-reduced signal contains a pulse signal, and the pulse signal is rewritten. Repeat the above process to extract pulse signals from all detected sequence signals.

6. The method for detecting partial discharge in distribution network equipment under strong noise conditions according to claim 1, characterized in that, The specific steps for multi-source pulse separation are as follows: In the Gaussian mixture model, pulses within the 95% confidence interval of each Gaussian distribution are defined as the same pulse source; otherwise, they are identified as noise or other pulse sources.

7. The method for detecting partial discharge in distribution network equipment under strong noise conditions according to claim 1, characterized in that, Partial discharge power source identification: First, the phase distribution spectrum of all pulse signals in the same pulse source is plotted. Then, determine whether the spectrum has power frequency phase concentration. If not, it is a noise signal; if so, it is a partial discharge signal.

Images