Ultrahigh frequency partial discharge signal denoising method and device and storage medium

Abstract

The invention relates to an ultrahigh-frequency partial discharge signal denoising method and device and a storage medium. The method comprises the following steps: S1, acquiring an ultrahigh-frequency original signal; s2, carrying out wavelet decomposition on the collected ultrahigh frequency original signal; s3, extracting an upper envelope spectrum and a lower envelope spectrum of the decomposed highest-layer signal by adopting Hilbert transform to obtain an envelope signal, processing the envelope signal by adopting a moving mean smoothing method, and determining an error band based on a mean value of fragments without partial discharge in the smoothed envelope signal, determining the specific starting point and the specific ending point of the partial discharge signal based on the intersection point of the error band and the smoothed envelope signal; s4, extracting a partial discharge signal from the ultrahigh frequency original signal based on the specific starting and ending points of the partial discharge signal, performing singular value decomposition on the extracted partial discharge signal, and dividing singular values by adopting a clustering mode; and S5, based on a clustering result, performing zero setting on the signal of the non-partial discharge segment to obtain a de-noised ultrahigh frequency signal. Compared with the prior art, the noise reduction effect is improved by cutting more accurate starting and ending points of partial discharge signals.

Description

Technical Field

[0001] The present invention relates to the technical field of ultra-high frequency partial discharge detection of power equipment, and particularly relates to an ultra-high frequency partial discharge signal denoising method, device and storage medium. Background Art

[0002] Defects that are not easily detectable are likely to occur inside power equipment. Partial discharge will exist in the defective parts, which is not only an omen of defect deterioration but also can reflect the location of the defect. Therefore, the detection and diagnosis of partial discharge in power equipment are important criteria for judging whether the equipment can operate for a long time.

[0003] The ultra-high frequency (UHF) method has high detection sensitivity and strong anti-interference ability, and is widely used in the partial discharge detection of various power equipment, such as insulating bushings, GIS, switch cabinets, etc. The UHF method has good detection effects for both the location of defect positions and the identification and classification of defect types in partial discharge detection. However, in practice, the UHF sensor will be affected by electromagnetic interference outside the equipment, such as corona discharge. In addition, there will be thermal noise during the observation of UHF signals using equipment such as oscilloscopes and acquisition cards. When the UHF signal generated by partial discharge at the defective part is relatively weak, the aliased noise will greatly affect the subsequent feature extraction and fault diagnosis of the UHF signal.

[0004] Most of the research on the noise reduction method for partial discharge signals is for the high-frequency current partial discharge signals of cables and the ultrasonic partial discharge signals of power equipment, and there is relatively little research on the noise reduction of ultra-high frequency partial discharge signals. Common noise reduction methods include short-time Fourier transform (STFT) and empirical mode decomposition (EMD), but the effects are often not good. The disadvantage of the Fourier transform is that the fixed window size causes poor time resolution in the low-frequency band and poor frequency resolution in the high-frequency band of the signal. The empirical mode decomposition and its improved algorithm, the ensemble empirical mode decomposition method (EEMD), will have the problems of end effect and mode component aliasing. To solve the problems of the empirical mode decomposition and its improved algorithms, the variational mode decomposition (VMD) method has also emerged, but the performance of VMD decomposition is greatly affected by its own parameters such as the number of decomposition layers and the penalty factor.

[0005] In response to this, some improved existing technologies use wavelet transform to replace Fourier transform and combine singular value decomposition to attempt to solve the above problems. For example, Chinese Patent CN 116383609 A discloses a partial discharge signal denoising method combining singular value decomposition and wavelet transform, belonging to the field of signal processing. The method includes: selecting an appropriate Hankel matrix to construct an adaptive singular value decomposition model, which is responsible for the singular value decomposition of the original noisy partial discharge signal; calculating the singular entropy increment, and finding the position with the largest bending degree of the singular entropy curve according to its asymptotic properties; using the position with the largest curvature as the threshold, and using the singular values greater than the threshold to reconstruct the original signal; selecting an appropriate wavelet basis function to construct a one-dimensional two-level dual-tree complex wavelet transform, which is responsible for the decomposition and reconstruction of the signal; using the q-shift scheme to jointly construct the filter bank inside the DT-CWT. The above denoising method expects to improve the denoising effect by suppressing narrowband noise and white noise on the PD signal and retaining the waveform mutation details. However, the above denoising method still has the following defects:

[0006] 1. This existing technology mainly relies on wavelet transform and singular value decomposition, and it is difficult to comprehensively remove various types of noise.

[0007] 2. It is difficult to accurately define the start and end moments of the partial discharge signal in the ultra-high frequency signal, which will affect the accuracy of subsequent data analysis;

[0008] 3. It may cause key information to be lost during noise reduction, affecting the accurate analysis of the partial discharge signal. Summary of the Invention

[0009] The purpose of the present invention is to provide an ultra-high frequency partial discharge signal denoising method, device and storage medium.

[0010] The purpose of the present invention can be achieved by the following technical solutions:

[0011] An ultra-high frequency partial discharge signal denoising method includes:

[0012] Step S1: Collect the ultra-high frequency original signal;

[0013] Step S2: Perform wavelet decomposition on the collected ultra-high frequency original signal;

[0014] Step S3: Use Hilbert transform to extract the upper and lower envelope spectra of the highest-level signal after decomposition to obtain the envelope signal, process the envelope signal using the moving average smoothing method, determine the error band based on the mean value of the segments without partial discharge in the smoothed envelope signal, and determine the specific start and end points of the partial discharge signal based on the intersection points of the error band and the smoothed envelope signal;

[0015] Step S4: Extract the partial discharge signal from the UHF original signal based on the specific start and end points of the partial discharge signal, perform singular value decomposition on the extracted partial discharge signal, and divide the singular values by using the clustering method;

[0016] Step S5: Set the signals of the non-partial discharge segments to zero based on the clustering result to obtain the denoised UHF signal.

[0017] The UHF original signal is a discrete signal. The wavelet function used in the wavelet decomposition process in Step S2 is the db4 wavelet, the decomposition level is 5 layers, and the wavelet decomposition coefficients are:

[0018]

[0019] where: W ψ (j,k) is the wavelet decomposition coefficient at scale j and displacement k, and x[n] is the UHF original signal. is the complex conjugate.

[0020] The mathematical expression of the Hilbert transform in Step S3 is:

[0021]

[0022] where: H{x5(t)} is the Hilbert transform of the highest layer signal x5(t) after wavelet decomposition, A(t) is the analytic signal, and E(t) is the envelope signal.

[0023] The mathematical expression of the moving average smoothing method in Step S3 is:

[0024]

[0025] where: is the smoothed envelope signal, N is the size of the smoothing window, and N is an odd number.

[0026] In Step S3, the error band is centered on the mean value of the segments without partial discharge and is established with a 5% up and down fluctuation as the boundary. The starting point of the partial discharge signal is the first intersection point of the error band and the smoothed envelope signal, and the ending point of the partial discharge signal is the last intersection point of the error band and the smoothed envelope signal;

[0027] The specific steps of the singular value decomposition in Step S4 are:

[0028] Step S4-1: Construct a matrix, and transform the partial discharge signal into a decomposable Hankel matrix H;

[0029] Step S4-2: Perform singular value decomposition on the Hankel matrix H to obtain the singular values σ of the matrix H i

[0030] In the step S4-2, the mathematical expression of singular value decomposition is as follows:

[0031] H = UΔV T

[0032] Δ = diag(σ i )

[0033] where: U is an m-dimensional orthogonal matrix, Δ is the diagonal element of the diagonal matrix A, V is an n-dimensional orthogonal matrix, and σ1 > σ2 > … > σ min(n,m) .

[0034] The clustering method adopted in the step S4 is K-means clustering.

[0035] The step S5 specifically includes:

[0036] Step S5-1: Select the optimal class among all clustering results;

[0037] Step S5-2: And determine the appropriate i′ based on the optimal class. By retaining σ1, σ2,..., σ i′ , setting all the values after σ i′ to 0, reconstructing to obtain the reconstructed Hankel matrix, and obtaining the denoised ultra-high frequency signal based on the reconstructed Hankel matrix.

[0038] An ultra-high frequency partial discharge signal denoising device includes a memory, a processor, and a program stored in the memory. When the processor executes the program, the above-mentioned method is implemented.

[0039] A storage medium stores a program, and when the program is executed, the above-mentioned method is implemented.

[0040] Compared with the prior art, the present invention has the following beneficial effects:

[0041] 1. Better denoising effect: A variety of methods such as Hilbert transform, moving average smoothing method, and K-means clustering are comprehensively used. The Hilbert transform extracts the upper and lower envelope spectra, and the moving average smoothing method processes the envelope signal, which can effectively remove some noise interference, make the signal smoother, and reduce the influence of noise on the signal characteristics. K-means clustering is used to divide the singular values. According to the clustering results, the optimal clustering is selected and retained, which can more reasonably retain the singular values related to the partial discharge signal and remove the singular values related to noise. Compared with the prior art that simply selects singular values based on the singular entropy increment and curvature, the denoising effect is better. In the test of a certain substation, the noise suppression ratio of the method of the present application reaches 16.35, which is higher than the method of directly using wavelet decomposition and other methods, strongly proving its superior denoising performance.

[0042] 2. More accurate positioning of signal start and end points: By selecting the signal at Level 5 through wavelet decomposition, after processing with Hilbert transform and moving average smoothing method, the mean value of the first 1500 points of the fragment signal without partial discharge is used to establish a ±5% error band to determine the start and end points of the partial discharge signal. This method fully considers the local characteristics and overall trend of the signal. Compared with the existing technology, it can more accurately locate the occurrence and end moments of the partial discharge signal in the entire UHF signal, providing a more accurate data range for subsequent analysis.

[0043] 3. Better preservation of signal features: During the noise reduction process, by reasonably setting the parameters of singular value decomposition and K-means clustering, such as specifying the number of clusters and the random number seed, while reducing white noise and narrowband noise, it can effectively preserve the characteristic frequency bands near 200 MHz, 360 MHz, and 720 MHz of the partial discharge signal, retain the detailed part of the signal, and avoid signal distortion. The existing technology does not mention such effective measures for preserving the signal characteristic frequency bands, which may lead to the loss of some key information when processing the signal. Description of the Drawings

[0044] Figure 1 Schematic diagram of the detected UHF original signal in the embodiment;

[0045] Figure 2 Schematic diagram of the 5-level wavelet decomposition of the UHF original signal in the embodiment;

[0046] Figure 3 Schematic diagram of the signal envelope and smoothed graph after wavelet decomposition in the embodiment;

[0047] Figure 4 Schematic diagram of the start and end positions of the partial discharge signal in the embodiment;

[0048] Figure 5 Schematic diagram of the singular value decomposition of the partial discharge signal in the embodiment;

[0049] Figure 6 Schematic diagram of the UHF signal after noise reduction in the embodiment; Figure 7 Schematic diagram of the main step flow of the method of the present invention. Detailed Embodiment

[0050] The present invention will be described in detail below with reference to the drawings and specific embodiments. This embodiment is implemented on the premise of the technical solution of the present invention, and gives the detailed implementation manner and specific operation process, but the protection scope of the present invention is not limited to the following embodiments.

[0051] Embodiment 1

[0052] A method for denoising UHF partial discharge signals, as Figure 7 shown, includes:

[0053] Step S1: Collect the UHF raw signal;

[0054] In this embodiment, an oscilloscope is specifically used to detect the UHF signal. The sampling rate of the oscilloscope is 10 G S / s, the sampling time is 500 ns, and the number of sampling points is 5,000, so as to obtain the UHF raw signal. Based on this, the UHF raw signal is a discrete signal.

[0055] Step S2: Perform wavelet decomposition on the collected UHF raw signal. Among them, the wavelet function used in the wavelet decomposition process is the db4 wavelet, the decomposition level is 5 layers, and the wavelet decomposition coefficients are:

[0056]

[0057] Among them: W ψ (j,k) is the wavelet decomposition coefficient at scale j and displacement k, and x[n] is the UHF raw signal. is the complex conjugate.

[0058] Wavelet decomposition obtains the information of the signal at different scales and positions by changing the scale parameter and displacement parameter. Through the above means, the low-frequency part can be separated from the high-frequency partial discharge signal.

[0059] Step S3: Use Hilbert transform to extract the upper and lower envelope spectra of the highest layer signal after decomposition to obtain the envelope signal, use the moving average smoothing method to process the envelope signal, determine the error band based on the mean value of the segments without partial discharge in the smoothed envelope signal, and determine the specific start and end points of the partial discharge signal based on the intersection points of the error band and the smoothed envelope signal;

[0060] The mathematical expression of the Hilbert transform in Step S3 is:

[0061]

[0062] Among them: H{x5(t)} is the Hilbert transform of the highest layer signal x5(t) after wavelet decomposition, A(t) is the analytic signal, and E(t) is the envelope signal.

[0063] The mathematical expression of the moving average smoothing method in Step S3 is:

[0064]

[0065] Among them: is the smoothed envelope signal, N is the size of the smoothing window, and N is an odd number.

[0066] In step S3, the error band is centered on the mean value of the segment without partial discharge and is established with a 5% fluctuation up and down as the boundary. The starting point of the partial discharge signal is the first intersection point of the error band and the smoothed envelope signal, and the ending point of the partial discharge signal is the last intersection point of the error band and the smoothed envelope signal;

[0067] Step S4: Extract the partial discharge signal from the UHF original signal based on the specific starting and ending points of the partial discharge signal, perform singular value decomposition on the extracted partial discharge signal, and divide the singular values by the clustering method;

[0068] The specific steps of singular value decomposition in step S4 are as follows:

[0069] Step S4-1: Construct a matrix, transform the partial discharge signal into a decomposable Hankel matrix H,

[0070]

[0071] where m + n - 1 = N2; m = N2 / 10, and N2 is a natural number

[0072] Step S4-2: Perform singular value decomposition on the Hankel matrix H to obtain the singular value σ of matrix H i

[0073] In step S4-2, the mathematical expression of singular value decomposition is:

[0074] H = UΔV T

[0075] Δ = diag(σ i )

[0076] where: U is an m-dimensional orthogonal matrix, Δ is the diagonal element of the diagonal matrix A, V is an n-dimensional orthogonal matrix, and σ1 > σ2 > … > σ min(n,m) .

[0077] The clustering method adopted in step S4 is K-means clustering.

[0078] Step S5: Set the signals of the non-partial discharge segments to zero based on the clustering results to obtain the denoised UHF signal, which specifically includes:

[0079] Step S5-1: Select the optimal class among all clustering results;

[0080] Step S5-2: Determine the appropriate i′ based on the optimal class. By retaining σ1, σ2,..., σ i′ , set all the σ i′ after to zero, perform reconstruction to obtain the reconstructed Hankel matrix, and obtain the denoised UHF signal based on the reconstructed Hankel matrix.

[0081] 1. In this embodiment, wavelet decomposition (wavelet function: db4) is used to process the original UHF signal. After the 5th layer of decomposition, the UHF partial discharge signal (the detail signal part of wavelet decomposition) and the useless low-frequency component (the approximate signal part of wavelet decomposition) can be just separated. For the high-frequency Level5 layer with the highest energy after decomposition, the Hilbert envelope is obtained and smoothed by moving average to get the contour. According to the contour line, the occurrence and end times of the partial discharge signal in the entire UHF signal can be found more accurately.

[0082] 2. In this embodiment, the singular value decomposition method is used to denoise the partial discharge signal, and K-means clustering is used to divide the singular value sequence to find a reasonable number of singular values to retain, which improves the problem of the conventional method - selecting the place with the largest difference in singular values, resulting in too few singular values of the retained signal and signal distortion after denoising.

[0083] In this embodiment, tests are carried out on the 110 kV GIS in a substation based on the denoising method in Embodiment 1. The original data is taken from the built-in UHF sensor of a certain interval. The effective frequency band of the UHF sensor is 200 - 1500 MHz, the highest sensitivity is 1 pC, and the effective height ≥ 20 cm. An oscilloscope is used to detect the UHF pulse signal of partial discharge. The sampling rate of the oscilloscope is 10 G S / s, the sampling time is 500 ns, and the number of sampling points is 5000.

[0084] Figure 1 is the measured UHF signal. Although the built-in sensor is used to avoid some external interferences, but by Figure 1 the time domain visible signal in (a) is still contaminated by noise, and it is impossible to distinguish the occurrence and duration of partial discharge, and it is difficult to observe the detail information. By Figure 1 the frequency domain in (b) (after removing the DC component), it can be seen that the effective frequency band of the signal collected by the sensor is roughly 200 - 1500 MHz. There are some characteristic frequency bands for partial discharge, but obvious white noise and weak narrowband noise are mixed. It is speculated that the source of the noise is that the partial discharge signal is relatively weak, and the thermal noise during the oscilloscope acquisition will be particularly obvious. At the same time, some narrowband noise and white noise will propagate along the internal busbar segment of the GIS and enter, and these noises are unavoidable.

[0085] Combined with Figure 2 As shown, in this embodiment, the original signal is first subjected to wavelet decomposition. The wavelet function uses the db4 wavelet that is closest to the partial discharge signal, and the decomposition layer is 5 layers. The reason is that after the 5th layer of decomposition, the low-frequency part (0 - 0.16 GHz) can be just separated from the high-frequency partial discharge signal. The signal of Level 5 layer containing the most energy is selected to search for the specific start and end points of the partial discharge signal.

[0086] After that, the Hilbert transform is used to extract the envelope spectrum of Level 5. The very high frequency partial discharge signal is almost symmetric up and down. Combining with Figure 3 As shown, in this embodiment, the upper and lower envelopes are extracted simultaneously, and then the moving average smoothing method is used for the upper and lower envelope signals, and the smoothing factor is selected as 0.2.

[0087] For the smoothed signal, the average value is taken for the first 1500 points of the segment signal where no partial discharge obviously occurs, and then an error band of ±5% is established. The first intersection point of the error band and the partial discharge signal is the starting point of the partial discharge, and the last intersection point of the error band and the signal is the ending point of the partial discharge signal.

[0088] Combining with Figure 4 As shown, the basis for the upper and lower envelope lines and their respective partial discharge selection points is the peak of their respective envelope lines. The starting point of the partial discharge is the first intersection point of the error band on the left side of the peak and the envelope line. The intersection points of the upper envelope and the lower envelope are both at the 2208th point. The ending point of the partial discharge is the first intersection point of the error band on the right side of the peak and the envelope line. The intersection points of the upper envelope and the lower envelope are both at the 4198th point. It can be seen that the partial discharge starts at about 220 ns and ends at about 419 ns. At the same time, it can be seen that the oscillation of the very high frequency signal is also very symmetric.

[0089] Combining with Figure 5 As shown, subsequently, the singular value decomposition is performed on the partial discharge signal (220 ns - 419 ns), and then the K-means clustering is used to classify the singular value sequence σ i . Although the K-means clustering algorithm is simple and easy to implement, it has some limitations: it is sensitive to the selection of the initial clustering center, may fall into a local optimum, and requires the clustering number k to be specified in advance.

[0090] From Figure 5 (a)'s singular value sequence σ i It can be observed that as the sequence increases, the singular value first drops rapidly, then drops slowly, and finally tends to 0. Therefore, the given clustering number is 3, and σ i is divided into three categories, and the initial clustering center will affect the clustering effect, resulting in inconsistent clustering each time. Therefore, when actually programming, the random number seed is specified as 0 to ensure that the clustering results are basically the same after repeated clustering.

[0091] The clustering result is that the sequences σ1 - σ3 are one category, σ4 - σ39 are one category, and σ40 and all subsequent σ are one category. As mentioned above, the more σ is retained, the closer the signal is to the original, and the weaker the noise reduction effect. According to the clustering result, in this embodiment, the first two clusters are selected to be retained, that is, σ1 - σ39 are retained, and all σ from σ40 onwards are set to zero. The partial discharge signal is reconstructed, and the signal of the non-partial discharge segment is set to zero. Finally, the noise-reduced very high frequency signal y' is obtained, as specifically shown in Figure 6 As shown.

[0092] In the time domain, the wavelet decomposition of the UHF signal combined with the Hilbert envelope and smoothing operations retains the partial discharge signals (220 ns - 419 ns) and removes the signals of the remaining non-partial discharge segments, improving the signal-to-noise ratio. In the frequency domain, after the singular value decomposition of the partial discharge signal combined with K-means clustering for noise reduction, the white noise and narrowband noise of the partial discharge signal are reduced. At the same time, the characteristic frequency bands near 200 MHz, 360 MHz, and 720 MHz of the partial discharge signal are retained, and the details of the partial discharge signal are retained without signal distortion after noise reduction.

[0093] For the measured signal, the noise suppression ratio μ can be used for quantitative analysis of the noise reduction effect. The noise suppression ratio formula is: In the formula, α1 and α2 are the standard deviations of the signal before and after noise reduction respectively. μ reflects the prominence of the signal after noise reduction. To a certain extent, the larger μ is, the better the noise reduction is. For the original UHF signal, the μ of this embodiment is compared with the three noise reduction methods of directly using wavelet decomposition (DWT, wavelet function: db4), empirical mode decomposition (EMD), and singular value decomposition (SVD). The results are shown in Table 1.

[0094] Table 1 Calculation results of the noise suppression ratio of the noise reduction method

[0095]

[0096] Combined with Table 1, it can be seen that the μ of the noise reduction method in this embodiment is larger. In this embodiment, if more singular values σ are selected to be removed, the noise suppression ratio will be even larger. However, for the noise reduction of UHF partial discharge signals, the method in this embodiment has better superiority in terms of time-frequency domain and noise suppression ratio.

[0097] If the above functions are implemented in the form of software function units and sold or used as independent products, they can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art, or a part of this technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to enable a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The foregoing storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

Claims

1. A method for denoising ultra-high frequency partial discharge signals, characterized in that, Including: Step S1: Collect the UHF raw signal; Step S2: Perform wavelet decomposition on the collected UHF raw signal; Step S3: Use Hilbert transform to extract the upper and lower envelope spectra of the highest-level signal after decomposition to obtain the envelope signal, process the envelope signal using the moving average smoothing method, determine the error band based on the mean value of the segments without partial discharge in the smoothed envelope signal, and determine the specific start and end points of the partial discharge signal based on the intersection points of the error band and the smoothed envelope signal; Step S4: Extract the partial discharge signal from the UHF raw signal based on the specific start and end points of the partial discharge signal, perform singular value decomposition on the extracted partial discharge signal, and divide the singular values using the clustering method; Step S5: Set the signals of the non-partial discharge segments to zero based on the clustering results to obtain the denoised UHF signal.

2. The method for denoising ultra-high frequency partial discharge signals according to claim 1, characterized in that The UHF raw signal is a discrete signal. The wavelet function used in the wavelet decomposition process in Step S2 is the db4 wavelet, the decomposition level is 5 layers, and the wavelet decomposition coefficients are: Where: W ψ (j, k) is the wavelet decomposition coefficient at scale j and displacement k, and x[n] is the ultra-high frequency original signal, is the complex conjugate.

3. A method for denoising ultra-high frequency partial discharge signals according to claim 2, characterized in that, The mathematical expression of the Hilbert transform in Step S3 is: Where: H{x5(t)} is the Hilbert transform of the highest-level signal x5(t) after wavelet decomposition, A(t) is the analytic signal, and E(t) is the envelope signal.

4. A method for denoising very high frequency partial discharge signals according to claim 2, characterized in that, The mathematical expression of the moving average smoothing method in Step S3 is: Wherein: is the smoothed envelope signal, N is the size of the smoothing window, which is an odd number.

5. A method for denoising ultra-high frequency partial discharge signals according to claim 2, characterized in that, In Step S3, the error band is centered on the mean value of the segments without partial discharge and is established with a 5% upper and lower floating boundary. The first intersection point of the error band and the smoothed envelope signal is the start point of the partial discharge signal, and the last intersection point of the error band and the smoothed envelope signal is the end point of the partial discharge signal; The specific steps of the singular value decomposition in Step S4 are: Step S4-1: Perform matrix construction to transform the partial discharge signal into a decomposable Hankel matrix H; Step S4-2: Perform singular value decomposition on the Hankel matrix H to obtain the singular value σ of matrix H i .

6. A method for denoising ultra-high frequency partial discharge signals according to claim 5, characterized in that, The mathematical expression of the singular value decomposition in Step S4-2 is: H = UAV T Δ = diag(σ i ) where: U is an m-dimensional orthogonal matrix, Δ is the diagonal element of the diagonal matrix A, V is an n-dimensional orthogonal matrix, and σ1 > σ2 > … > σ min(n，m) .

7. A method for denoising ultra-high frequency partial discharge signals according to claim 5, characterized in that The clustering method used in Step S4 is K-means clustering.

8. A method for denoising very high frequency partial discharge signals according to claim 5, characterized in that, Step S5 specifically includes: Step S5-1: Select the optimal class among all clustering results; Step S5-2: And determine the appropriate i' based on the optimal class. By retaining σ1, σ2,..., σ i′ , set all the elements after σ i′ to 0, perform reconstruction to obtain the reconstructed Hankel matrix, and obtain the denoised UHF signal based on the reconstructed Hankel matrix.

9. A very high frequency partial discharge signal denoising device, comprising a memory, a processor, and a program stored in the memory, characterized in that, When the processor executes the program, it implements the method described in any one of claims 1-8.

10. A storage medium having a program stored thereon, characterized in that, When the program is executed, it implements the method described in any one of claims 1-8.

Images