A cable partial discharge positioning method and device based on multi-element modal noise reduction repair

Abstract

This application discloses a method and apparatus for locating partial discharge in high-voltage AC cables, relating to the field of partial discharge detection technology. The location method includes: performing a first VMD decomposition on the acquired noisy PD signal of the high-voltage AC cable to remove noise decomposition modes, and then performing secondary denoising using wavelet decomposition; repairing the waveform of the denoised signal using the peak-valley energy difference per unit time; extracting the computational features of the repaired denoised signal and inputting them into a CNN model to evaluate the partial discharge defect level; and based on the partial discharge defect level evaluation result, performing a second VMD decomposition on the repaired denoised signal, and identifying the wavefront of the mode component with the maximum kurtosis value to achieve partial discharge location. This application can quickly and accurately locate partial discharge in high-voltage AC cables.

Description

Technical Field

[0001] This application relates to the field of partial discharge detection technology, specifically to a method and apparatus for locating partial discharge in a high-voltage AC cable. Background Technology

[0002] The main methods for online fault location of high-voltage AC cables include impedance method, traveling wave method, and artificial intelligence algorithm. Impedance method is easily affected by excessive resistance and the synchronization accuracy of measuring device. Artificial intelligence algorithm relies on a large amount of historical data. The traveling wave method achieves location by calculating the time difference of the traveling wave, which has the advantages of fast and accurate location, but requires a series of signal noise reduction and analysis methods.

[0003] A known fault location method based on multivariate mode decomposition (VMD) and kurtosis fusion decomposes noisy signals, extracts the mode corresponding to the maximum kurtosis value for traveling wave front calibration, and then uses a two-end method for traveling wave ranging to locate the fault location (PD). This method selects the mode with the largest kurtosis for wavefront calibration when the decomposition mode number K is set to 2. However, this method has the following shortcomings: high-voltage cables in actual operation contain multiple types of noise, requiring in-depth white noise removal; the key indicator of the decomposition mode number K needs optimization and adjustment; and the denoised signal requires distortion correction. To address this, another study directly selects the 5th decomposition mode for Hilbert transform to calculate the traveling wave arrival time. This method also requires in-depth denoising and distortion correction, and the wavefront calibration requires optimization and adjustment of the decomposition mode number K.

[0004] Furthermore, some studies first perform wavelet thresholding denoising on noisy PD signals, then apply VMD adaptive decomposition to optimize and select the high-frequency mode components that best reflect the characteristics of the traveling wave front. The NTEO energy operator is then used to calibrate these high-frequency mode components. When the PD signal arrives, an energy singularity signal appears in the energy spectrum of the energy operator, thus calibrating the wavefront position. Finally, the two-end method of traveling wave ranging is used to determine the PD position. While this method's energy operator can preliminarily determine the wavefront based on the arrival time of the PD, it is easily affected by interference from different mode components, such as noise modes, leading to misjudgments of the wavefront timing. Summary of the Invention

[0005] This application aims to disclose a method and apparatus for locating partial discharge in high-voltage AC cables, so as to quickly and accurately locate partial discharge in high-voltage AC cables.

[0006] Firstly, a method for locating partial discharge in cables using multimodal noise reduction and repair is provided, including: The noise-containing PD signal of the high-voltage AC cable is subjected to the first VMD decomposition to remove the noise decomposition mode, and then the wavelet decomposition method is used for secondary denoising. The peak-valley energy difference per unit time is used to repair the denoised signal waveform; The computational features of the denoised signal after repair are extracted and input into a CNN model to evaluate the level of partial discharge defects. Based on the partial discharge defect level assessment results, the repaired denoised signal is subjected to a second VMD decomposition, and the mode component with the maximum kurtosis value is identified by wavefront localization to achieve partial discharge localization.

[0007] In one example, the optimal number of decomposition modes Kp for the first VMD decomposition is determined using the following method: Using a preset fixed kurtosis threshold, for each decomposition mode number K within a predetermined numerical range, decomposition modes of the sample PD signal containing noise that are less than the fixed kurtosis threshold are deleted. The remaining decomposition modes are then summed, and the cosine similarity between the summation result and the noise-free sample PD signal is calculated. The K value with the highest similarity is selected as the optimal decomposition mode number Kp.

[0008] In one example, the kurtosis threshold of the noise-removing decomposition mode. Determined based on the decomposition mode number K, and expressed as: In the formula, , These are the upper and lower limits of the kurtosis threshold, respectively. K is the empirical adjustment coefficient. max This represents the maximum value within the range of decomposition mode number K.

[0009] In one example, the optimal restoration coefficient for restoring the denoised signal waveform is determined by the following method: extracting the peak and trough times of the sample PD signal after secondary denoising, calculating the peak-trough energy difference between the sample PD signal after secondary denoising and the sample PD signal without noise, using an equidistant trial-and-error method, taking values ​​from 0 to a predetermined value at predetermined intervals, and taking the restoration coefficient corresponding to the minimum peak-trough energy difference as the optimal restoration coefficient.

[0010] In one example, for the same type of cable, multiple calculations were performed, and the average of the optimal repair coefficients obtained from each calculation was taken as the final empirical repair coefficient.

[0011] In one example, the computed features include maximum value, average value, root mean square, phase, skewness, kurtosis, crest factor, waveform factor, and peak energy per unit time.

[0012] In one example, the optimal number of decomposition modes Kp for the second VMD decomposition is determined using the following method: With the goal of obtaining the maximum kurtosis value from the selected decomposition mode number K, the kurtosis value of each decomposition mode under each decomposition mode number K within a predetermined numerical range is calculated, and the mode decomposition mode number K corresponding to the maximum kurtosis value is selected as the optimal decomposition mode number Kp.

[0013] In one example, after locating the moment when the wavefront appears, a two-end traveling wave ranging method is used to locate the partial discharge position.

[0014] Secondly, a cable partial discharge locating device for multimodal noise reduction and repair is provided, comprising: The noise reduction module performs a first VMD decomposition to remove the noise decomposition modes from the acquired noisy PD signal of the high-voltage AC cable, and then uses wavelet decomposition for secondary noise reduction. The repair module uses the peak-valley energy difference per unit time to repair the denoised signal waveform; The evaluation module extracts the computational features of the repaired and denoised signal and inputs them into the CNN model to evaluate the level of partial discharge defects. The positioning module performs a second VMD decomposition on the repaired denoised signal based on the partial discharge defect level assessment results, and identifies the wavefront of the mode component with the maximum kurtosis value to achieve partial discharge positioning.

[0015] For example, the cable is a high-voltage photovoltaic AC cable.

[0016] Beneficial effects: This application fully utilizes kurtosis values ​​to effectively distinguish waveform characteristics during signal denoising. It employs a VMD denoising method with an adaptive kurtosis threshold, dynamically adjusting the kurtosis threshold to keep pace with the number of decomposed modes, ensuring the denoised signal retains more of the original PD signal information. A cosine similarity method is used to comprehensively evaluate the denoising effect, and the optimal number of decomposed modes can be selected based on this effect. Furthermore, wavelet decomposition is used for deep denoising, significantly filtering out white noise interference while retaining all PD signal components, further improving the denoising effect. Simultaneously, this application uses a peak-valley energy difference function to effectively correct the amplitude distortion problem of the actual denoised signal, providing a valid reference for CNN models to assess the severity of partial discharge. This allows for secondary VMD decomposition based on the assessed severity of partial discharge, and the use of the global maximum kurtosis value enables rapid selection of the optimal number of decomposed modes and optimal modal components. The NTEO operator accurately calibrates the wavefront position, providing effective data support for subsequent maintenance work.

[0017] Other features and advantages of this application will become clear from the following detailed description of exemplary embodiments with reference to the accompanying drawings. Attached Figure Description

[0018] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1 This is a schematic flowchart of the cable partial discharge signal noise reduction method according to an embodiment of this application; Figure 2 This is a schematic diagram comparing signal changes during the PD denoising process; Figure 3 This is a schematic diagram of the distribution of maximum kurtosis values ​​in quadratic VMD decomposition; Figure 4 This is a schematic diagram of the NTRO energy operator curve based on maximum kurtosis. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. The following description of at least one exemplary embodiment is merely illustrative and is in no way intended to limit this application or its application or use. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0020] Unless otherwise specifically stated, the numerical expressions and values ​​set forth in these embodiments do not limit the scope of this application. Furthermore, techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values.

[0021] According to embodiments of this application, a method for locating partial discharge in cables using multimodal noise reduction and repair is provided, for locating partial discharge in high-voltage AC cables, especially high-voltage photovoltaic AC cables. Figure 1 As shown, the method includes the following steps: Step S1: Perform VMD decomposition on the noisy PD signal of the high-voltage AC cable to remove the noise decomposition mode, and then use wavelet decomposition method for secondary denoising. High-voltage AC cables operate under harsh conditions, including high temperatures, laying conditions, environmental electromagnetic interference, and frequent system switching. As a result, signals collected using online monitoring methods are prone to contain a high proportion of noise signals, the amplitude and frequency of which can mask the PD signal. Therefore, in-depth noise reduction is required.

[0022] In the embodiments of this application, a single exponential oscillation decay model and a double exponential oscillation decay model are used as the PD signals under study, and their mathematical models are as follows: (1) Single exponential oscillation decay model (1) (2) Double exponential oscillation decay model (2) In the formula, A i τ is the signal amplitude. i f is the attenuation coefficient; ci The oscillation frequency is given. The desired signal waveform can be obtained by adjusting the above two parameters. The PD signal is obtained by superimposing the two signals, as shown in the following formula: (3) In addition, the PD signal also contains periodic narrowband interference of different amplitudes and frequencies, as well as white noise signals of a certain amplitude. Let the PD signal containing the above noise be... .

[0023] VMD is a quasi-orthogonal, completely non-recursive signal decomposition method. The signal can be decomposed into K decomposition modes (modal components). The process of denoising using adaptive VMD and cosine similarity is as follows: First, the signal is decomposed into K decomposition modes. Then, a kurtosis threshold is used to determine invalid decomposition modes, which are then deleted. Finally, the remaining decomposition modes are superimposed and summed. Cosine similarity is used to determine whether the superimposed PD signal can effectively retain the original information. If the similarity is low, it indicates that the threshold K and kurtosis threshold are not appropriately chosen. The formulas for kurtosis and cosine similarity can be found in publicly available literature and will not be repeated here.

[0024] In the above process, firstly, the selection of the number of decomposed modes K is very critical. When the value of K is small, the signal decomposition is incomplete, which will lead to overlap of information contained in the modal signals. When the value of K is large, redundant and invalid decomposed modes will be generated, which will interfere with mode filtering and reduce the efficiency of the algorithm.

[0025] Therefore, embodiments of this application combine the enumeration method and the cosine similarity method, using sample signals. (Real signal or simulated signal), by setting a fixed kurtosis threshold, such as 3.2dB, and taking values ​​of K from 2 to 20, at each K value, the decomposed modes with a kurtosis threshold are deleted, and the remaining decomposed modes are summed. The cosine similarity between the summation result and the PD signal is calculated. Finally, the K value with the highest similarity is selected as the optimal number of decomposed modes Kp.

[0026] Secondly, when using kurtosis thresholding to determine invalid decomposed modes, a reasonable kurtosis threshold needs to be selected. Mode components greater than the threshold are retained, while those less than the threshold are deleted. The selection of the kurtosis threshold directly affects the accuracy of denoising. To address this, this application employs an adaptive kurtosis threshold function, dynamically adjusting the kurtosis threshold based on the selected value of K, expressed as: (4) In the formula, The kurtosis threshold when the number of decomposition modes is K; , These are the upper and lower limits of the kurtosis threshold, for example, values ​​of 3.6 and 2.8 respectively; This is an empirical adjustment coefficient, for example, a value of 3; K max To maximize the number of decomposition modes K, for example, 20.

[0027] After removing noise by decomposing the modes using the above method, the signals are superimposed and summed to obtain the denoised signal. However, this still contains white noise, which needs to be further removed. According to an embodiment of this application, wavelet decomposition is used for secondary denoising to obtain the final denoised signal. Wavelet decomposition denoising is a signal processing technique based on wavelet transform. It involves performing wavelet transform to obtain wavelet coefficients of different frequencies, then thresholding these coefficients to remove noise. Finally, the processed coefficients undergo inverse wavelet transform to reconstruct the deeply denoised signal. Specific models for wavelet decomposition can be found in publicly available literature and will not be elaborated upon here.

[0028] The embodiments of this application fully utilize kurtosis values ​​to effectively distinguish waveform characteristics. The VMD denoising method with adaptive kurtosis threshold can dynamically adjust the kurtosis threshold in line with the number of decomposed modes, so that the denoised signal retains more of the original PD signal information. At the same time, the cosine similarity method is used to evaluate the overall denoising effect, and the optimal number of decomposed modes can be selected based on the denoising effect. In addition, the wavelet decomposition method is used for deep denoising, which can significantly filter out white noise interference and retain all PD signal components, further improving the denoising effect and laying the foundation for accurate positioning in the future.

[0029] Step S2: The peak-valley energy difference per unit time is used to repair the denoised signal waveform; The denoised signal obtained by the above method can extract waveform features, but the signal amplitude will be significantly reduced and distorted, requiring amplitude amplification to repair the denoised waveform features. According to an embodiment of this application, the peak-to-valley energy difference per unit time is used to repair the denoised signal waveform. Specifically, the process includes the following: First, the denoised signal Extract peak and valley amplitudes; for example, using the following method: (5) In the formula, t0 represents the time of the peak and trough. Because there is some small-amplitude interference signal, only signals with peak and trough amplitudes greater than 0.05mV are extracted.

[0030] Then, based on the denoised signal The repair coefficient is determined by minimizing the difference between the peak and trough amplitudes per unit time and that of the noise-free PD signal, expressed as: (6) In the formula, This represents the difference in peak and trough amplitude per unit time. T1 and T2 are the denoised signal restoration coefficients; T1 and T2 are the denoised signal values, respectively. PD signal The total number of peak and trough moments.

[0031] Take values ​​in sequence Seeking to make smallest The value is the final repair coefficient.

[0032] For the same type of cable, multiple repeated tests can be conducted to determine the optimal value. The average value is used as the final empirical repair coefficient.

[0033] Finally, based on the final determined repair coefficient Denoising signal for all times Repair: (7) In the formula, This is the denoised signal after repair.

[0034] The embodiments of this application employ a signal distortion repair method based on peak-valley energy difference, which calculates the optimal repair coefficient to ensure that the peak-valley amplitude after denoising remains consistent with the PD signal.

[0035] Step S3: Extract the computational features of the repaired and denoised signal and input them into the CNN model to evaluate the level of partial discharge defects; For example, the severity of partial discharge is divided into five assessment levels: critical defect, severe defect, general defect, potential risk, and continuous monitoring, which are respectively labeled as Type 1 to Type 5. For each type, a massive amount of denoised and repaired signals are collected. A convolutional neural network (CNN) model is trained to preliminarily assess the types of partial discharge defects in cables. The input of the CNN model consists of the calculated features of the signal, and the output is the defect type number.

[0036] For example, the signal input to the CNN model is used to calculate the feature quantity. The maximum value, average value, root mean square (RMS), phase, skewness, kurtosis, crest factor, waveform factor, and peak energy per unit time are defined as follows: Skewness measures the direction and degree of skewness in the distribution of statistical data; kurtosis describes the peak or flatness of the data distribution; crest factor is the ratio of the instantaneous amplitude of the waveform to its RMS value; waveform factor is the ratio of the maximum amplitude of the signal to its average value within a certain period; and peak energy per unit time is: (8) In the formula, This represents the peak energy per unit time.

[0037] The training sample size consists of 100 typical denoised repair signals for each of the five defect types. A total of 500 signals were collected, and nine features were calculated for each signal. The specific model training techniques are standard and will not be elaborated upon here.

[0038] Step S4: Based on the partial discharge defect level assessment results, the repaired denoised signal is decomposed using VMD, and the modal component with the maximum kurtosis value is identified by wavefront localization to achieve partial discharge localization. For example, signals with assessment levels of general defects, serious defects, and critical defects need to be precisely located. The specific process includes: First, regarding the repaired partial discharge signal A second VMD decomposition is performed to select the K value and mode components that best reflect the characteristics of the partial discharge wavefront.

[0039] In the second-order VMD decomposition process, the value of K is chosen to obtain the maximum kurtosis. Under this objective, the value of K ranges from 2 to 20. The kurtosis value of each modal component is calculated for each K value, and finally, the modal decomposition number K value corresponding to the maximum kurtosis value is selected as the optimal number of decomposed modes.

[0040] Then, after determining the optimal number of modes K for the quadratic VMD decomposition and its corresponding maximum kurtosis mode component, wavefront localization and identification are performed on the mode component.

[0041] For example, an NTEO energy operator is proposed, which introduces a resolution parameter to enhance the detection capability of signal singularities: (9) In the formula, The optimal modal component (the modal component with the maximum kurtosis value) is the one corresponding to the second VMD decomposition. The energy operator corresponding to the optimal modal component at time t; For resolution parameters, and , , These are the sampling frequency and the fundamental frequency, for example. The value is 2.

[0042] Calculate the optimal modal components sequentially At each moment, the energy operator exhibits significant amplitude and frequency fluctuations when reaching the wavefront position, at which point the NTEO energy value is at its maximum, corresponding to that moment. This represents the initial wavefront of the partial discharge signal.

[0043] Finally, the location of the partial discharge signal at the distance from the first end point was determined using the two-end traveling wave ranging method. After determining the moment when the wavefront appears, the next step is to use the double-ended traveling wave ranging formula, combined with the actual parameters of the high-voltage cable, to locate the partial discharge position.

[0044] The formula for two-end distance measurement is as follows: (10) In the formula, The location of the partial discharge signal is at a distance from the first end point, i.e., the location of the partial discharge. The distance between the two measuring devices; This represents the waveform propagation speed, for example, a value of 290 meters per microsecond. , These represent the times when the wavefront with the maximum calibration of the energy operator receives the signal at the beginning and end of the wave, respectively.

[0045] The following simulation example verifies the scheme disclosed in this application.

[0046] Using a high-voltage AC cable as the test object, the signals of a single exponential decay oscillation function and a double exponential decay oscillation function were first superimposed, with parameters A1 and A2 being 8 mV and 3 mV respectively, and the attenuation coefficients being... , The oscillation frequencies are 0.4 and 0.7 respectively. , Both are 3MHz. The simulation frequency is set to 1MHz, and the simulation time is 2 milliseconds, so there are 2000 sampling points T0. The two PD functions are added at the 400th and 1400th time points, respectively. Sine functions with different amplitudes, frequencies, and initial phases are superimposed to form periodic narrowband interference, and 15dB of white noise interference is added at the same time.

[0047] First, VMD is used to decompose the signal. The number of decomposed modes K is calculated sequentially from 2 to 20, and the corresponding adaptive scheduling thresholds are as follows: Table 1. Kurtosis thresholds for different decomposition modes

[0048] For each decomposition mode number K, the kurtosis value of each decomposition mode is calculated. Components with kurtosis values ​​less than the kurtosis threshold are deleted, and the remaining decomposition modes that meet the conditions are then superimposed. The superimposed components are then compared with the original noise-free signal using cosine similarity calculation. The similarity is calculated. The higher the similarity, the more thorough the denoising, and the better the selected decomposition mode number. Finally, it is calculated that the cosine similarity is the largest when K is 9.

[0049] The initially denoised signal still contains white noise and local high-frequency fluctuations. Therefore, wavelet decomposition is used for denoising. The VMD denoising effect based on adaptive kurtosis threshold is as follows: Figure 2 As shown.

[0050] Depend on Figure 2 It can be seen that by selecting the optimal number of decomposition modes using cosine similarity, filtering out noisy modes through adaptive kurtosis thresholding, and then performing superposition and summation, a noisy PD signal is obtained, but the characteristics of the PD signal can already be initially seen at the response time. Further denoising using wavelet decomposition yields the final denoised signal, which shows a high degree of restoration of the PD signal curve and accurate identification of the occurrence time. However, the amplitude after denoising is significantly reduced, from 1mV to around 0.2mV, requiring further improvement.

[0051] For the denoised signal The next step requires repair. First, extract the peak and trough times, and calculate... Compared with the original noise-free PD signal Peak-valley energy difference Using the equidistant trial method, Calculate the minimum peak-valley energy difference using a spacing of 0.01, ranging from 0 to 50. Time corresponding , which is the optimal restoration coefficient for the denoised signal. In this example, the calculation is... The value is 4.17.

[0052] Five assessment levels for partial discharge defects in cables were established: critical defect, severe defect, general defect, potential risk, and continuous monitoring, labeled as type 1 to type 5 respectively. A sample curve library was built, with 50 curves for each defect level, totaling 250 curves, each labeled with defect level 1 to 5. For each curve, nine parameters were calculated: maximum value, average value, root mean square, phase, skewness, kurtosis, crest factor, waveform factor, and peak energy per unit time. A convolutional neural network (CNN) was used for training and computation, with the nine parameters of each waveform as input and the corresponding defect level as output. After CNN training was completed, one curve to be detected was input. If the assessment level of the signal is classified as a general defect, a serious defect, or a critical defect, then the next step of defect location calculation must be carried out immediately.

[0053] For the signal curves that need to be located mentioned above A secondary VMD decomposition is performed because this decomposition needs to have significant discharge characteristics, which is more helpful for accurate defect localization. Therefore, K values ​​are calculated from 2 to 20, and the maximum kurtosis of each decomposition mode is calculated for each K value. Then, the maximum kurtosis value among all K values ​​is calculated, and the decomposition mode corresponding to this maximum kurtosis value is... The maximum kurtosis values ​​for different decomposition modes are as follows: Figure 3 As shown.

[0054] Depend on Figure 3 It can be seen that the maximum kurtosis value after the quadratic VMD decomposition occurs at K=12. Therefore, the final selection of the quadratic decomposition mode number is 12. The NTEO energy operator curve corresponding to the maximum kurtosis value at K=12 is then calculated as follows: Figure 4 As shown.

[0055] according to Figure 4 It can be concluded that the energy operator value is at its maximum at 422 microseconds, which is the wavefront time, ts=422 microseconds. Using the double-ended traveling wave ranging method, the time of the wavefront is measured at the other end as tm=418 microseconds. The total cable length is 5000 meters. The partial discharge location is calculated to be 3080 meters away from the first segment using the ranging formula.

[0056] According to one embodiment of this application, a cable partial discharge locating device for multimodal noise reduction and repair is provided, comprising: The noise reduction module performs a first VMD decomposition to remove the noise decomposition modes from the acquired noisy PD signal of the high-voltage AC cable, and then uses wavelet decomposition for secondary noise reduction. The repair module uses the peak-valley energy difference per unit time to repair the denoised signal waveform; The evaluation module extracts the computational features of the repaired and denoised signal and inputs them into the CNN model to evaluate the level of partial discharge defects. The positioning module performs a second VMD decomposition on the repaired denoised signal based on the partial discharge defect level assessment results, and identifies the wavefront of the mode component with the maximum kurtosis value to achieve partial discharge positioning.

[0057] Furthermore, it also includes a parameter optimization module, used to determine the optimized parameters of the above modules based on the sample signals, such as the optimal number of decomposed modes Kp, kurtosis threshold, and optimal repair coefficient. The sample signals here can be simulated signals or actual cable PD signals processed using existing methods. These optimized parameters are stored in a storage module and continuously updated for use by the modules during execution.

[0058] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application and not to limit them; although this application has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications can still be made to the specific implementation of this application or equivalent substitutions can be made to some technical features without departing from the spirit of the technical solutions of this application, and all such modifications and substitutions should be covered within the scope of the technical solutions claimed in this application.

Claims

1. A method for locating partial discharge in cables using multimodal noise reduction and repair, characterized in that, include: The noise-containing PD signal of the high-voltage AC cable is subjected to the first VMD decomposition to remove the noise decomposition mode, and then the wavelet decomposition method is used for secondary denoising. The peak-valley energy difference per unit time is used to repair the denoised signal waveform; The computational features of the denoised signal after repair are extracted and input into a CNN model to evaluate the level of partial discharge defects. Based on the partial discharge defect level assessment results, the repaired denoised signal is subjected to a second VMD decomposition, and the mode component with the maximum kurtosis value is identified by wavefront localization to achieve partial discharge localization.

2. The cable partial discharge location method according to claim 1, characterized in that, The optimal number of decomposition modes Kp for the first VMD decomposition is determined using the following method: Using a preset fixed kurtosis threshold, for each decomposition mode number K within a predetermined numerical range, decomposition modes of the sample PD signal containing noise that are less than the fixed kurtosis threshold are deleted. The remaining decomposition modes are then summed, and the cosine similarity between the summation result and the noise-free sample PD signal is calculated. The K value with the highest similarity is selected as the optimal decomposition mode number Kp.

3. The cable partial discharge location method according to claim 2, characterized in that, The kurtosis threshold of the noise removal decomposition mode Determined based on the decomposition mode number K, and expressed as: In the formula, , These are the upper and lower limits of the kurtosis threshold, respectively. K is the empirical adjustment coefficient. max This represents the maximum value within the range of decomposition mode number K.

4. The cable partial discharge location method according to claim 2, characterized in that, The optimal restoration coefficient for restoring the denoised signal waveform is determined by the following method: extracting the peak and trough times of the sample PD signal after secondary denoising, calculating the peak-trough energy difference between the sample PD signal after secondary denoising and the sample PD signal without noise, and using the equidistant trial method to take the restoration coefficient from 0 to a predetermined value at predetermined intervals, and taking the restoration coefficient corresponding to the minimum peak-trough energy difference as the optimal restoration coefficient.

5. The cable partial discharge location method according to claim 4, characterized in that, For the same type of cable, multiple calculations are performed, and the average of the optimal repair coefficients obtained from each calculation is taken as the final empirical repair coefficient.

6. The cable partial discharge location method according to claim 1, characterized in that, The calculated characteristic quantities include maximum value, average value, root mean square, phase, skewness, kurtosis, crest factor, waveform factor, and peak energy per unit time.

7. The cable partial discharge location method according to claim 1, characterized in that, The optimal number of decomposition modes Kp for the second VMD decomposition is determined using the following method: With the goal of obtaining the maximum kurtosis value from the selected decomposition mode number K, the kurtosis value of each decomposition mode under each decomposition mode number K within a predetermined numerical range is calculated, and the mode decomposition mode number K corresponding to the maximum kurtosis value is selected as the optimal decomposition mode number Kp.

8. The cable partial discharge location method according to claim 1, characterized in that, After determining the moment when the wavefront appears, the partial discharge location is located using dual-end traveling wave ranging.

9. The cable partial discharge location method according to any one of claims 1-8, characterized in that, The cable is a high-voltage photovoltaic AC cable.

10. A cable partial discharge locating device for multimodal noise reduction and repair, characterized in that, include: The noise reduction module performs a first VMD decomposition to remove the noise decomposition modes from the acquired noisy PD signal of the high-voltage AC cable, and then uses wavelet decomposition for secondary noise reduction. The repair module uses the peak-valley energy difference per unit time to repair the denoised signal waveform; The evaluation module extracts the computational features of the repaired and denoised signal and inputs them into the CNN model to evaluate the level of partial discharge defects. The positioning module performs a second VMD decomposition on the repaired denoised signal based on the partial discharge defect level assessment results, and identifies the wavefront of the mode component with the maximum kurtosis value to achieve partial discharge positioning.

Images