CN113726318A

CN113726318A - WM-based partial discharge white noise self-adaptive suppression method

Info

Publication number: CN113726318A
Application number: CN202110770537.6A
Authority: CN
Inventors: 孟鹏飞; 周凯; 龚薇; 李原; 朱光亚; 曹晓燕; 冉立
Original assignee: Sichuan University
Current assignee: Sichuan University
Priority date: 2021-07-08
Filing date: 2021-07-08
Publication date: 2021-11-30

Abstract

The invention discloses a WM-based self-adaptive suppression method for white partial discharge noise, which comprises the steps of obtaining an original signal generated in a partial discharge test process; decoupling and separating relatively high-frequency components and relatively low-frequency components in the original signal; correcting the relative high-frequency component to obtain a real relative high-frequency fluctuation amount; and taking an absolute value of the relative low-frequency component, and scanning by adopting a sliding threshold window to obtain the PD signal after white noise suppression. Realizing the separation of aliasing signals, and further denoising the separated PD signals through a self-adaptive threshold window, thereby realizing the suppression of white noise in the noisy signals; the method can effectively inhibit white noise in the actually measured signal, has small distortion of the pulse starting point, and is obviously superior to the signal denoising effect of the conventional DWT and ASVD methods.

Description

WM-based partial discharge white noise self-adaptive suppression method

Technical Field

The invention relates to the field of power cable detection, in particular to a WM-based partial discharge white noise self-adaptive suppression method.

Background

Partial Discharge (PD) detection is an effective measure for diagnosing the insulation state of a power cable, and is also one of the main means for ensuring the normal operation of the cable. However, in-situ cable partial discharge detection is often in a complex environment, and the detected partial discharge signal is often mixed with various interference signals, such as white noise interference, high-frequency pulse interference, and periodic narrow-band interference, so as to affect the partial discharge detection result. White noise is the dominant interference in PD detection. Therefore, how to effectively suppress white noise in the partial discharge signal is a key problem of improving the accuracy of the partial discharge detection.

At present, a plurality of scholars at home and abroad research and make great contribution to the problem. Discrete Wavelet Transform (DWT) has been widely used for suppressing white noise in PD signals. In the prior art, white noise in a local-playing signal is effectively suppressed based on DWT. The wavelet transform is utilized to obtain good effect in periodic narrow-band interference suppression. However, actually measured PD signals are complex and diverse, the difficulty in selecting a wavelet basis function of a suitable PD pulse waveform is increased, and the selection of the decomposition scale and the threshold of the wavelet transform is not unique. With respect to the problem of selecting a Decomposition scale and a threshold in wavelet Decomposition, researchers have proposed an Empirical Mode Decomposition (EMD) noise suppression method. By comparing the denoising effects of DWT and EMD, the method discovers that EMD can effectively suppress noise in oscillation attenuation PD signals, but the denoising effect of non-oscillation pulses is poor, and the EMD algorithm is poor in stability, and has the problems of endpoint effect, mode aliasing and the like. Singular Value Decomposition (SVD) is also a commonly used method for suppressing partial discharge noise at present. The SVD method reconstructs signals by selecting proper singular value threshold values to realize noise suppression, but the singular value threshold values are not properly selected and are easy to cause signal distortion. White noise in the PD signal is suppressed by SVD technology and a good effect is obtained. However, for the actually measured signals, the number of SVD decomposition layers and the effective number of singular values are difficult to select, and are easily affected by human factors, and the SVD needs to perform a large amount of matrix decomposition calculation, so that the algorithm consumes a long time. Therefore, an Adaptive Singular Value Decomposition (ASVD) is proposed to suppress white noise in a noisy PD signal and effectively retain signal detail information, but the Singular Value threshold remains difficult to determine and the pulse end point position is likely to be distorted.

Disclosure of Invention

The invention aims to provide a WM-based self-adaptive suppression method for partial discharge white noise. Secondly, white noise suppression is carried out on the MATLAB simulation signal and the actually measured PD signal by utilizing the method, the DWT method and the ASVD method, and the denoising effects of the MATLAB simulation signal and the actually measured PD signal are compared, and the result shows that the effect of suppressing the white noise in the PD signal by the method is superior to that of the two.

The patent provides a WM-based partial discharge white noise self-adaptive suppression method, which comprises the following steps:

s1, acquiring an original signal generated in the partial discharge test process;

s2, decoupling and separating a relatively high-frequency component and a relatively low-frequency component in the original signal;

s3, correcting the relative high-frequency component to obtain a real relative high-frequency fluctuation amount;

and S4, taking an absolute value of the relative low-frequency component, and scanning by adopting a sliding threshold window to obtain a PD signal after white noise suppression.

Further preferably, in S2, decoupling and separating the original signal to obtain the high frequency components includes the following steps: sequentially solving local variation and accumulated variation for the original data D; taking a data starting point as a first characteristic point, and sequentially judging whether the next data is effective fluctuation or not, wherein the accumulated variation is 0; recording feature points of the effective fluctuation; amplitude H corresponding to all the characteristic points of the effective fluctuation_AInterpolation is carried out to obtain S_HF。

Further preferably, when sequentially judging whether the next data is valid fluctuation, the following method is adopted:

local variance andaccumulate the product of the change, Δ d × c>At 0, the point is not recorded and the calculation of Δ d and c continues for the next point. When Δ d × c is less than or equal to 0, the accumulated variation c and the fluctuation H of the previous feature point are further determined_V(k) (k is 1,2,3 …), if the two are in the same direction, updating the original feature point to the current data point, and setting c to zero; if the two are reversed, it is determined whether the absolute value of the current c is larger than the fluctuation threshold Th₁. If greater than Th₁Then, marking the current data point as a new characteristic point, and setting c to zero; when less than Th₁Then the data point is not recorded and the accumulation c continues for the next point.

Further preferably, the original signal is decoupled and separated, and when the low-frequency component is relatively low, the following method is adopted:

the extraction is S_HFThe curve formed by the fluctuation centers is used as the original data of the relatively low-frequency components;

judging where the original signals are overlapped;

sequentially solving local variation and accumulated variation at the overlapped part; taking the initial point of the data as a first characteristic point, and sequentially judging whether the next data is effective fluctuation or not, wherein the accumulated variation is 0; recording the characteristic points T of the effective fluctuation_L，

Obtaining the characteristic point T of the relative low-frequency component_LAnd corresponding amplitude L_ABy interpolation to obtain S_LF

Further preferably, the following formula is used to determine where the original signals overlap:

H_V(k)＝H_A(k)-H_A(k-1)

wherein H_V(k) Representing the amount of fluctuation of the kth effective fluctuation, H_A(k) Amplitude corresponding to the kth feature point, where Th₂Usually 0.2. T is_CIs S_HFCenter of fluctuation of A_CIs S_HFThe amplitude of (c).

Further preferably, the correcting the relatively high frequency component includes correcting a signal overlapping portion according to the following formula:

wherein H_D(k) The corrected characteristic point amplitude value is obtained; h_A(k) The amplitude corresponding to the kth characteristic point; l is_A(k) The corresponding amplitudes of the feature points for the low frequency components.

Further preferably, the correcting the relatively high frequency component includes correcting a non-overlapping portion of the signals according to the following formula:

H_D(k)＝H_A(k)-L_A(k)

Further preferably, a sliding threshold window is adopted, and during scanning, the method comprises the step of adaptively selecting a fluctuation threshold and an amplitude threshold by adopting an OSTU algorithm.

Further preferably, the method also comprises simulating two typical noisy PD signals by using MATLAB, and performing simulation on the partial discharge signal by using two models of single-exponential damped oscillation and double-exponential damped oscillation; and selecting a 35kV XLPE cable with a terminal longitudinal tool mark defect to build a test platform for PD test.

Compared with the prior art, the invention has the beneficial effects that:

1. according to the suppression method provided by the invention, the separation of aliasing signals is realized by utilizing a fluctuation method, and then the separated PD signals are further de-noised through a self-adaptive threshold window, so that the suppression of white noise in the noise-containing signals is realized; the method can effectively inhibit white noise in the actually measured signal, has small distortion of the pulse starting point, and is obviously superior to the signal denoising effect of the conventional DWT and ASVD methods.

2. The suppression method provided by the invention utilizes the OSTU algorithm to select the optimal threshold, and has better self-adaptability compared with DWT and ASVD methods.

3. According to the suppression method provided by the invention, under different signal-to-noise ratios, the denoised waveform has higher similarity, the mean square error is smaller, and the algorithm consumes shorter time.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

FIG. 1 is a flow chart of the present invention for extracting a signal by a wave method;

FIG. 2 is a flow chart of denoising according to the present invention;

FIG. 3(a) is a diagram of a type I PD pulse waveform in simulation of the present invention;

FIG. 3(b) is a white noise waveform diagram in simulation according to the present invention;

FIG. 3(c) is a waveform diagram of a mixed type I PD pulse and white noise in simulation according to the present invention;

FIG. 4(a) is a diagram of type II PD pulse waveforms in simulation of the present invention;

FIG. 4(b) is a white noise waveform diagram in simulation according to the present invention;

FIG. 4(c) is a waveform diagram of a mixed type II PD pulse and white noise in simulation according to the present invention;

FIG. 5(a) is a waveform diagram of a noisy signal in a denoising process of a type I simulation signal in simulation according to the present invention;

FIG. 5(b) is a graph of the denoising effect of db8 for comparison in the type I simulation signal denoising process;

FIG. 5(c) is an ASVD denoising effect for comparison in the type I simulation signal denoising process;

FIG. 5(d) is a graph of the denoising effect of the I-type simulation signal in the simulation of the present invention;

FIG. 6(a) is a waveform diagram of a noisy signal in a type II simulation signal denoising process in simulation of the present invention;

FIG. 6(b) is a graph of the denoising effect of db8 for comparison in the type II simulation signal denoising process;

FIG. 6(c) is an ASVD denoising effect for comparison in the type II simulation signal denoising process;

FIG. 6(d) is a graph of the denoising effect of type II simulation signals in simulation of the present invention;

FIG. 7(a) is a denoising performance evaluation diagram under signal-to-noise ratio in simulation of the present invention;

FIG. 7(b) is a diagram of denoising performance evaluation under another signal-to-noise ratio in simulation according to the present invention;

FIG. 8 is a schematic diagram of a partial discharge detection platform according to the present invention

FIG. 9(a) is a graph of PD signals measured according to the test platform shown in FIG. 8;

FIG. 9(b) is a diagram illustrating a PD signal actually measured after noise is superimposed according to the detection platform shown in FIG. 8;

FIG. 9(c) illustrates de-noising according to the inspection platform db8 shown in FIG. 8;

FIG. 9(d) is an ASVD de-noising according to the inspection platform shown in FIG. 8;

FIG. 9(e) is a diagram illustrating the de-noising effect of the detection platform shown in FIG. 8;

FIG. 10 measured signal denoising algorithm time-consuming comparison

Detailed Description

The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the drawings in the embodiments of the present invention, and it is obvious that the described embodiments are only a part of the embodiments of the present invention, and not all of the embodiments. All other embodiments, which can be obtained by a person skilled in the art without any inventive step based on the embodiments of the present invention, are within the scope of the present invention.

In the present invention, unless otherwise expressly stated or limited, the terms "mounted," "connected," "secured," and the like are to be construed broadly and can, for example, be fixedly connected, detachably connected, or integrally connected; can be mechanically or electrically connected; they may be connected directly or indirectly through intervening media, or they may be interconnected between two elements. The specific meanings of the above terms in the present invention can be understood by those skilled in the art according to specific situations.

Referring to fig. 1-10, the present invention provides a technical solution: the patent provides a WM-based partial discharge white noise self-adaptive suppression method, which comprises the following steps: s1, acquiring an original signal generated in the partial discharge test process;

The signal amplitude changes due to the superposition of PD pulses and white noise in the partial discharge test. A method of separating an aliasing signal by using the fluctuation of a signal to achieve noise suppression is defined as a fluctuation method. The change of the signal from one extreme point to another adjacent extreme point is called fluctuation, and the positions corresponding to the two extreme points are called characteristic points. The principle of suppressing noise by the fluctuation method is to decouple and separate relatively high-frequency components and relatively low-frequency components in an original signal. The steps of noise suppression of the original data D with length N are as follows:

relatively high frequency component S_HFThe local variation and the accumulated variation of the data at the point i are obtained by the formula (1) and the formula (2).

Δ D (i) ═ D (i +1) -D (i) (formula 1)

Where i is 1,2,3, … N-1, Δ d (i) represents a local variation of i point, which is a difference in data amplitude between i +1 point and i point, and c (i) represents a sum of all local variations of data before i point (excluding the local variation of data at i point).

The fluctuation range of the background noise is small and varies around 0, and in order to extract the effective fluctuation, an appropriate fluctuation amount threshold Th is set₁Only when the fluctuation amount is larger than the threshold value, it can be judged as effective fluctuation. The specific extraction process is shown in FIG. 1.

when the product of the local variation and the cumulative variation, Δ d c>At 0, the point is not recorded and the calculation of Δ d and c continues for the next point. When Δ d × c is less than or equal to 0, the accumulated variation c and the fluctuation H of the previous feature point are further determined_V(k) (k is 1,2,3 …), if the two are in the same direction, updating the original feature point to the current data point, and setting c to zero; if the two are reversed, it is determined whether the absolute value of the current c is larger than the fluctuation threshold Th₁. If greater than Th₁Then, marking the current data point as a new characteristic point, and setting c to zero; when less than Th₁Then the data point is not recorded and the accumulation c continues for the next point.

Decoupling and separating the original signal, and when the low-frequency component is relatively low, adopting the following method:

the extraction is S_HFThe curve formed by the fluctuation centers is used as the original data of the relatively low-frequency components; the raw data from which the relatively low frequency components are extracted can be approximated as S_HFThe centers of the respective fluctuations constitute a curve. And S_HFCenter of fluctuation TP of_CAnd its amplitude A_CMay be represented by equation (3) and equation (4), respectively.

Judging where the original signals are overlapped; when the relative amplitude of the two fluctuation centers is greatly changed, the original signals are overlapped. The specific judgment formula is as (5)

H_V(k)＝H_A(k)-H_A(k-1) (equation 6)

The relatively high frequency component S in the original signal_HFIs subjected to a relatively low frequency component S_LFIs therefore required to be applied to S_HFAnd correcting to obtain the real relative high-frequency fluctuation amount. To S_HFCharacteristic sequence T of_H(k) And H_A(k) (k is 1,2,3, …) the formula (5) is determined, and if it is larger than the threshold Th₂Then the current point is considered to overlap. If less than threshold Th₂Then the current point is considered to be non-overlapping.

Obtaining the characteristic point T of the relative low-frequency component_LAnd corresponding amplitude L_ABy interpolation to obtain S_LF。

H_D(k)＝H_A(k)-L_A(k) (formula 8)

Further preferably, a sliding threshold window is adopted, and during scanning, the method comprises the step of adaptively selecting a fluctuation threshold and an amplitude threshold by adopting an OSTU algorithm. White noise is mainly distributed in a high frequency portion of a noise-containing signal, and a PD signal is mainly distributed in a low frequency portion. So that the relatively low-frequency component S extracted by the wave method is utilized_LFNamely the partial discharge signal after white noise suppression.

Last pair of S_LFAdaptive selection of the amplitude threshold Th₃And using a length of T_wTime window of (1) in T_wStep size pair S of/2_LFScanning is performed sequentially, wherein the time window is selected to be about the duration of the pulse. If the data in the time window is smaller than the threshold value, setting the data in the window to zero; if data greater than the threshold occurs within the time window, the time window continues to be moved until data less than the threshold occurs, and the data for the several time windows is retained. And repeating the scanning process until all data are scanned, and obtaining the PD signal after white noise suppression. In summary, the flow of suppressing white noise in the noise-containing partial discharge signal by the method is shown in fig. 2. In order to effectively suppress white noise, a threshold value of the fluctuation amount and a threshold value of the amplitude are selected as appropriate. The method introduces Otsu algorithm (OSTU) to fluctuation threshold Th₁And a magnitude threshold Th₃And self-adaptive selection is carried out, so that the influence of artificial selection is avoided.

The fluctuation and amplitude variation of the white noise are small and concentrated near the 0 value, while the fluctuation and amplitude variation of the PD signal are large and dispersed, and a certain threshold value T exists to enable the inter-class variance of the fluctuation amount of the PD signal and the fluctuation amount of the white noise signal to be maximum. The threshold T at this time is the optimal division threshold of the effective fluctuation and the ineffective fluctuation.

The discrete sequence of known length N has x_i(i ═ 1,2,3 … N) the maximum and minimum values are each x_max、x_minThen there is d_x＝(x_max-x_min) and/M, wherein M is a gray scale. If the discrete data falls within [ (m-1) d_x,md_x]The number of the intervals is n_mWhere M (M is 1,2,3 … M) is called the gray scale value, n is_mThe total number of pixels for the entire gray scale is equal to the sequence length N, i.e. N is equal to N, the sum of the number of pixels for each gray scale₁+n₂+…+n_m. So that the probability that the gray value m appears in the number of pixels is P_m＝n_m/N。

Assume that the optimal threshold T ═ k × d_xDividing discrete sequences into two classes of sequences C₁And C₂In which C is₁Indicates that the falling within the interval [0, k + d ]_x]The number of discrete data of (2); c₂The expression falls within the interval [ (k +1) × d_x,k*d_x]The number of discrete data of (2). The probabilities ω and the mean u of the two classes are obtained by equations (9-10) and (11-13), respectively:

then C is₁And C₂Is between classes of²Is composed of

From the above formula, σ²Is a function of k, there is an optimum threshold T ═ k ═ d_xThe variance between the two classes is maximized. That is to say that the first and second electrodes,

in order to quantitatively evaluate the denoising effect, two parameters, namely a Normalized Correlation Coefficient (NCC) and a Mean Square Error (MSE), are introduced to evaluate the denoising waveform. If for the noise-free PD signal x with the length of N₁(i) (i ═ 1,2,3, …, N) and denoised PD signal x₂(i) For the evaluation of the denoising effect, NCC and MSE can be determined by equations (16) and (17), respectively.

The NCC represents the waveform similarity of signals before and after denoising, and the closer the value is to 1, the better the denoising effect is; and MSE represents the mean square error of the signals before and after denoising, and the smaller the value of the MSE is, the better the denoising effect is.

The volume mathematical model is shown in the formula (18) and the formula (19), respectively.

Wherein A is₁And A₂Representing the amplitude, fc, of the partial discharge signal₁And fc₂Representing the frequency of oscillation, τ₁And τ₂Is the attenuation coefficient. Two kinds of partial discharge signals generated by MATLAB according to the formula (18) and the formula (19) are shown in fig. 3(b) and fig. 4(b), respectively, and the specific parameter settings thereof are shown in table 1.

TABLE 1 PD two model parameter settings

Wherein the sampling time is about 10us and the sampling frequency is 200 MSa/s. Gaussian white noise shown in fig. 3(b) and fig. 4(b) is added to the two PD pulses, respectively, to obtain two noisy PD signals with a signal-to-noise ratio of 2, and waveforms thereof are shown in fig. 3(c) and fig. 4(c), respectively.

Firstly, the optimal fluctuation threshold Th of two noisy signals is solved by using the OSTU algorithm₁. Second, db8(4 layers), ASVD (artificially preserving the first 5 singular values), and text method (T)_W180) to perform noise suppression on the two noisy PD simulation signals, and the denoising results are shown in fig. 5 and fig. 6. And finally, quantitatively evaluating the denoising effect by using the NCC and the MSE of the three. The de-noising evaluation of type I noisy PD signals is shown in Table 2, and the de-noising evaluation of type II noisy PD signals is shown in Table 3

From the denoising results shown in fig. 5 and fig. 6, the waveform denoised by db8 retains a certain residual noise, the waveform denoised by ASVD is prone to generate the distortion of the pulse end point, and the residual noise of the denoised waveform and the distortion of the pulse end point can be reduced by the method.

TABLE 2 type I simulation signal denoising assessment

TABLE 3 type II simulated signal denoising assessment

As can be seen from tables 2 and 3, for the two simulation signals, the waveform of the denoised signal by the method is closer to the PD signal without noise, and the mean square error of the denoised signal is smaller, so the white noise suppression effect of the method is better than that of the DWT method and the ASVD method.

In order to study the denoising effect of the method under different signal-to-noise ratios, white noise with different degrees is added into the simulated PD signal shown in FIG. 3(a) to make the signal-to-noise ratio thereof be-4, -2, 0dB, and dB8(4 layers), ASVD (artificially preserving the first 5 singular values) and the text method (T)_W180) are respectively denoised, resulting in a denoised estimate as shown in fig. 7.

As can be seen from FIG. 7, under different signal-to-noise ratios, the waveform similarity of the denoised method is higher, and the mean square error of the denoised method is smaller, which indicates that the denoising effect of the method is better than that of DWT and ASVD methods for different signal-to-noise ratios.

In order to verify the effectiveness of the method for suppressing white noise in the actually measured partial discharge signal, a power frequency partial discharge test platform shown in fig. 8 is set up in a laboratory, and a PD test is performed on a 35kV XLPE cable with terminal longitudinal tool mark defects (length 100mm, width 1mm, height 1 mm). Wherein, the bandwidth of a High Frequency Current Transducer (HFCT) -6dB is 2.5-216 MHz, and the maximum sensitivity is 5.83 mV/Ma. The oscilloscope is of a type Rigol DS6104, the maximum sampling rate is 5GSa/s, and the bandwidth is up to 1 GHz. The partial discharge signal obtained by the test is shown in FIG. 9(a), but measuredTo verify the white noise suppression effect of this method, gaussian white noise was added to the actual measurement signal to make the snr 2, and a signal as shown in fig. 9(b) was obtained. Wherein the sampling rate is 200 MSa/s. Using db8(4 layers), ASVD (take the first 3 singular values), and the method herein (T)_w300) the measured signal is denoised, and the denoising effect is shown in fig. 9 (c-e).

As shown in fig. 9(c), residual noise still exists in the waveform after wavelet denoising; as shown in fig. 9(d), after ASVD denoising, the starting point of the pulse is distorted; as can be seen from fig. 9(e), the method can effectively suppress white noise in the measured signal, and the distortion of the pulse starting point is small.

TABLE 4 De-noising evaluation of measured partial discharge signals

As can be seen from the analysis of table 4 and fig. 10, for the measured signal, the method also has higher waveform similarity and smaller mean square error compared with DWT and ASVD methods, and the algorithm is less time-consuming. Therefore, the white noise suppression effect of the method on the actually measured signal is superior to that of the DWT method and the ASVD method. And the method is more advantageous when processing larger data volumes.

In the description of the present invention, it is to be understood that the indicated orientations or positional relationships are based on the orientations or positional relationships shown in the drawings and are only for convenience in describing the present invention and simplifying the description, but are not intended to indicate or imply that the indicated devices or elements must have a particular orientation, be constructed and operated in a particular orientation, and thus should not be construed as limiting the present invention.

In the description herein, references to the description of the term "one embodiment," "some embodiments," "an example," "a specific example," or "some examples," etc., mean that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the invention. In this specification, the schematic representations of the terms used above do not necessarily refer to the same embodiment or example. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that changes, modifications, substitutions and alterations can be made in these embodiments without departing from the principles and spirit of the invention, the scope of which is defined in the appended claims and their equivalents.

Claims

1. A partial discharge white noise self-adaptive suppression method based on WM is characterized by comprising the following steps:

2. The WM-based adaptive suppression method for partial discharge white noise according to claim 1, wherein in S2, decoupling separation of the original signal with respect to high frequency components is performed, comprising the steps of:

sequentially solving local variation and accumulated variation for the original data D;

taking a data starting point as a first characteristic point, accumulating the variation quantity to be 0, and sequentially judging whether the next data is effective fluctuation or not; the characteristic points of the effective undulations are recorded,

amplitude H corresponding to all the characteristic points of the effective fluctuation_AInterpolation is carried out to obtain S_HF。

3. The WM-based adaptive suppression method for white partial discharge noise according to claim 2, wherein the following method is used when sequentially determining whether the next data is valid fluctuation:

when the product of the local variation and the cumulative variation, Δ d c>At 0, the point is not recorded and the calculation of Δ d and c continues for the next point. When Δ d × c is less than or equal to 0, the accumulated variation c and the fluctuation H of the previous feature point are further determined_V(k) (k is 1,2,3 …), if the two are in the same direction, updating the original feature point to the current data point, and setting c to zero; if the two are opposite, judging whether the absolute value of the current c is larger than the fluctuation threshold Th₁. If greater than Th₁Then, marking the current data point as a new characteristic point, and setting c to zero; when less than Th₁Then the data point is not recorded and the accumulation c continues for the next point.

4. The WM-based adaptive suppression method for partial discharge white noise according to claim 2, characterized in that the original signal is decoupled and separated, and when the relative low frequency component is determined, the following method is adopted:

judging where the original signals are overlapped;

sequentially solving local variation and accumulated variation at the overlapped part; taking a data starting point as a first characteristic point, accumulating the variation quantity to be 0, and sequentially judging whether the next data is effective fluctuation or not; recording the characteristic points T of the effective fluctuation_L，

5. The WM-based adaptive suppression method for local discharge white noise according to claim 4, wherein the following formula is used to determine where the original signal overlaps:

H_V(k)＝H_A(k)-H_A(k-1)

6. The WM-based adaptive suppression method for local discharge white noise according to claim 4, wherein the modifying the relatively high frequency component comprises modifying the signal overlap according to the following formula:

7. The WM-based adaptive suppression method for local discharge white noise according to claim 4, wherein the modifying the relatively high frequency component comprises modifying the non-overlap of signals according to the following formula:

H_D(k)＝H_A(k)-L_A(k)

8. The WM-based adaptive suppression method for white partial discharge noise according to claim 1, using a sliding threshold window for scanning, comprising using an OSTU algorithm to adaptively select the fluctuation amount threshold and the amplitude threshold.

9. The adaptive suppression method for partial discharge white noise based on WM of claim 1 further comprising simulating two typical noisy PD signals with MATLAB and simulating the partial discharge signal with two models of single exponential ringing and double exponential ringing.