CN116522080A - Partial discharge signal noise reduction method - Google Patents
Partial discharge signal noise reduction method Download PDFInfo
- Publication number
- CN116522080A CN116522080A CN202310489104.2A CN202310489104A CN116522080A CN 116522080 A CN116522080 A CN 116522080A CN 202310489104 A CN202310489104 A CN 202310489104A CN 116522080 A CN116522080 A CN 116522080A
- Authority
- CN
- China
- Prior art keywords
- signal
- partial discharge
- signals
- noise
- frequency
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
- 238000000034 method Methods 0.000 title claims abstract description 67
- 230000009467 reduction Effects 0.000 title claims abstract description 35
- 238000012360 testing method Methods 0.000 claims abstract description 47
- 230000009466 transformation Effects 0.000 claims abstract description 22
- 238000000354 decomposition reaction Methods 0.000 claims abstract description 17
- 230000000694 effects Effects 0.000 claims abstract description 16
- 230000001629 suppression Effects 0.000 claims abstract description 9
- 239000011159 matrix material Substances 0.000 claims description 31
- 239000013598 vector Substances 0.000 claims description 16
- 238000001914 filtration Methods 0.000 claims description 12
- 230000006870 function Effects 0.000 claims description 11
- 238000009826 distribution Methods 0.000 claims description 9
- 238000002474 experimental method Methods 0.000 claims description 9
- 230000008569 process Effects 0.000 claims description 9
- 238000013016 damping Methods 0.000 claims description 8
- 230000010355 oscillation Effects 0.000 claims description 8
- 230000000737 periodic effect Effects 0.000 claims description 8
- 238000007781 pre-processing Methods 0.000 claims description 8
- 238000012545 processing Methods 0.000 claims description 8
- 230000007246 mechanism Effects 0.000 claims description 7
- 238000001228 spectrum Methods 0.000 claims description 7
- 238000006243 chemical reaction Methods 0.000 claims description 6
- 238000005070 sampling Methods 0.000 claims description 6
- 238000012216 screening Methods 0.000 claims description 6
- 238000010183 spectrum analysis Methods 0.000 claims description 6
- 239000004973 liquid crystal related substance Substances 0.000 claims description 4
- 238000004088 simulation Methods 0.000 claims description 4
- 230000002441 reversible effect Effects 0.000 claims description 3
- 238000012549 training Methods 0.000 claims description 3
- 230000001131 transforming effect Effects 0.000 claims description 3
- 238000000605 extraction Methods 0.000 abstract description 7
- 238000001514 detection method Methods 0.000 abstract description 2
- 238000003745 diagnosis Methods 0.000 abstract 2
- 208000028659 discharge Diseases 0.000 description 52
- 238000010586 diagram Methods 0.000 description 9
- 230000008859 change Effects 0.000 description 4
- 230000007547 defect Effects 0.000 description 2
- 230000005684 electric field Effects 0.000 description 2
- 238000009413 insulation Methods 0.000 description 2
- 238000005259 measurement Methods 0.000 description 2
- 238000005457 optimization Methods 0.000 description 2
- 238000011160 research Methods 0.000 description 2
- 230000009471 action Effects 0.000 description 1
- 230000009286 beneficial effect Effects 0.000 description 1
- 230000008033 biological extinction Effects 0.000 description 1
- 230000015556 catabolic process Effects 0.000 description 1
- 239000000470 constituent Substances 0.000 description 1
- 238000012217 deletion Methods 0.000 description 1
- 230000037430 deletion Effects 0.000 description 1
- 230000006872 improvement Effects 0.000 description 1
- 238000012417 linear regression Methods 0.000 description 1
- 238000010801 machine learning Methods 0.000 description 1
- 230000014759 maintenance of location Effects 0.000 description 1
- 238000004519 manufacturing process Methods 0.000 description 1
- 239000000463 material Substances 0.000 description 1
- 238000012544 monitoring process Methods 0.000 description 1
- 238000003058 natural language processing Methods 0.000 description 1
- 230000000149 penetrating effect Effects 0.000 description 1
- 238000013139 quantization Methods 0.000 description 1
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/10—Pre-processing; Data cleansing
-
- G—PHYSICS
- G01—MEASURING; TESTING
- G01R—MEASURING ELECTRIC VARIABLES; MEASURING MAGNETIC VARIABLES
- G01R31/00—Arrangements for testing electric properties; Arrangements for locating electric faults; Arrangements for electrical testing characterised by what is being tested not provided for elsewhere
- G01R31/12—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing
- G01R31/1227—Testing dielectric strength or breakdown voltage ; Testing or monitoring effectiveness or level of insulation, e.g. of a cable or of an apparatus, for example using partial discharge measurements; Electrostatic testing of components, parts or materials
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/14—Fourier, Walsh or analogous domain transformations, e.g. Laplace, Hilbert, Karhunen-Loeve, transforms
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F17/00—Digital computing or data processing equipment or methods, specially adapted for specific functions
- G06F17/10—Complex mathematical operations
- G06F17/16—Matrix or vector computation, e.g. matrix-matrix or matrix-vector multiplication, matrix factorization
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F18/00—Pattern recognition
- G06F18/20—Analysing
- G06F18/21—Design or setup of recognition systems or techniques; Extraction of features in feature space; Blind source separation
- G06F18/213—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods
- G06F18/2131—Feature extraction, e.g. by transforming the feature space; Summarisation; Mappings, e.g. subspace methods based on a transform domain processing, e.g. wavelet transform
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/02—Preprocessing
- G06F2218/04—Denoising
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F2218/00—Aspects of pattern recognition specially adapted for signal processing
- G06F2218/08—Feature extraction
Abstract
The invention discloses a partial discharge signal noise reduction method, which comprises the following steps: building a partial discharge signal detection platform and a partial signal collection platform; and measuring and detecting characteristic signals in the partial discharge signals. The tested signals are screened and classified into characteristic signals through Gaussian test, the signals are processed and noise reduced through a rapid S transformation algorithm and a Singular Value Decomposition (SVD) algorithm, the noise reduction effect is tested and diagnosed by utilizing two coefficients of a noise suppression ratio and an amplitude attenuation ratio, and the diagnosis analysis result is displayed on a platform display interface. The invention discloses a partial discharge signal diagnosis system for rapid S transformation and self-adaptive singular values, which solves the problems that the denoising effect and the extraction speed are difficult to be compatible in the characteristic extraction of partial discharge signals and the characteristic extraction is difficult in the prior art method.
Description
Technical Field
The invention relates to the technical field of noise reduction of electric signals, in particular to a partial discharge signal noise reduction method.
Background
Partial discharge (Partial Discharge, PD) refers to a discharge form occurring between electrodes but not penetrating the electrodes, and has weak points inside the insulation of the device or defects caused in the production process, and repeated breakdown and extinction occur under the action of high electric field intensity.
Common PD signal denoising methods are: the denoising effect of the wavelet denoising method is improved compared with that of the traditional denoising algorithm, but the wavelet transformation is difficult to select proper wavelet basis functions and decomposition layer numbers; the Empirical Mode Decomposition (EMD) is used for denoising the partial discharge signal, and the empirical mode decomposition denoising algorithm has certain self-adaptability, but the algorithm itself can add new white noise to the processed signal, so that the white noise is not thoroughly filtered, and the mode aliasing phenomenon exists.
Disclosure of Invention
The present invention aims to solve at least one of the technical problems in the related art to some extent. Therefore, an object of the present invention is to provide a partial discharge signal noise reduction method capable of improving the partial discharge noise reduction processing effect and improving the denoising time.
The invention provides a partial discharge signal noise reduction method, which comprises the following steps:
s1: building a partial discharge signal noise reduction experiment platform;
s2: by adopting a double-exponential oscillation damping model and a single-exponential oscillation damping model, the simulation of the partial discharge signals is completed, a plurality of groups of partial discharge noise-containing signals are set as experimental groups, and each group of partial discharge noise-containing signals are different;
s3: each group of experiment groups is provided with a control group according to a preset number for training, a kurtosis test is determined, a confidence interval of a Gaussian test is set, gaussian test pretreatment is perfected, and noise-containing partial discharge signals of each group of experiment groups are subjected to Gaussian test pretreatment to obtain a frequency range where characteristic signals are located;
s4: the signals after pretreatment are subjected to rapid S transformation, and a denoising algorithm based on the combination of a time-frequency spectrum analysis filtering method and a threshold denoising method is adopted to remove most of periodic narrowband interference signals in the signals;
s5: the signal after the rapid S transformation is subjected to a fitting threshold value approximation singular value denoising algorithm, a signal two-dimensional mode time-frequency matrix is analyzed, and Gaussian white noise in the signal is filtered;
s6: and testing the noise suppression ratio and the amplitude attenuation coefficient of the noise-reduced signal, analyzing the test result, and displaying the test result on a display interface.
Preferably, the method for performing gaussian test pretreatment on the noisy partial discharge signal in step S3 comprises the following steps:
knowing the random variable X of mean μ and variance σ, gaussian test was performed on the random variable X, and the distribution interval of the random variable X satisfies the probability inequality according to the Chebyshev inequality:
in the above formula: q=1- (σ) 2 /ε 2 ) Epsilon > 0, assuming that a group of observation samples X (t) with the mean value of 0 and the number of N exist in random variables X conforming to Gaussian distribution, when sigma 2 /ε 2 Sufficiently small, the estimate of the random variable X kurtosis value and variance can be expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,is the mean value of kurtosis value, +.>Variance of kurtosis value,/->Kurtosis coefficient +.>The early failure of the reaction system is determined by the number N of observation points and the confidence q, and if the kurtosis value estimated by the variable sequence x (t) is in the interval, the detected signal has obvious Gaussian characteristic and needs to be eliminated.
Preferably, in the preprocessing stage, obtaining the frequency range of the useful PD signal, setting a characteristic frequency retention mechanism and a non-characteristic signal rejection mechanism through an envelope extremum algorithm, and screening out characteristic frequencies through an envelope spectrum so as to independently process the characteristic frequencies;
because the fast S-transform is reversible and linear, a time-frequency filter is designed based on the time-frequency signal after the transform, preserving the time range t E [ t ] 1 ,t 2 ]Frequency range f e f 1 ,f 2 ]While suppressing the time range t e t 3 ,t 4 ]Data f e f of frequency range 3 ,f 4 ]Discretizing the fast S transform, let f=n/NT, τ=jt:
wherein f is frequency, T and τ are time, w (T- τ, f) is a Gaussian window function, λ and p are adjustment factors, λ=0.55, p=0.8, S [ jT, N/NT ] is a fast S transform in discrete case, m, N is a constant, T is a sampling time interval, N is a sampling total point number, and X [ (m+n)/NT ] is a signal X (nT) discrete Fourier transform;
in step S4, the signal after preprocessing is subjected to fast S transformation, and the method of fast S transformation includes the following steps:
s41: screening out useful frequency points k through a Gaussian test according to an envelope spectrum obtained through pretreatment i ，i＝1,2,3,...,n；
S42: fourier transforming the frequency points of the selected frequency band, wherein the window function W' is:
s43: performing inverse Fourier transform to obtain a point k i Fast S-transformation of (a);
s44: step S43 is repeated until the rapid S-transformation of all frequency points is completed.
Preferably, in step S5, the two-dimensional mode time-frequency matrix of the signal is analyzed, and the method for filtering the white gaussian noise in the signal comprises the following steps:
s51: selecting the first n singular values which are not 0 from a singular value matrix lambda obtained by singular value decomposition, and converting the singular values into singular value vectors;
s52: selecting m data from the singular value vector, equally dividing the m data into n/m sections of intervals to be processed, and selecting m to be even according to the property that singular values appear in pairs;
s53: calculating regression coefficients of the processing intervals, determining a point with the maximum second-order difference coefficient, selecting two adjacent intervals of the point, and performing curve fitting;
s54: and determining intersection points of curve fitting, and selecting data slightly smaller than the intersection points from the singular value vectors as a threshold value of a singular value noise reduction algorithm.
Preferably, in step S6, the filtering process is performed for detectionEffect of introducing noise suppression ratio ρ NRR And an amplitude decay ratio ρ ARR Two coefficients are used for comprehensively evaluating the denoising performance:
middle sigma 1 For standard deviation of pre-denoising signal, σ 2 To the standard deviation of the denoised signal A m1 For maximum amplitude of signal before denoising, A m2 Is the maximum amplitude of the denoised signal.
The beneficial effects of the invention are as follows:
(1) The invention provides a method for combining a time-frequency spectrum analysis filtering method and a threshold denoising method based on Gaussian test pretreatment and rapid S transformation, which can effectively identify whether a detected signal has research significance or not, wherein the rapid S transformation is an improvement on the algorithm operation efficiency of generalized S transformation, the frequency range of PD signals is obtained by carrying out pretreatment on the signals, and then the rapid S transformation is carried out according to the frequency range, so that the rapid optimization of the algorithm can be realized, the problem of low feature extraction efficiency when the data volume of partial discharge signals is overlarge is solved, and the method is suitable for noise feature extraction in partial discharge signals of transformers. SVD is a matrix decomposition method for transforming a matrix into its constituent parts, decomposing the vector space of a noisy signal into a "signal subspace" and a "noise subspace", and employing an orthogonal matrix decomposition technique in linear algebra to facilitate matrix computation.
(2) The traditional singular value algorithm determines the threshold value of the singular value by a manual selection or experience prediction method, so that larger errors can be caused, if the threshold value of the singular value is determined by using cubic spline interpolation, the correct singular value threshold value can be selected when the data is less in noise and simpler; when data points are too many and noise is complex, the threshold selection effect is not good, and the problem that the threshold cannot be self-adaptive can be solved by the denoising method for improving singular value decomposition by fitting the threshold.
(3) Because a clean partial discharge signal cannot be measured in the actual test process, the invention introduces two coefficients of the noise suppression ratio and the amplitude attenuation ratio to comprehensively evaluate the denoising performance.
Drawings
In the drawings:
FIG. 1 is a Gaussian test diagram of partial discharge signals according to the present invention;
FIG. 2 is a flow chart of an experimental platform according to the present invention;
FIG. 3 is a partial discharge waveform diagram according to the present invention;
FIG. 4 is a diagram showing the comparison of the fast S-transform and S-transform according to the present invention;
FIG. 5 is a waveform diagram of a time-frequency analysis according to the present invention;
FIG. 6 is a graph of singular values in accordance with the present invention;
FIG. 7 is a simulation of a linear regression function according to the present invention;
FIG. 8 is a graph showing the comparison of denoising effects according to the present invention;
FIG. 9 is a graph of the denoising effect of wavelet transform according to the present invention;
FIG. 10 is a graph showing the effect of empirical mode decomposition on denoising according to the present invention;
FIG. 11 is a graph showing the effect of singular value denoising according to the present invention;
FIG. 12 is a graph showing the denoising effect of the actual measurement signal according to the present invention;
FIG. 13 is a diagram showing the actual measurement results according to the present invention;
FIG. 14 is a diagram showing a real-time monitoring interface of a partial discharge diagnostic system according to the present invention;
FIG. 15 is a diagram showing a historical fault query interface for a partial discharge diagnostic system according to the present invention.
Detailed Description
As shown in fig. 1, a method for reducing noise of partial discharge signals comprises the following steps:
(1) Building a partial discharge signal noise reduction experiment platform;
(2) And the simulation of the partial discharge signal is completed by adopting a double-exponential oscillation damping model and a single-exponential oscillation damping model. Simultaneously, four groups of different partial discharge noise signals are respectively arranged;
(3) Setting four groups of 50 groups of noise signals for training, determining kurtosis test, setting confidence intervals of Gaussian test, perfecting Gaussian test pretreatment, and carrying out Gaussian test pretreatment on four groups of noise-containing partial discharge signals to obtain a frequency range where characteristic signals are located;
(4) The signals after pretreatment are subjected to rapid S transformation, and a denoising algorithm based on the combination of a time-frequency spectrum analysis filtering method and a threshold denoising method is adopted to remove most of periodic narrowband interference signals in the signals;
(5) And (3) the signal after the rapid S transformation is subjected to a fitting threshold value approximation singular value denoising algorithm, a signal two-dimensional mode time-frequency matrix is analyzed, and Gaussian white noise in the signal is filtered.
(6) And testing the noise suppression ratio and the amplitude attenuation coefficient of the noise-reduced signal, analyzing the test result, and displaying the test result on a display interface.
In the step (3), the signal to be processed is subjected to Gaussian test pretreatment, wherein the pretreatment is specifically operated as follows:
the random variables X of the mean μ and variance σ are known, which can be subjected to gaussian tests, from the Chebyshev inequality (Chebyshev inequality), whose distribution interval is known to satisfy the probability inequality:
in the above formula: q=1- (σ) 2 /ε 2 ) Epsilon > 0, assuming that a group of observation samples X (t) with the mean value of 0 and the number of N exist in random variables X conforming to Gaussian distribution, when sigma 2 /ε 2 Sufficiently small, the estimation of its kurtosis value and variance can be expressed as:
is the mean value of kurtosis value, +.>Variance of kurtosis value,/->Kurtosis coefficient +.>The early failure of the reaction system is determined by the number N of observation points and the confidence q, and if the kurtosis value estimated by the variable sequence x (t) is in the interval, the detected signal has obvious Gaussian characteristic and needs to be eliminated.
And (4) performing rapid S conversion on the preprocessed signals, wherein the specific processing flow of the rapid S conversion is as follows:
in the preprocessing stage, the frequency range of the useful PD signal is obtained, a characteristic frequency retaining mechanism and a non-characteristic signal removing mechanism are set through an envelope extremum algorithm, and characteristic frequencies are screened through an envelope spectrum so as to carry out independent processing on the characteristic frequencies.
Because the fast S-transform is reversible and linear, a time-frequency filter is designed based on the time-frequency signal after the transform, preserving the time range t E [ t ] 1 ,t 2 ]Frequency range f e f 1 ,f 2 ]While suppressing the time range t e t 3 ,t 4 ]Data f e f of frequency range 3 ,f 4 ]For quick speedThe S transform is discretized, letting f=n/NT, τ=jt:
wherein f is frequency, T and τ are time, w (T- τ, f) is a Gaussian window function, λ and p are adjustment factors, λ=0.55, p=0.8, S [ jT, N/NT ] is a fast S transform in discrete case, m, N is a constant, T is a sampling time interval, N is a sampling total point number, and X [ (m+n)/NT ] is a signal X (nT) discrete Fourier transform;
the fast S-transform algorithm flow is as follows:
1) Screening out useful frequency points k through a Gaussian test according to an envelope spectrum obtained through pretreatment i ，i＝1,2,3,...,n。
And carrying out Fourier transformation of Gauss self-adaptive optimization window on the frequency points of the selected frequency band, wherein the window function W' is as follows:
3) Performing inverse Fourier transform to obtain a point k i Fast S-transform of (C)
4) Repeating 3) until the rapid S-change of the frequency bin is completed.
The step (5) carries out singular value decomposition on a two-dimensional mode time-frequency matrix of the signal, and fits a specific algorithm flow of a threshold value approximation singular value denoising algorithm:
singular value decomposition is an algorithm widely applied in the field of machine learning, and can be used for feature decomposition in a dimension reduction algorithm, a recommendation system, natural language processing and other fields.
The basic formula of singular value decomposition is:
SVD：A＝UΛV
in the above formula, A is m multiplied by n real number track matrix
A∈R m*n ，U∈R m*m ，V∈R n*n ，Λ∈R m*n All 0 except the elements on the main diagonal, each element on the main diagonal being called a singular value and being ordered in descending order, the column vector of U being AA T Each feature vector in U is generally called the left singular vector of a; the column vector of V is A T The eigenvectors of a, each of which is generally referred to as the right singular vector of a.
Since the positions other than the singular values on the diagonal are all 0 s, solving the singular value matrix Λ is equivalent to solving for each singular value.
The singular values are similar to the eigenvalues in the eigenvalue decomposition, the diagonal matrix elements in the singular value matrix are also arranged in descending order, and the singular value reduction occurs very rapidly, and in most cases, the sum of the first 1% -10% of the singular values accounts for the vast majority of the sum of all the singular values. The matrix can thus be approximated by k singular values and their corresponding left and right singular vectors. That is, the real trajectory matrix may be truncated, and the eigenvalues may be subjected to dimension reduction.
Before SVD noise reduction treatment is carried out on the signals, the signals are subjected to rapid S transformation to obtain a two-dimensional mode time-frequency matrix of the signals, so that the time defect that the traditional singular value noise reduction algorithm can only process one-dimensional time-frequency signals is overcome, the two-dimensional mode time-frequency matrix is increased in time and power information quantity relative to the one-dimensional time-frequency matrix, further research and analysis on partial discharge signals are facilitated, the problem of partial characteristic quantity distortion of PD signals after noise reduction treatment due to time signal deletion is avoided, and therefore the two-dimensional mode time-frequency matrix of the signals is obtained through rapid S change on original signals, and the two-dimensional mode time-frequency matrix with more information is selected as a track matrix of the singular value noise reduction algorithm for noise reduction analysis.
The gap between the approximate matrix and the actual matrix fitted with singular values and the choice of k values are critical. The limitation of the singular value algorithm is how to choose the appropriate threshold.
And adaptively selecting the threshold value of the singular value by adopting a fitting threshold value approximation method. Noise reduction is performed by using a singular value algorithm, and because the singular values of the main signal and the noise signal have larger differences, the singular value of the main signal is generally far larger than the singular value of the noise signal, so that noise reduction can be performed through the singular value, and the maximum point of curve slope change is found through fitting threshold approximation, so that a threshold function of the singular value is determined.
Fitting a threshold value approximation method step;
1) And selecting the first n singular values which are not 0 from the singular value matrix lambda obtained by singular value decomposition, and converting the singular values into singular value vectors.
2) And selecting m data from the singular value vector, equally dividing the m data into n/m sections of intervals to be processed, and selecting m to be even according to the property that singular values appear in pairs.
3) And calculating regression coefficients of the processing intervals, determining a point with the maximum second-order difference coefficient, selecting two adjacent intervals of the point, and performing curve fitting.
4) And determining intersection points of curve fitting, and selecting data slightly smaller than the intersection points from the singular value vectors as a threshold value of a singular value noise reduction algorithm.
In the step (6), in order to detect the filtering effect, a noise suppression ratio rho is introduced NRR And an amplitude decay ratio ρ ARR Two coefficients are used for comprehensively evaluating the denoising performance:
in the actual test process, a clean partial discharge signal without noise cannot be measured, and the partial discharge signal cannot pass through imitationThe pure partial discharge signal in the field environment is really simulated, so that two parameters are required to be reintroduced to evaluate the denoising performance. Thus introducing a noise suppression ratio ρ NRR And an amplitude decay ratio ρ ARR The two coefficients are used to comprehensively evaluate the denoising performance.
Middle sigma 1 For standard deviation of pre-denoising signal, σ 2 To the standard deviation of the denoised signal A m1 For maximum amplitude of signal before denoising, A m2 Is the maximum amplitude of the denoised signal.
The invention adopts a double-exponential oscillation damping model and a single-exponential oscillation damping model to simulate the pure partial discharge signal, a periodic function simulates periodic narrow-band interference noise, and Gaussian white noise is automatically generated by adopting a mathematical algorithm. And preprocessing the signals by using a Gaussian test, and realizing denoising by using a fast S-transformation and fitting threshold singular value approximation algorithm.
The method can finish the noise reduction treatment of the partial discharge noise signal of the transformer.
The method comprises the steps of building a partial discharge signal noise reduction experiment platform, and firstly simulating partial discharge noise to verify the partial discharge signal noise reduction experiment platform. Simulating a large amount of partial discharge noise of the transformer, selecting four typical partial discharge noise for experimental detection, and specific partial discharge noise parameters are shown in table 1
Table 1 four partial discharge noise parameters
And carrying out a Gaussian test on the processed signal, wherein the Gaussian test is required to calculate the kurtosis value of the detected signal, and judging whether the detected signal can pass the Gaussian test or not according to the kurtosis interval, so that the useful signal and the useless signal are distinguished. The choice of confidence parameters required for the test becomes particularly important when the gaussian test is performed, and the larger the confidence value is, the larger the confidence interval is, the more severe the reflected gaussian test is, and if the confidence interval is too large, the useful signal is easily misexcluded. In order to select the appropriate confidence level and thus to strictly distinguish between useful and noise signals, the measured data are shown in the following table:
TABLE 2 kurtosis value estimation
| Signal type | Signal type | Number of | Kurtosis value range |
| A first part | 0-50dB Gaussian white noise | 50 | 0～4 |
| Two (II) | Periodic narrowband interference of 0.3Ghz-1.5Ghz | 50 | 0～4 |
| Three kinds of | White gaussian noise + periodic narrowband interference | 50 | 0～4 |
| Fourth, fourth | Noisy partial discharge signal | 50 | 20～70 |
And sequentially screening all points which cannot pass the Gaussian test from the point with the maximum envelope curve through the Gaussian test, thereby carrying out rapid S transformation. Contrast generalized S-transforms with fast S-transforms.
The generalized S-transform is generally composed of two denoising methods, namely a threshold denoising method and a time-frequency spectrum analysis filtering method. The threshold denoising method is to quantize data according to the maximum value and the minimum value of the two-dimensional matrix distribution, and set a proper threshold for each quantization interval to denoise. The time-frequency analysis filtering method is based on the noise range and the time range of the signal to filter and denoise. The method is characterized in that a time-frequency spectrum analysis filtering method and a threshold denoising method are combined, the frequency range of useful signals is screened out in the Gaussian test process, the time range of the signals is determined according to the energy distribution of the signals, the time range of PD signals is determined approximately, noise signals outside the range are filtered, the specific time range and the frequency range are reserved, and signals in other time-frequency regions are filtered. Fig. 4 is a comparison waveform diagram of the fast S-transform and S-transform.
The gap between the approximate matrix and the actual matrix fitted with singular values and the choice of k values are critical. The limitation of the singular value algorithm is how to choose the appropriate threshold. Noise reduction is performed by using a singular value algorithm, and because the singular values of the main signal and the noise signal have larger differences, the singular value of the main signal is generally far larger than the singular value of the noise signal, so that noise reduction can be performed through the singular value, and the maximum point of curve slope change can be found through fitting threshold approximation, thereby determining a threshold function of the singular value. And (3) finding two adjacent sections with the largest gradient second-order difference spectrum change by using a fitting threshold approximation method, respectively linearly regression-fitting the two sections of curves according to regression coefficients of the two sections, searching an intersection point between the two sections of curves, and finally searching a singular value slightly smaller than the intersection point value in a singular value sequence as a threshold value of the singular value noise reduction algorithm, so that singular value reconstruction is carried out. Fig. 6 and 7 are waveform diagrams of the fitting threshold value approaching singular value denoising effect.
Partial discharge signal denoising processing should follow two main criteria: and the distortion of the two denoising signals is small when the requirements of the specified signal to noise ratio are met. Partial discharge denoising effect is characterized by introducing an introduced signal-to-noise ratio (Signal Noise Ration, SNR), square root mean error (Root Mean Square Error, RMSE), waveform similarity coefficient (Normalized Correlation Coefficient, NCC) and variation trend parameters (Variation Trend Parameter, VTP), wherein the parameters are set as follows:
TABLE 3 denoising Effect parameter table
And (5) comprehensive comparison. By comparing the method with wavelet transformation denoising method, empirical mode decomposition denoising method, singular value denoising method, denoising effect is shown in the following table
TABLE 4RMSE data Table
| Denoising method | Waveform 1 | Waveform 2 | Waveform shape3 | Waveform 4 |
| Methods herein | 0.093 | 0.105 | 0.137 | 0.143 |
| Hard threshold | 0.321 | 2.024 | 0.397 | 0.274 |
| Soft threshold | 0.414 | 1.513 | 0.469 | 0.358 |
| Fixed threshold | 0.463 | 1.426 | 0.574 | 0.390 |
| EMD | 0.421 | 0.457 | 0.506 | 0.442 |
| SVD | 0.336 | 0.205 | 0.529 | 0.449 |
Table 5SNR data table
TABLE 6NCC data sheet
| Denoising method | Waveform 1 | Waveform 2 | Waveform 3 | Waveform 4 |
| Methods herein | 0.995 | 0.993 | 0.990 | 0.995 |
| Hard threshold | 0.955 | 0.363 | 0.453 | 0.342 |
| Soft threshold | 0.907 | 0.324 | 0.467 | 0.357 |
| Fixed threshold | 0.875 | 0.309 | 0.469 | 0.359 |
| EMD | 0.554 | 0.696 | 0.731 | 0.632 |
| SVD | 0.833 | 0.357 | 0.381 | 0.229 |
TABLE 7VTP data tables
| Denoising method | Waveform 1 | Waveform 2 | Waveform 3 | Waveform 4 |
| Methods herein | 1.004 | 0.995 | 1.078 | 1.060 |
| Hard threshold | 1.157 | 1.357 | 0.751 | 0.684 |
| Soft threshold | 1.297 | 1.533 | 1.174 | 0.750 |
| Fixed threshold | 1.284 | 1.235 | 0.784 | 0.822 |
| EMD | 0.846 | 1.324 | 1.452 | 0.855 |
| SVD | 0.885 | 0.950 | 0.702 | 0.642 |
The insulation system of the transformer is complex, the involved materials are numerous, and the electric field distribution is uneven, so that a plurality of types of partial discharge exist inside the transformer. The partial discharges occurring in transformers can be broadly divided into three basic types, depending on the location, phenomena and mechanism of the partial discharges occurring: (1) partial discharge inside the insulating medium; (2) partial discharge of the surface of the insulating medium; (3) corona discharge at the tip of the high voltage electrode. Therefore, the transformer needs to be measured in the field, the field measuring field is shown as the field space of a transformer of a certain substation, and an antenna is used for receiving the partial discharge UHF signal.
Table 8 actual filter parameters
| ρ NRR | ρ ARR | Time/s |
| 23.60 | 0.052 | 0.341 |
In summary, the partial discharge signal denoising processing is realized by the Gaussian test preprocessing, the rapid S transformation and the fitting threshold approximation denoising method of the singular value, and Gaussian white noise and periodic narrow-band interference noise are filtered. Aiming at the problem of long time consumption of local signal feature extraction, a rapid S conversion algorithm is provided, the operation speed of the algorithm is reduced, the time-frequency resolution of the algorithm is improved, the traditional singular value denoising algorithm can only process one-dimensional data, and the two-dimensional mode time-frequency matrix is processed by the improved singular value algorithm, so that signal distortion is compensated, noise in a local discharge signal can be effectively filtered in field test, feature extraction time is shortened, meanwhile, distortion generated in the denoising process can be reduced, and original features of the signal are reserved to the greatest extent.
Claims (5)
1. A method for reducing noise of partial discharge signals, characterized by comprising the following steps:
s1: building a partial discharge signal noise reduction experiment platform;
s2: by adopting a double-exponential oscillation damping model and a single-exponential oscillation damping model, the simulation of the partial discharge signals is completed, a plurality of groups of partial discharge noise-containing signals are set as experimental groups, and each group of partial discharge noise-containing signals are different;
s3: each group of experiment groups is provided with a control group according to a preset number for training, a kurtosis test is determined, a confidence interval of a Gaussian test is set, gaussian test pretreatment is perfected, and noise-containing partial discharge signals of each group of experiment groups are subjected to Gaussian test pretreatment to obtain a frequency range where characteristic signals are located;
s4: the signals after pretreatment are subjected to rapid S transformation, and a denoising algorithm based on the combination of a time-frequency spectrum analysis filtering method and a threshold denoising method is adopted to remove most of periodic narrowband interference signals in the signals;
s5: the signal after the rapid S transformation is subjected to a fitting threshold value approximation singular value denoising algorithm, a signal two-dimensional mode time-frequency matrix is analyzed, and Gaussian white noise in the signal is filtered;
s6: and testing the noise suppression ratio and the amplitude attenuation coefficient of the noise-reduced signal, analyzing the test result, and displaying the test result on a display interface.
2. The method for noise reduction of partial discharge signals according to claim 1, wherein the step S3 of performing gaussian test preprocessing on the partial discharge signals containing noise comprises the following steps:
knowing the random variable X of mean μ and variance σ, gaussian test was performed on the random variable X, and the distribution interval of the random variable X satisfies the probability inequality according to the Chebyshev inequality:
in the above formula: q=1- (σ) 2 /ε 2 ) Epsilon > 0, assuming that a group of observation samples X (t) with the mean value of 0 and the number of N exist in random variables X conforming to Gaussian distribution, when sigma 2 /ε 2 Sufficiently small, the estimate of the random variable X kurtosis value and variance can be expressed as:
wherein, the liquid crystal display device comprises a liquid crystal display device,is the mean value of kurtosis value, +.>Variance of kurtosis value,/->Kurtosis coefficient +.>The early failure of the reaction system is determined by the number N of observation points and the confidence q, and if the kurtosis value estimated by the variable sequence x (t) is in the interval, the detected signal has obvious Gaussian characteristic and needs to be eliminated.
3. The partial discharge signal noise reduction method according to claim 1, wherein:
in the preprocessing stage, obtaining the frequency range of a useful PD signal, setting a characteristic frequency retaining mechanism and a non-characteristic signal removing mechanism through an envelope extremum algorithm, and screening out characteristic frequencies through an envelope spectrum so as to independently process the characteristic frequencies;
because the fast S-transform is reversible and linear, a time-frequency filter is designed based on the time-frequency signal after the transform, preserving the time range t E [ t ] 1 ,t 2 ]Frequency range f e f 1 ,f 2 ]While suppressing the time range t e t 3 ,t 4 ]Data f e f of frequency range 3 ,f 4 ]Discretizing the fast S transform, let f=n/NT, τ=jt:
wherein f is frequency, T and τ are time, w (T- τ, f) is a Gaussian window function, λ and p are adjustment factors, λ=0.55, p=0.8, S [ jT, N/NT ] is a fast S transform in discrete case, m, N is a constant, T is a sampling time interval, N is a sampling total point number, and X [ (m+n)/NT ] is a signal X (nT) discrete Fourier transform;
in step S4, the signal after preprocessing is subjected to fast S transformation, and the method of fast S transformation includes the following steps:
s41: screening out useful frequency points k through a Gaussian test according to an envelope spectrum obtained through pretreatment i ，i＝1,2,3,...,n；
S42: fourier transforming the frequency points of the selected frequency band, wherein the window function W' is:
s43: performing inverse Fourier transform to obtain a point k i Fast S-transformation of (a);
s44: step S43 is repeated until the rapid S-transformation of all frequency points is completed.
4. The partial discharge signal noise reduction method according to claim 1, wherein: in the step S5, the two-dimensional mode time-frequency matrix of the signal is analyzed, and the method for filtering Gaussian white noise in the signal comprises the following steps:
s51: selecting the first n singular values which are not 0 from a singular value matrix lambda obtained by singular value decomposition, and converting the singular values into singular value vectors;
s52: selecting m data from the singular value vector, equally dividing the m data into n/m sections of intervals to be processed, and selecting m to be even according to the property that singular values appear in pairs;
s53: calculating regression coefficients of the processing intervals, determining a point with the maximum second-order difference coefficient, selecting two adjacent intervals of the point, and performing curve fitting;
s54: and determining intersection points of curve fitting, and selecting data slightly smaller than the intersection points from the singular value vectors as a threshold value of a singular value noise reduction algorithm.
5. The partial discharge signal noise reduction method according to claim 1, wherein in step S6, in order to detect the filtering effect, a noise suppression ratio ρ is introduced NRR And an amplitude decay ratio ρ ARR Two coefficients toComprehensively evaluating denoising performance:
middle sigma 1 For standard deviation of pre-denoising signal, σ 2 To the standard deviation of the denoised signal A m1 For maximum amplitude of signal before denoising, A m2 Is the maximum amplitude of the denoised signal.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310489104.2A CN116522080A (en) | 2023-05-04 | 2023-05-04 | Partial discharge signal noise reduction method |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202310489104.2A CN116522080A (en) | 2023-05-04 | 2023-05-04 | Partial discharge signal noise reduction method |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN116522080A true CN116522080A (en) | 2023-08-01 |
Family
ID=87398957
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202310489104.2A Pending CN116522080A (en) | 2023-05-04 | 2023-05-04 | Partial discharge signal noise reduction method |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN116522080A (en) |
Cited By (1)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407650A (en) * | 2023-12-13 | 2024-01-16 | 中汽研新能源汽车检验中心(天津)有限公司 | Noise quality level evaluation method for driving motor system |
-
2023
- 2023-05-04 CN CN202310489104.2A patent/CN116522080A/en active Pending
Cited By (2)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN117407650A (en) * | 2023-12-13 | 2024-01-16 | 中汽研新能源汽车检验中心(天津)有限公司 | Noise quality level evaluation method for driving motor system |
| CN117407650B (en) * | 2023-12-13 | 2024-04-09 | 中汽研新能源汽车检验中心(天津)有限公司 | Noise quality level evaluation method for driving motor system |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| Smulko | Methods of trend removal in electrochemical noise data–Overview | |
| CN113777650B (en) | Sparse time-frequency spectrum decomposition method, device and equipment based on mixed norm and wavelet transform and storage medium | |
| CN112014692A (en) | Partial discharge ultrahigh frequency signal blind source separation denoising method based on principal component analysis | |
| CN106443183B (en) | A kind of method that frequency source quantity and fundamental frequency value are extracted from electromagnetic spectrum | |
| CN107315991B (en) | IFRA frequency response curve denoising method based on wavelet threshold denoising | |
| CN116522080A (en) | Partial discharge signal noise reduction method | |
| Govindarajan et al. | Development of Hankel‐SVD hybrid technique for multiple noise removal from PD signature | |
| CN113887398A (en) | GPR signal denoising method based on variational modal decomposition and singular spectrum analysis | |
| CN114035007A (en) | Non-contact transformer bushing partial discharge detection system | |
| CN113486750A (en) | Oil-immersed transformer partial discharge signal denoising method based on improved VMD algorithm and wavelet packet denoising algorithm | |
| Song et al. | An improved reconstruction method based on auto-adjustable step size sparsity adaptive matching pursuit and adaptive modular dictionary update for acoustic emission signals of rails | |
| CN116559110A (en) | Self-adaptive near infrared spectrum transformation method based on correlation and Gaussian curve fitting | |
| Thibault et al. | Comparison of signal processing methods considering their optimal parameters using synthetic signals in a heat exchanger network simulation | |
| CN111505709B (en) | Attenuation qualitative analysis method based on sparse spectral decomposition | |
| CN110780162B (en) | Method for extracting partial discharge signal of primary and secondary fusion power distribution switch and detection device | |
| Chaudhuri et al. | Denoising of partial discharge signal using a hybrid framework of total variation denoising-autoencoder | |
| CN112183260A (en) | One-way valve fault diagnosis method based on total variation noise reduction and RQA | |
| CN111079591A (en) | Bad data restoration method and system based on improved multi-scale principal component analysis | |
| CN112183297B (en) | Ultrasonic phased array signal sparse feature extraction method | |
| CN112729531B (en) | Fault studying and judging method and system for distribution transformer equipment | |
| Xin et al. | De-noising of GIS UHF Partial Discharge monitoring based on wavelet method | |
| CN116028840A (en) | Marine rotor fault diagnosis method for maximum overlapped discrete wavelet packet transformation time spectrum | |
| Hejtmanek et al. | Comparison of two denoising techniques to improve UHF partial discharge localization | |
| CN113378614B (en) | Partial discharge signal denoising processing method and system | |
| Xu et al. | Research on partial discharge de-noising for transformer based on synchro-squeezed continuous wavelet transform |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |