Partial discharge signal wavelet denoising method based on all-parameter space traversal optimization

Abstract

The invention discloses a partial discharge signal wavelet denoising method based on all-parameter space traversal optimization. The method comprises the following steps: S1, initializing parameters; s2, a noisy signal is read; s3, constructing a wavelet basis traversal cycle; s4, constructing a decomposition layer number traversal cycle; s5, denoising and index calculation are executed; s6, comparing and updating a global optimal solution; and S7, carrying out loop iteration until all preset wavelet base order and decomposition layer combinations are traversed. According to the method, the parameter search space containing various wavelet bases and different decomposition layer numbers is constructed, so that automation and optimal matching of denoising parameters are realized; compared with a traditional method of selecting wavelet parameters depending on artificial experience, the method has the advantages that the influence of subjective factors on the denoising effect is effectively avoided, and the adaptability and robustness of the algorithm to different field environments and different types of partial discharge signals are remarkably improved.

Description

TECHNICAL FIELD

[0001] The application belongs to the technical field of high-voltage insulation detection and digital signal processing, and particularly relates to a partial discharge signal wavelet denoising method based on full-parameter space traversal optimization. BACKGROUND

[0002] Partial discharge detection is an important means to evaluate the insulation state of power equipment (such as transformers, GIS, cables, etc.). However, the partial discharge signals collected on site are usually very weak (millivolt level), and inevitably mixed with a large amount of environmental background noise, especially Gaussian white noise, which seriously affects the subsequent feature extraction and fault diagnosis.

[0003] Wavelet transform, due to its excellent time-frequency localization characteristics, is the mainstream method for PD signal denoising at present, mainly including the following steps:

[0004] a. Determine the appropriate wavelet basis and decomposition level through experiments. Then, according to the selected wavelet basis and decomposition level, perform wavelet decomposition on the noisy signal, and decompose the signal into approximate coefficients (low-frequency information) and detail coefficients (high-frequency noise and details) at different scales.

[0005] b. Denoising processing is performed on the decomposed wavelet coefficients (mainly detail coefficients). That is, by calculating the threshold to determine the boundary between noise and signal, and combining the selected threshold function (hard threshold, soft threshold function) to quantize and shrink or truncate the coefficients, to eliminate the noise components.

[0006] c. Using the detail coefficients processed by threshold and the unprocessed approximate coefficients, perform wavelet reconstruction to restore the denoised signal that retains the effective features.

[0007] However, in practical applications, wavelet denoising has "two selection problems": one is the difficulty in selecting wavelet basis, different types of PD pulses (single exponential, double exponential decay oscillation, etc.) have different similarities with different wavelet bases (such as Daubechies, Symlet, conif series), and artificial selection lacks standards; the second is the difficulty in determining the decomposition level: too few levels will not denoise thoroughly, and too many levels will cause the loss of high-frequency details of useful signals.

[0008] In the prior art, in order to determine these two key parameters, artificial experience method or simple "trial and error method" is usually used. However, partial discharge signals are transient and non-stationary, and the noise environment on site is complex and variable, so fixed parameters or parameters selected by experience are difficult to match the changing signal characteristics. If the parameters are not properly selected, not only the signal-to-noise ratio cannot be effectively improved, but also the start and end time of PD pulses may be blurred due to excessive smoothing, which seriously affects the positioning accuracy of the discharge source.

[0009] In addition, the selection of the wavelet basis and the determination of the decomposition layer number are often coupled with each other. In the absence of objective evaluation indexes, it is difficult to find the optimal combination of the two parameters at the same time. At present, there is a lack of an efficient method capable of adaptively performing joint optimization of the wavelet basis and the decomposition layer number according to the characteristics of the collected noisy signal. SUMMARY

[0010] The present application aims at the problem that the parameter selection in the existing wavelet denoising method depends on artificial experience and lacks adaptiveness, and provides a partial discharge signal wavelet denoising method based on full parameter space traversal optimization, which aims to realize adaptive matching of the wavelet basis and the decomposition layer number.

[0011] In order to achieve the above-mentioned purpose, the present application adopts the following technical solutions:

[0012] The partial discharge signal wavelet denoising method based on full parameter space traversal optimization comprises the following steps:

[0013] Step S1: parameter initialization;

[0014] First, the global variable is initialized, that is, the current global minimum index residual autocorrelation RAC is set to infinity (RAC = inf), and the best denoised signal PDdenoise is initialized to a zero vector of 1xN, and N is the data length of the noisy signal, that is, the number of sampling points.

[0015] Step S2: reading a noisy signal;

[0016] Read the field collected partial discharge noisy signal to be processed, denoted as PDSignal;

[0017] Step S3: constructing a wavelet basis traversal loop;

[0018] The initial value of the order variable Order of the wavelet basis is set to 1; enter the outer loop, judge whether the current Order is less than log2(N); if the judgment is "no", it means that all the preset wavelet basis orders have been traversed, the algorithm ends, and the final saved PDdenoise is output; if the judgment is "yes", the current numerical value Order is spliced with the string "db", "sym" or "coif" (wavelet basis type, namely Daubechies, Symlet, conif), so as to generate a specific wavelet basis name;

[0019] Step S4: constructing a decomposition layer number traversal loop;

[0020] Set the initial value of the decomposition level variable Level as 1; enter the inner loop, judge whether the current Level is less than log2(N); if the judgment is "no", it means that all levels under the current wavelet base have been traversed, execute Order = Order + 1, return to step S3 to continue the traversal of the next order wavelet base; if the judgment is "yes", use the current determined wavelet base and the current decomposition level Level to perform wavelet denoising processing on the noisy signal PDSignal;

[0021] Step S5: execute denoising and index calculation;

[0022] According to the current parameter combination (i.e. wavelet base type + order + decomposition level), wavelet denoising is performed to obtain the current denoised partial discharge signal, denoted as PDdenoise'; then, the residual autocorrelation of the denoised partial discharge signal is calculated using the preset evaluation algorithm, denoted as RAC';

[0023] Step S6: global optimal solution comparison and update;

[0024] Compare the index RAC' calculated in the current step S5 with the global minimum index RAC: if RAC' < RAC, it means that the current parameter combination is better than the previous best combination; at this time, the update operation is performed: update the global minimum index to the current value (RAC = RAC'), and update the current denoised partial discharge signal to the best result (PDdenoise = PDdenoise'); if RAC' ≥ RAC, keep the global parameters unchanged;

[0025] Step S7: loop iteration;

[0026] After the comparison is completed, execute Level = Level + 1, return to step S4 to continue testing the next decomposition level; repeat steps S4 to S7 until all preset wavelet base order and decomposition level combinations are traversed, and the optimal denoised partial discharge signal is obtained.

[0027] The method of the application firstly constructs a full parameter traversal space of wavelet base order and decomposition layer number; then performs multi-scale discrete wavelet decomposition on the noisy partial discharge signal, preferably utilizes a general threshold value based on noise level estimation and a soft threshold function to perform quantization processing and reconstruction; introduces residual autocorrelation (RAC) as a no-reference evaluation index, and evaluates the fidelity by detecting whether the residual sequence conforms to the statistical characteristics of white noise; finally, through global iterative comparison, the parameter combination corresponding to the minimum RAC is locked, and the optimal denoising signal is output. The application can realize adaptive optimal matching of denoising parameters in the absence of ideal true values, effectively suppress white noise while perfectly preserving the steep wave front and oscillation details of the partial discharge signal, and significantly improve the accuracy of insulation fault diagnosis.

[0028] As a further description of the application, in step S4, the wavelet denoising processing is performed on the noisy signal PDSignal, specifically including the following steps:

[0029] Step S41: wavelet decomposition;

[0030] The noisy signal PDSignal is subjected to multi-scale discrete wavelet decomposition using the "wavelet base (type+Order) and decomposition layer number Level" determined in the current cycle; after decomposition, the approximation coefficient cA Level of the Level layer and the detail coefficients cD1, cD2, ···, cD Level of each layer from the 1st layer to the Level layer are obtained; wherein the detail coefficients contain high-frequency noise components and part of the mutation details of the signal;

[0031] Step S42: threshold estimation;

[0032] The denoising threshold λ is calculated for each layer of detail coefficients;

[0033] Step S43: threshold quantization processing;

[0034] The denoising threshold λ calculated is used to perform soft threshold processing on all levels of detail coefficients cD j to remove noise components; compared with hard threshold, the soft threshold function has continuity, can effectively avoid pseudo Gibbs oscillation of the denoised signal, and is more suitable for the recovery of partial discharge pulses; the calculation formula of the processed detail coefficients is as follows:

[0035] ;

[0036] Wherein, New(cD j ) is the processed detail coefficient of the jth level; j is the jth level of the decomposition layer number; sgn() is the sign function;

[0037] For the approximation coefficients, considering that they mainly contain the low frequency trend of the signal, usually no threshold processing or only weak processing is performed, and they are directly reserved;

[0038] Step S44: wavelet reconstruction;

[0039] Using the reserved approximation coefficients and the soft threshold processed detail coefficients of each layer, the signal is reconstructed using the wavelet inverse transform algorithm, and the obtained reconstructed signal is the current temporary denoising signal PDdenoise'.

[0040] As a further illustration of the application, in step S42, in order to adapt to the non-stationary characteristics of the partial discharge signal, the denoising threshold λ is determined using a general threshold formula based on noise level estimation:

[0041] First, the first layer detail coefficient cD1 is used to estimate the noise standard deviation σ:

[0042] ;

[0043] Where median() is the median function, i.e. the value in the middle position after sorting the sequence elements by numerical value;

[0044] Then, the denoising threshold λ is calculated in combination with the signal length N:

[0045] .

[0046] As a further illustration of the application, in step S5, the calculation process of the residual autocorrelation RAC' of the denoised partial discharge signal is as follows:

[0047] Step S51: obtain the residual sequence;

[0048] First, define the residual sequence r (which is the difference between the original noisy PD signal and the signal processed by the denoising algorithm, representing the "noise" component removed by the algorithm):

[0049] ;

[0050] Step S52: de-meaning processing;

[0051] Eliminate the DC component of the residual sequence r[N] to ensure that the subsequent correlation calculation only reflects the fluctuation characteristics of the signal:

[0052] ;

[0053] Where μ r =mean(r) is the mean of the residual sequence, mean() is the mean function, and r' is the residual sequence after eliminating the DC component;

[0054] Step S53: calculating normalized autocorrelation;

[0055] The lag one order correlation coefficient of the de-mean residual sequence is calculated by using the sample autocorrelation function, in order to eliminate the influence of signal energy difference on the index, normalization processing (so that the autocorrelation coefficient at zero lag is 1) is adopted, and the calculation formula is defined as follows:

[0056] ;

[0057] Wherein, R is the correlation coefficient, i is the index of the residual sequence r';

[0058] Index definition: taking the absolute value of the lag one order coefficient as the final evaluation index residual autocorrelation RAC', that is:

[0059] ;

[0060] If RAC' =0, it indicates that the adjacent points of the residual sequence are not correlated with each other, which meets the statistical characteristics of white noise, and it is indicated that the algorithm removes noise without damaging the original signal structure;

[0061] If RAC' is significantly greater than 0, it indicates that there are signal components with deterministic rules (that is, effective signal leakage) in the residual, which reflects the problem of excessive denoising of the algorithm.

[0062] Advantages of the present application:

[0063] The wavelet denoising method for partial discharge signal based on full parameter space traversal optimization proposed in the present application realizes the automation and best matching of denoising parameters by constructing a parameter search space containing multiple wavelet bases and different decomposition layers. Compared with the traditional method of selecting wavelet parameters depending on artificial experience, the present application effectively avoids the influence of subjective factors on the denoising effect, and significantly improves the adaptability and robustness of the algorithm to different field environments and different types of partial discharge signals. BRIEF DESCRIPTION OF DRAWINGS

[0064] Figure 1 It is a flow chart of the wavelet denoising method for partial discharge signal based on full parameter space traversal optimization;

[0065] Figure 2 It is a wavelet denoising flow chart;

[0066] Figure 3 It is a schematic diagram of a partial discharge simulation test platform;

[0067] Figure 4 It is a schematic diagram of measuring and detecting the voltage waveform at both ends of the impedance by using an oscilloscope;

[0068] Figure 5The algorithm iterates through the preset wavelet base library and decomposition level, and calculates the corresponding residual autocorrelation diagram.

[0069] Figure 6 This is a schematic diagram of the waveform of the denoised signal after processing;

[0070] Figure 7 The PRPS spectrum before denoising;

[0071] Figure 8 This is the PRPS spectrum after denoising. Detailed Implementation

[0072] The invention will be further described below with reference to the accompanying drawings.

[0073] Example: A wavelet denoising method for partial discharge signals based on full-parameter space ergonomic optimization, such as... Figure 1 The steps shown are as follows:

[0074] Step S1: Parameter initialization;

[0075] First, initialize the global variables, that is: set the current global minimum index residual autocorrelation RAC to infinity (RAC = inf), and initialize the best denoised signal PDdenoise to a 1×N zero vector, where N is the data length of the noisy signal, i.e. the number of sampling points;

[0076] Step S2: Read the noisy signal;

[0077] Read the noisy partial discharge signal acquired on-site and denoted as PDSignal;

[0078] Step S3: Construct the wavelet basis traversal loop;

[0079] The initial value of the wavelet basis order variable Order is set to 1; the outer loop is entered to check if the current Order is less than log2(N); if the check is "no", it means that all preset wavelet basis orders have been traversed, the algorithm ends, and the final saved PDdenoise is output; if the check is "yes", the next step is to concatenate the current value Order with the string "db", "sym" or "coif" to generate a specific wavelet basis name;

[0080] Step S4: Construct a decomposition level traversal loop;

[0081] Set the initial value of the decomposition level variable Level as 1; enter the inner loop, judge whether the current Level is less than log2(N); if the judgment is "no", it means that all levels under the current wavelet base have been traversed, execute Order = Order + 1, return to step S3 to continue the traversal of the next order wavelet base; if the judgment is "yes", use the current determined wavelet base and the current decomposition level Level to perform wavelet denoising processing on the noisy signal PDSignal;

[0082] Step S5: execute denoising and index calculation;

[0083] According to the current parameter combination, wavelet denoising is performed to obtain the current denoised partial discharge signal, denoted as PDdenoise'; then, the residual autocorrelation of the denoised partial discharge signal is calculated using the preset evaluation algorithm, denoted as RAC';

[0084] Step S6: global optimal solution comparison and update;

[0085] Compare the index RAC' calculated in the current step S5 with the global minimum index RAC: if RAC' < RAC, it means that the current parameter combination is better than the previous best combination; at this time, the update operation is performed: update the global minimum index to the current value (RAC = RAC'), and update the current denoised partial discharge signal to the best result (PDdenoise = PDdenoise'); if RAC' ≥ RAC, keep the global parameters unchanged;

[0086] Step S7: loop iteration;

[0087] After the comparison is completed, execute Level = Level + 1, return to step S4 to continue testing the next decomposition level; repeat steps S4 to S7 until all preset wavelet base order and decomposition level combinations are traversed to obtain the optimal denoised partial discharge signal.

[0088] As a further illustration of the present embodiment, in step S4, the wavelet denoising processing on the noisy signal PDSignal includes the following steps as shown in Figure 2

[0089] Step S41: wavelet decomposition;

[0090] Use the "wavelet base (type + Order) and decomposition level Level" determined in the current loop to perform multi-scale discrete wavelet decomposition on the noisy signal PDSignal; after decomposition, the approximation coefficient cA Level ​and the detail coefficients cD1, cD2, ···, cDof each level from the 1st level to the Levelth level Level ; wherein the detail coefficients contain high frequency noise components and partial abrupt details of the signal;

[0091] Step S42: threshold estimation;

[0092] Calculate the denoising threshold λ for each layer of detail coefficients;

[0093] Step S43: threshold quantization processing;

[0094] Use the calculated denoising threshold λ to perform soft threshold processing on the detail coefficients cD j of all levels to remove noise components; compared with hard threshold, soft threshold function has continuity, which can effectively avoid pseudo Gibbs oscillation of the denoised signal, and is more suitable for recovery of partial discharge pulses; the processed detail coefficients are calculated according to the following formula:

[0095]

[0096] wherein New(cD j ) is the processed detail coefficient of the jth level; j is the jth level of the decomposition level number; sgn( ) is a sign function;

[0097] For the approximation coefficients, considering that they mainly contain low frequency trends of the signal, usually no threshold processing or only weak processing is performed, and they are directly reserved;

[0098] Step S44: wavelet reconstruction;

[0099] Use the reserved approximation coefficients and the soft threshold processed detail coefficients of each level to perform signal reconstruction using the wavelet inverse transform algorithm, and the obtained reconstructed signal is the current tentative denoised signal PDdenoise'.

[0100] As a further illustration of the embodiment, in step S42, in order to adapt to the non-stationary characteristics of the partial discharge signal, the denoising threshold λ is determined using a general threshold formula based on noise level estimation:

[0101] First, estimate the noise standard deviation σ using the first layer detail coefficient cD1:

[0102]

[0103] wherein median( ) is a median function, that is, the numerical value in the middle position after sorting the sequence elements according to numerical value;

[0104] Subsequently, calculate the denoising threshold λ in combination with the signal length N: ​​

[0105] .

[0106] As a further illustration of the present embodiment, in step S5, the calculation process of the residual autocorrelation RAC' of the de-noised partial discharge signal is as follows:

[0107] Step S51: Obtain the residual sequence;

[0108] First, define the residual sequence r (which is the difference between the original noisy PD signal and the signal processed by the de-noising algorithm, representing the "noise" component removed by the algorithm):

[0109] ;

[0110] Step S52: De-meaning processing;

[0111] Eliminate the DC component of the residual sequence r[N] to ensure that the subsequent correlation calculation only reflects the fluctuation characteristics of the signal:

[0112] ;

[0113] Where μ r =mean(r) is the mean of the residual sequence, mean() is the mean function, and r' is the residual sequence after eliminating the DC component;

[0114] Step S53: Calculate the normalized autocorrelation;

[0115] Use the sample autocorrelation function to calculate the lag one order correlation coefficient of the de-meaned residual sequence. In order to eliminate the influence of signal energy difference on the index, normalization processing (so that the autocorrelation coefficient at zero lag is 1) is adopted, and the calculation formula is defined as follows:

[0116] ;

[0117] Where R is the correlation coefficient, and i is the index of the residual sequence r';

[0118] Index definition: take the absolute value of the lag one order coefficient as the final evaluation index residual autocorrelation RAC', that is:

[0119] ;

[0120] If RAC' =0, it indicates that the adjacent points of the residual sequence are not correlated with each other, which meets the statistical characteristics of white noise, indicating that the algorithm removes noise without destroying the original signal structure;

[0121] If RAC' is significantly greater than 0, it indicates that there is a signal component with deterministic law remaining in the residual (i.e., effective signal leakage), which reflects that the algorithm has the problem of over denoising.

[0122] Application example:

[0123] A partial discharge simulation test platform as shown in Figure 3 is built, which includes a test transformer control box A, a no corona power frequency test transformer T, an artificial defect simulation module B, a coupling capacitor Ck and a detection impedance Zd; the coupling capacitor Ck has a loop for providing a pulse current generated by the artificial defect simulation module B, and avoids the direct action of the power frequency high voltage on the detection impedance Zd.

[0124] The oscilloscope is used to measure the voltage waveform at both ends of the detection impedance Zd as shown in Figure 4 From the figure, it can be seen that due to the complex electromagnetic interference in the test site, the original partial discharge signal collected is mixed with significant broadband background noise. The existence of the noise seriously covers the steep wave front and oscillation tail characteristics of the partial discharge pulse, resulting in extremely low signal-to-noise ratio of the signal, which is difficult to be directly used for subsequent feature extraction and insulation fault diagnosis.

[0125] In order to verify the effectiveness of the present application, the above-mentioned noise-containing signal is taken as an input PDSignal, and the wavelet denoising method based on full parameter space traversal optimization described in the embodiment of the present application is used for processing. The algorithm automatically traverses the preset wavelet basis library and decomposition layer number, and calculates the corresponding residual autocorrelation as shown in Figure 5 During the traversal process, the algorithm finds the parameter combination "the basis function and the decomposition layer number are db6 and 6 layers of decomposition" with the minimum RAC value as the global optimal solution.

[0126] The denoised signal waveform processed by the method of the present example is shown in Figure 6 Compared with the noise-containing signal, the background noise of the processed signal is effectively suppressed, and the baseline tends to be flat. At the same time, the amplitude and phase information of the partial discharge pulse are well preserved, and there is no obvious waveform distortion or oscillation attenuation. In addition, the residual sequence calculated in the time domain shows disordered random fluctuation characteristics, and its autocorrelation coefficient is significantly close to zero, which verifies that the method of the present application does not cause effective signal leakage while removing noise, and achieves a high-fidelity denoising effect.

[0127] Figure 7 and Figure 8 are PRPS spectrum diagrams before and after denoising, respectively. As Figure 8As shown in the original PRPS spectrum before denoising, due to the existence of serious electromagnetic interference in the field environment, a large number of chaotic background noise points are filled in the whole power frequency cycle. These high-density noise points seriously cover the real phase distribution characteristics of the partial discharge signal, resulting in that the outline of the discharge spectrum is blurred and it is difficult to identify the unique discharge mode from it.

[0128] As shown in the original PRPS spectrum before denoising, due to the existence of serious electromagnetic interference in the field environment, a large number of chaotic background noise points are filled in the whole power frequency cycle. These high-density noise points seriously cover the real phase distribution characteristics of the partial discharge signal, resulting in that the outline of the discharge spectrum is blurred and it is difficult to identify the unique discharge mode from it. Figure 8 As shown in the original PRPS spectrum before denoising, due to the existence of serious electromagnetic interference in the field environment, a large number of chaotic background noise points are filled in the whole power frequency cycle. These high-density noise points seriously cover the real phase distribution characteristics of the partial discharge signal, resulting in that the outline of the discharge spectrum is blurred and it is difficult to identify the unique discharge mode from it. Figure 7 As shown in the original PRPS spectrum before denoising, due to the existence of serious electromagnetic interference in the field environment, a large number of chaotic background noise points are filled in the whole power frequency cycle. These high-density noise points seriously cover the real phase distribution characteristics of the partial discharge signal, resulting in that the outline of the discharge spectrum is blurred and it is difficult to identify the unique discharge mode from it. Figure 8 As shown in the original PRPS spectrum before denoising, due to the existence of serious electromagnetic interference in the field environment, a large number of chaotic background noise points are filled in the whole power frequency cycle. These high-density noise points seriously cover the real phase distribution characteristics of the partial discharge signal, resulting in that the outline of the discharge spectrum is blurred and it is difficult to identify the unique discharge mode from it.

[0129] Obviously, the above embodiments are only examples for clearly illustrating the present application, and are not intended to limit the implementation of the present application. For those skilled in the art, other different forms of changes or variations can be made on the basis of the above description; here, it is not necessary and impossible to exhaust all the embodiments; the obvious changes or variations derived therefrom are still within the protection scope of the present application.

Claims

1. A wavelet denoising method for partial discharge signals based on full-parameter spatial ergonomic optimization, characterized in that... It includes the following steps: Step S1: Parameter initialization; First, initialize the global variables, that is: set the current global minimum index residual autocorrelation RAC to infinity, and initialize the best denoised signal PDdenoise as a zero vector of 1×N, where N is the data length of the noisy signal, that is, the number of sampling points; Step S2: Read the noisy signal; Read the locally collected partial discharge noisy signal to be processed, denoted as PDSignal; Step S3: Construct a wavelet basis traversal loop; Set the initial value of the wavelet basis order variable Order to 1; enter the outer loop, and judge whether the current Order is less than log2(N); if the judgment is "no", it means that all preset wavelet basis orders have been traversed, the algorithm ends, and the finally saved PDdenoise is output; if the judgment is "yes", then enter the next step, and splice the current value of Order with the strings "db", "sym", or "coif" to generate a specific wavelet basis name; Step S4: Construct a decomposition level traversal loop; Set the initial value of the decomposition level variable Level to 1; Enter the inner loop, and judge whether the current Level is less than log2(N); if the judgment is "no", it means that all levels under the current wavelet basis have been traversed, execute Order = Order + 1, and return to Step S3 to continue traversing the next order of wavelet basis; if the judgment is "yes", then perform wavelet denoising on the noisy signal PDSignal using the currently determined wavelet basis and the current decomposition level Level; Step S5: Perform denoising and index calculation; Perform wavelet denoising according to the current parameter combination to obtain the current denoised partial discharge signal, denoted as PDdenoise'; subsequently, calculate the residual autocorrelation of the denoised partial discharge signal using a preset evaluation algorithm, denoted as RAC'; Step S6: Compare and update the global optimal solution; Compare the index RAC' calculated in the current Step S5 with the global minimum index RAC: if RAC' < RAC, it means that the current parameter combination is better than the previous best combination; at this time, perform an update operation: update the global minimum index to the current value, and update the current denoised partial discharge signal to the best result; if RAC' ≥ RAC, the global parameters remain unchanged; Step S7: Loop iteration; After the comparison is completed, execute Level = Level + 1, return to Step S4, and continue to test the next decomposition level; repeat Steps S4 to S7 until all preset wavelet basis orders and decomposition level combinations are traversed to obtain the optimal denoised partial discharge signal.

2. The wavelet denoising method for partial discharge signals based on full-parameter spatial ergonomic optimization according to claim 1, characterized in that: In Step S4, the performing wavelet denoising on the noisy signal PDSignal specifically includes the following steps: Step S41: Wavelet decomposition; The noisy signal PDSignal is decomposed using the wavelet basis and decomposition level determined by the current loop; the approximation coefficients cA of the Level are obtained after decomposition. Level And the detail coefficients cD1, cD2, ..., cD for each level from level 1 to level 2. Level Among them, the detail coefficients include the high-frequency noise components of the signal and some abrupt change details; Step S42: Threshold estimation; Calculate the denoising threshold λ for each layer of detail coefficients; Step S43: Threshold quantization processing; Using the calculated denoising threshold λ, the detail coefficients cD at all levels are... j Soft thresholding is performed to remove noise components; the formula for calculating the detail coefficients after processing is as follows: ； Among them, New(cD) j ) represents the detail coefficients of the j-th level after processing; j represents the j-th level of the decomposition level; sgn() is the sign function; Step S44: Wavelet reconstruction; Using the retained approximation coefficients and the detail coefficients of each layer after soft thresholding, the signal is reconstructed using the inverse wavelet transform algorithm. The reconstructed signal is the current provisional denoised signal PDdenoise'.

3. The wavelet denoising method for partial discharge signals based on full-parameter spatial ergonomic optimization according to claim 2, characterized in that: In step S42, the denoising threshold λ is determined using a general threshold formula based on noise level estimation: First, the noise standard deviation σ is estimated using the first layer detail coefficients cD1: ； Where median() is the median function, which is the value in the middle position after sorting the elements of the sequence by numerical value. Subsequently, the denoising threshold λ is calculated based on the signal length N: 。 4. The wavelet denoising method for partial discharge signals based on full-parameter spatial ergonomic optimization according to claim 1, characterized in that: In step S5, the calculation process for the residual autocorrelation RAC' of the denoised partial discharge signal is as follows: Step S51: Obtain the residual sequence; First, define the residual sequence r: ； Step S52: Mean removal; Eliminate the DC component of the residual sequence r to ensure that subsequent correlation calculations only reflect the fluctuation characteristics of the signal: ； Where, μ r =mean(r) is the mean of the residual sequence, mean() is the mean function, and r' is the residual sequence after eliminating the DC component; Step S53: Calculate the normalized autocorrelation; The lagged first-order correlation coefficient of the mean-residual sequence is calculated using the sample autocorrelation function. To eliminate the influence of signal energy differences on the index, normalization is adopted. The calculation formula is defined as follows: ； Where R is the correlation coefficient and i is the index of the residual sequence r'; Index definition: The absolute value of the first-order lag coefficient is taken as the final evaluation index, residual autocorrelation RAC', that is: ； If RAC' = 0, it indicates that the adjacent points of the residual sequence are uncorrelated, which is consistent with the statistical characteristics of white noise. This means that the algorithm removes noise without destroying the original signal structure. If RAC' is significantly greater than 0, it indicates that there are signal components with deterministic patterns remaining in the residuals, reflecting that the algorithm has an over-denoising problem.

Images