Power data denoising method based on optimal wavelet base and improved wavelet threshold function

Abstract

The application provides a DC power data denoising method based on an optimal wavelet base and an improved wavelet threshold function. First, the optimal decomposition layer number of a sampling signal under different wavelet bases is determined according to noise power and noise power difference, and on this basis, the signal-to-noise ratio, the root mean square error and the cross correlation are used as evaluation indexes to determine the optimal wavelet base. Then, the optimal wavelet base is selected to perform optimal layer number decomposition on the sampling signal to obtain a group of wavelet coefficients. The improved threshold function and the threshold selection criterion in the application are used to perform threshold quantization processing on the detail coefficients of each layer and the approximation coefficients of the last layer. Finally, the residual coefficients are subjected to inverse wavelet transformation to obtain a denoising signal. Experiments prove that the optimal wavelet base and the decomposition layer number selection method are feasible, and the denoising effect of the application is better than that of the traditional threshold function, and the application is more conducive to retaining signal characteristics, can better filter out the noise components in the DC power signal, and improves the power metering accuracy.

Description

TECHNICAL FIELD

[0001] The application belongs to the field of electric energy metering and signal processing, and particularly relates to an electric energy data denoising method based on an optimal wavelet base and an improved wavelet threshold function. BACKGROUND

[0002] As the core function of an electric energy meter, the electric energy metering accuracy has always been the focus of engineers. At present, domestic direct-current meters are often used on direct-current charging piles (fast charging). Most direct-current charging piles obtain direct current by rectifying and filtering alternating current. However, under the current technical conditions, harmonics cannot be completely filtered out, so the output direct-current signal is mostly pulsating direct current. The unconverted alternating component superimposed on the electric energy data is called ripple. It is the existence of ripple and other pulsating noise that causes distortion of the direct-current voltage or current waveform, thereby affecting the accuracy of electric energy metering.

[0003] In terms of hardware, LC filter circuits are usually used to reduce ripple and pulsating noise interference in electric energy data signals, but will introduce leakage current, thereby reducing the accuracy of electric energy metering. In terms of digital filtering, wavelet transform is often used for denoising of direct-current noisy signals.

[0004] A cable partial discharge signal adaptive wavelet denoising method based on wavelet entropy and sparsity is disclosed in Chinese Patent No. CN112380934A, published on February 19, 2021. The method includes: collecting a cable partial discharge signal s(n) to be denoised and establishing a wavelet library; selecting an optimal base wavelet and an optimal decomposition scale for wavelet decomposition of s(n) according to the wavelet entropy and sparsity of s(n); wavelet decomposing s(n) to obtain approximation coefficients αJ and detail coefficients dj at each decomposition scale j; calculating the wavelet threshold thrj of the jth layer of wavelet coefficients; threshold quantizing the detail coefficients dj of the jth layer to filter out the wavelet coefficients with absolute values less than the wavelet threshold thrj and weaken the wavelet coefficients with absolute values greater than the wavelet threshold thrj, thereby obtaining the thresholded detail coefficients dj' of the jth layer; and reconstructing the approximation coefficients αJ and the thresholded detail coefficients dj' using a discrete wavelet inverse transform to obtain the denoised cable partial discharge signal s(n)',. The application improves the denoising ability of electric signals by selecting an optimal wavelet base and an optimal decomposition level. However, the selected threshold function will filter out the wavelet coefficients below the threshold, which will cause information loss of the electric signal.

[0005] A wavelet speech denoising method based on a new threshold function is disclosed in Chinese Patent Publication No. CN107274908A, published on October 20, 2017. The main steps of the method include: 1) noisy speech generation; 2) wavelet decomposition of the noisy speech; 3) wavelet threshold processing of the noisy speech; 4) wavelet reconstruction of the processed speech; and 5) obtaining the denoised speech signal. The step 3) is mainly processed, which directly determines the denoising effect. The step 3) is the threshold processing of the wavelet coefficients after the wavelet decomposition of the noisy speech, mainly involving the determination of the threshold and the selection of the threshold function. The new threshold function used in the present application alleviates the problems of signal oscillation caused by the discontinuity of the traditional hard threshold function and large distortion of the processed signal caused by the soft threshold function. However, the threshold value in the new threshold function is a fixed value, and the sparsity of the small coefficient noise of different detail coefficients in the denoising signal is different. If the fixed threshold value is set too small, the denoising will not be complete, and if it is set too large, the information of the signal will be lost. SUMMARY

[0006] To solve the above problems, the present application discloses an electric energy data denoising method based on an optimal wavelet basis and an improved wavelet threshold function.

[0007] To achieve the above effects, the technical scheme of the present application is as follows:

[0008] The electric energy data denoising method based on the optimal wavelet basis and the improved wavelet threshold function comprises the following steps:

[0009] Step 1: sampling the electric energy data at a sampling frequency f s The electric energy data is sampled, denoted as a source signal s(n), where s(n) = x(n) + noise(n), x(n) is a pure signal, and noise(n) is a noise signal.

[0010] Step 2: calculating the optimal decomposition layer number of the wavelet basis according to the evaluation index A plurality of

[0011] Step 3: calculating the denoising performance of multiple wavelet bases under the optimal decomposition layer number to obtain the optimal wavelet basis with the best denoising performance

[0012] Step 4: selecting the optimal wavelet basis according to the optimal decomposition layer to decompose the source signal, obtaining the detail coefficient cD j of each decomposition layer and the approximation coefficient cA j of the maximum decomposition layer

[0013] Step 5: calculating the local threshold thr j of the detail coefficient of each decomposition layer and estimating the threshold value of the approximation coefficient of the maximum decomposition layer

[0014] ​Step 6: Generate a wavelet threshold function based on the local threshold and the approximation coefficient threshold. Generating The wavelet threshold function compresses the wavelet coefficients of the source signal;

[0015] Step 7: After processing the source signal with the wavelet threshold function, perform inverse wavelet transform to obtain the reconstructed source signal, thus completing the noise reduction processing of the power data.

[0016] Preferably, in step 2, the evaluation index includes noise power P. noise and noise power difference ΔP i The noise power P noise and noise power difference ΔP i The formulas are expressed as follows:

[0017]

[0018]

[0019] Where N is the length of the signal sampling points, y i (n) is the denoised signal, i is the wavelet basis number, and x(n) = y0(n) is the clean signal.

[0020] Preferably, step 2 includes the following steps;

[0021] Step 21: Calculate the noise power of the same wavelet basis at multiple decomposition levels;

[0022] Step 22: After sorting the noise power, calculate the noise power difference between adjacent noise powers;

[0023] Step 23: Select the minimum noise power difference min(Δp) i ) and the smaller noise power involved in the corresponding calculations;

[0024] Step 24: Select the decomposition layer with the lower noise power as the optimal decomposition layer;

[0025] Step 25: Repeat steps 21 to 24 to determine the optimal decomposition layer for the remaining wavelet basis.

[0026] Preferably, step 3 includes the following steps:

[0027] Step 31: Evaluate the performance evaluation index of multiple wavelet bases under the optimal decomposition level and obtain the performance index data;

[0028] Step 32: Perform positive transformation and standardization on the performance index data to obtain adjusted performance index data;

[0029] Step 33: Calculate the information entropy of the adjusted performance index data;

[0030] Step 34: Calculate the weight of each performance evaluation index according to information entropy;

[0031] Step 35: Evaluate the denoising performance according to the weight, and determine the optimal wavelet base.

[0032] As a preferred embodiment, the performance evaluation index comprises the root mean square error and signal-to-noise ratio of the denoising data in the denoising process, and the cross-correlation between the denoising data and the original data.

[0033] As a preferred embodiment, the local threshold function in step 5 is represented as thr j :

[0034]

[0035] wherein j∈Z, N j is the data length of each decomposition layer coefficient, m is a threshold adjustment coefficient, is the noise variance, wherein w j，k = {cD 1，k cD 2，k …cD j，k cA j，k is the kth wavelet coefficient of the maximum wavelet entropy subband in the jth layer.

[0036] As a preferred embodiment, the formula of the threshold adjustment coefficient m is respectively:

[0037]

[0038] wherein, w j，k is the kth wavelet coefficient of the maximum wavelet entropy subband in the jth layer.

[0039] As a preferred embodiment, in step 5, the peak-to-average ratio of the approximation coefficient of the maximum decomposition layer is calculated, the peak-to-average ratio of the detail coefficient closest to the peak-to-average ratio of the approximation coefficient is determined, and the threshold value of the detail coefficient where the peak-to-average ratio of the detail coefficient is located is taken as the threshold value of the approximation coefficient.

[0040] As a preferred embodiment, the expression of the wavelet threshold function generated in step 6 is:

[0041]

[0042] wherein is the wavelet coefficient after threshold function processing, sgn(·) is a sign function, and α and τ are both adjustment factors, and α satisfies the first-order continuous derivative.

[0043] As a preferred embodiment, the expression of the adjustment factor α is:

[0044]

[0045] wherein, α and τ are both adjustment factors, and the local threshold function is represented as thr j .

[0046] The present application has the following advantages:

[0047] 1. The optimal decomposition layer of each wavelet base is determined by using noise power and noise power difference, and the optimal wavelet base is obtained according to the performance index evaluation and weight calculation of each wavelet base, thereby providing a method for selecting optimal wavelet base and optimal decomposition layer number.

[0048] 2. An automatic adjustment coefficient m is introduced into the local threshold function to dynamically adjust the threshold of the local function, and the appropriate threshold is automatically set according to the sparsity of small noise in different detail coefficients, compared with the general threshold (Sqtwolog criterion), the threshold in the present application can effectively avoid the choking phenomenon and improve the noise reduction performance.

[0049] 3. The threshold of the approximation coefficient is set to reduce the low-frequency noise in the approximation coefficient, and the threshold of the approximation coefficient is determined as the function threshold corresponding to the peak and ratio of the approximation coefficient and the closest detail coefficient.

[0050] 4. In view of the problems that the soft threshold in the traditional threshold function has constant error and the hard threshold causes the reconstructed signal to oscillate at the threshold, the wavelet threshold function generated by the present application is continuously derivable and has as the asymptote, which avoids the existence of constant error and discontinuity, further suppresses the useless components in the signal and enhances the useful part of the signal, in addition, by changing the size of the adjustment factor τ, the compression degree of the wavelet threshold function to the coefficient can be changed, and the effective components in the detail components can be effectively preserved. BRIEF DESCRIPTION OF DRAWINGS

[0051] Figure 1 The flow chart for determining the optimal wavelet base and the optimal decomposition layer number in the present application is shown;

[0052] Figure 2 The noise power graph of the Heavy Sine noise-containing signal under each decomposition layer of the wavelet base sym3 is shown;

[0053] Figure 3 The noise-reduced signal obtained by decomposing the Heavy Sine noise-containing signal by one layer of the wavelet base sym3 is shown;

[0054] Figure 4 The adjacent noise power difference graph of the Heavy Sine noise-containing signal under each decomposition layer of the wavelet base sym3 is shown;

[0055] Figure 5Fig. 1 shows the denoising performance score graph of multiple wavelet bases under the respective optimal decomposition level (Heavy Sine noise signal);

[0056] Figure 6 Fig. 4 shows the comparison graph of the processing results of wavelet coefficients under soft threshold, hard threshold and improved wavelet threshold function;

[0057] Figure 7 Fig. 5 shows the comparison graph of the denoising results of Heavy Sine noise signal under soft threshold, semi-soft threshold, hard threshold, Garrote threshold and improved wavelet threshold function;

[0058] Figure 8 Fig. 6 shows the comparison graph of the denoising results of DC voltage signal under soft threshold, semi-soft threshold, hard threshold, Garrote threshold and improved wavelet threshold function;

[0059] Figure 9 Fig. 7 shows the comparison graph of the denoising results of DC current signal under soft threshold, semi-soft threshold, hard threshold, Garrote threshold and improved wavelet threshold function. DETAILED DESCRIPTION

[0060] The present application will be further illustrated in combination with the accompanying drawings and specific implementation steps.

[0061] The power data denoising method based on optimal wavelet base and improved wavelet threshold function of the present application comprises the following steps:

[0062] Step 1: sampling the power data signal at a sampling frequency f s The power data signal is sampled and recorded as a source signal s(n), where s(n) = x(n) + noise(n), x(n) is the pure signal and noise(n) is the noise signal;

[0063] Step 2: calculating the denoising performance of multiple wavelet bases under the optimal decomposition level to obtain the optimal wavelet base with the best denoising performance, which comprises the following steps:

[0064] Step 21: calculating the noise power of the same wavelet base under multiple decomposition levels;

[0065] Step 22: calculating the noise power difference of adjacent noise powers after sorting the noise powers;

[0066] Step 23: selecting the minimum noise power difference min(Δp i ) and the smaller noise power involved in the corresponding calculation;

[0067] Step 24: taking the decomposition layer with the smaller noise power as the optimal decomposition layer;

[0068] Step 25: Repeat steps 21 to 24 to determine the optimal decomposition layer of the remaining wavelet bases.

[0069] In step 21, the noise power is taken as the first layer evaluation index, and the greater the noise power value, the better the noise reduction effect, as shown in Figure 2 .

[0070] Considering the existence of distortion, the noise power difference is taken as the second layer evaluation index. After sorting the noise power, the embodiment of the application is sorted in ascending order, and the adjacent noise power difference is calculated according to the noise power difference formula. The optimal decomposition layer of the wavelet base is the decomposition layer represented by the smaller power in the related calculation of the minimum noise power difference min(Δp i ).

[0071] The noise power P noise and the noise power difference ΔP i formula is as follows:

[0072]

[0073]

[0074] Where N is the length of the signal sampling point, y i (n) is the denoised signal, y0(n) = x(n) is the pure signal, and i is the wavelet base number, i ∈ Z.

[0075] Step 3: Calculate the denoising performance of the plurality of wavelet bases under the optimal decomposition layer to obtain the optimal wavelet base with the best denoising performance.

[0076] Step 3 includes steps:

[0077] Step 31: Evaluate the performance evaluation index of the plurality of wavelet bases under the optimal decomposition layer to obtain performance index data;

[0078] Step 32: Perform forward and standardization processing on the performance index data to obtain adjusted performance index data;

[0079] Step 33: Calculate the information entropy of the adjusted performance index data;

[0080] Step 34: Calculate the weight of each performance evaluation index according to the information entropy;

[0081] Step 35: Evaluate the denoising performance according to the weight to determine the optimal wavelet base.

[0082] The performance evaluation index includes the root mean square error and the signal-to-noise ratio of the denoising data in the denoising process, and the cross-correlation between the denoising data and the original data.

[0083] The signal-to-noise ratio SNR formula is represented as:

[0084]

[0085] where y i (n) is the reconstructed (denoised) signal, x(n) is the pure signal, is the power of the pure signal, is the noise power. The larger the SNR value, the better the denoising effect.

[0086] The formula of the root mean square error (RMSE) is:

[0087]

[0088] N is the number of sampling points, is the noise power. The smaller the RMSE, the better the denoising effect.

[0089] The formula for the forward processing of performance index data processing The range method formula r i，j and the information entropy E j , the weight ω i，j and the score Score i The formula is as follows:

[0090]

[0091]

[0092]

[0093] where

[0094]

[0095]

[0096] Step 4: Select the optimal wavelet basis according to the optimal decomposition layer to perform Mallat decomposition on the source signal, and use the appcoef function to extract the approximation coefficient cA j , and use the detcoef function to extract the detail coefficient cD j , where j∈Z.

[0097] Step 5: Calculate the local threshold value thr j of the detail coefficient of each decomposition layer, whose formula is as follows:

[0098]

[0099] where j is the layer number, j∈Z. N j is the data length of the jth layer detail coefficient, is the noise variance, where w j，k = {cD 1，k ; cD 2，k ;... cD j，k ; cA j，k} is the kth wavelet coefficient of the largest wavelet entropy subband in the jth layer.

[0100] The threshold adjustment coefficient m = lg(S j -1 ) / 2, and the peak and ratio of the jth layer The threshold adjustment coefficient m is used to determine the sparsity of the detail coefficient, and the smaller the value, the more small coefficient noise exists in the detail coefficient, and vice versa, meaning that only a small amount of signal with large coefficient value exists. According to the above characteristics, the threshold adjustment coefficient m can dynamically adjust the local threshold, so that a large amount of small coefficient noise is included in the threshold range, and the wavelet threshold function generated by the present application is combined for coefficient compression, further improving the noise reduction effect.

[0101] Considering that the approximation coefficient can also have low-frequency noise, the approximation part needs to be denoised. Since wavelet threshold denoising depends on the sparsity of the wavelet coefficient and the selection of the threshold, the detail coefficient of the low-frequency noise is non-sparse, which limits the effectiveness of the noise threshold. Therefore, in order to better remove the low-frequency noise in the approximation coefficient, the present application calculates the peak and ratio of the approximation component of the last layer, and according to the signal characteristics and the selection of the wavelet, the approximation component of the jth layer can represent the proportion of low-frequency noise and signal coefficient in a relatively sparse manner, and generally the value of the signal coefficient is greater than that of the noise coefficient, so in this embodiment, the local threshold of the peak and ratio of the detail coefficient closest to the peak and ratio of the approximation coefficient is used as the threshold of the approximation coefficient by comparing the peak and ratio of the approximation coefficient with the peak and ratio of the detail coefficient.

[0102] Step 6: generate a wavelet threshold function according to the local threshold and the approximation coefficient threshold, compress the wavelet coefficients of the source signal according to the generated wavelet threshold function, reduce the noise signal, and retain the useful signal. The wavelet threshold function generated by the present application is used as an improved threshold function and is expressed as follows:

[0103]

[0104] In the formula, w is the wavelet coefficient after threshold function compression, w j，k is the wavelet coefficient before compression, sgn(·) is the sign function, and α, τ are adjustment factors, thr j is the local threshold. In order to simplify the number of parameters and make it satisfy the first-order continuous differentiability, there are

[0105] The wavelet threshold function generated by the present application has the following properties:

[0106]

[0107]

[0108]

[0109] The wavelet threshold function generated by the application is asymptotic to i，j as The constant bias caused by the soft threshold function is effectively overcome, and the threshold function is continuous on the positive half-axis, and it can be proved that it is also continuous on the negative half-axis, effectively avoiding oscillation near the discontinuity point during signal reconstruction.

[0110] The processing effect of the soft threshold, hard threshold and improved threshold function on the wavelet coefficients is shown in Figure 6 By changing the value of the adjustment coefficient τ, the shape of the threshold function on both sides of the threshold can be dynamically adjusted. When τ is smaller, the coefficients greater than the threshold converge less, and the coefficients less than the threshold are not directly set to zero, which is beneficial to retaining the high-frequency characteristics of the signal. When τ is larger, the curve less than the threshold converges to 0 faster, and the curve greater than the threshold converges to which is beneficial to filtering noise signals.

[0111] Step 7: After the source signal processed by the wavelet threshold function is inverse transformed, the reconstructed source signal is obtained, and the power data denoising processing is completed.

[0112] Referring to Figures 1 to 9 , the improved threshold function is verified using Matlab software, and the Heavy Sine signal with different intensity white Gaussian noise is processed by wavelet threshold denoising. The threshold functions selected for comparison are: soft threshold semi-soft threshold hard threshold and Garrote threshold function The formula is:

[0113]

[0114]

[0115]

[0116]

[0117] Example 1: The implementation conditions include that the pure signal is a Heavy Sine signal, the noise is a white noise signal, the signal-to-noise ratio is 20.39 dB, and the threshold rule is the local threshold thr j, the threshold function selects the threshold function generated by the application.

[0118] Table 1: Heavy Sine noise signal under different wavelet base sym3 layer noise reduction results comparison

[0119]

[0120] According to Table 1 and Figure 4 , 5 It can be seen that the best decomposition layer number of the noise signal in the wavelet base sym3 is 6.

[0121] Table 2: Heavy Sine noise signal under optimal layer number of multiple wavelet bases Noise reduction results comparison

[0122]

[0123] According to Table 2 and Figure 5 It can be seen that the Heavy Sine noise signal has the maximum noise reduction performance score in the wavelet base sym3, db3. At this time, db3 corresponds to the best decomposition layer number 6, and the SNR, RMSE and NCC values are the same as sym3, which indicates that the best wavelet base and the optimal decomposition layer number of the noise signal have two groups.

[0124] Example two: using five threshold functions to denoise the noise signal in experiment 1, threshold rule: Sqtwolog threshold, Heursure threshold, MiniMaxi threshold, local threshold thr of the application j , the adjustment coefficient τ is selected as 20, 10, 5, 3, 2, 1, 1, 1 in turn with the increase of the layer number.

[0125] Table 3: Noise reduction performance index of five threshold functions under multiple threshold rules

[0126]

[0127] From the above table, it can be seen that the generated wavelet threshold function of the application as an improved threshold function has greater improvement in the noise reduction effect of the Heavy Sine noise signal, and compared with the other four threshold functions, the SNR of the generated wavelet threshold function of the application is larger and the RMSE is smaller.

[0128] Figure 7 The comparison chart of denoising the noise signal in example one by soft threshold, semi-soft threshold, hard threshold, Garrote threshold and improved threshold function, threshold rule: local threshold thr of the application j .

[0129] It can be seen from the figure that the waveform after hard threshold function denoising is not smooth, especially at the turning point, noise cannot be filtered well. Soft threshold function denoising and signal are very smooth, but the signal details exist overkill phenomenon. The semi-soft threshold and Garrote threshold function retains the signal detail characteristics to some extent, the denoising signal smoothness is better than the hard threshold function, but it can be found that the denoising signal and the pure signal still exist certain gap. The generated wavelet threshold function of the application better inhibits the overkill phenomenon of the detail coefficient and signal oscillation, retains the signal characteristics to the greatest extent, and the denoising effect is remarkable.

[0130] Example three: using five threshold functions to denoise the direct current signal, the threshold rule is the local threshold thr of the application j , the adjustment coefficient τ is selected as 20, 10, 5, 3, 2, 1, 1, 1 in turn with the layer increasing.

[0131] Figure 8 、 9 It is respectively the comparison of the denoising results of the direct current voltage and current signal using multiple threshold functions.

[0132] It can be seen from the figure that the hard threshold function has general denoising effect on the pulsating noise and ripple signal in the direct current signal, and the waveform comparison before and after the direct current voltage denoising is not obvious. The soft threshold function has obvious denoising effect on the voltage signal, but sacrifices the detail information. The semi-soft threshold and Garrote threshold function have similar denoising effect, which is between the hard threshold and soft threshold function. The generated wavelet threshold function in the application effectively filters the noise component in the direct current voltage and current signal, and better retains the source signal characteristics, which is beneficial to the measurement of electric energy data.

[0133] The above is the specific implementation method and simulation verification given by the application. According to the results of example one and example two, the feasibility of selecting the optimal wavelet basis and decomposition layer method is indicated; according to example two and example three, it is indicated that the improved threshold function can effectively retain the signal characteristics while denoising to the greatest extent, and effectively improve the electric energy data measurement accuracy.

Claims

1. A method for denoising power data based on an optimal wavelet basis and an improved wavelet threshold function, characterized in that, Including the following steps: Step 1: Using a sampling frequency f s The electrical energy data is sampled and denoted as the source signal s(n). The source signal s(n) = x(n) + noise(n), where x(n) is the clean signal and noise(n) is the noise signal; Step 2: Calculate the optimal decomposition level for multiple wavelet bases based on the evaluation criteria; Step 3: Calculate the denoising performance of multiple wavelet bases under the optimal decomposition level, and obtain the optimal wavelet base with the best denoising performance; Step 4: Select the optimal wavelet basis according to the optimal decomposition layer to decompose the source signal and obtain the detail coefficients cD of each decomposition layer. j The approximation coefficient cA of the maximum decomposition layer j ; Step 5: Calculate the local threshold thr for the detail coefficients of each decomposition layer. j And estimate the approximate coefficient threshold of the maximum decomposition layer; Step 6: Generate a wavelet threshold function based on the local threshold and the approximation coefficient threshold, and compress the wavelet coefficients of the source signal based on the generated wavelet threshold function; Step 7: After processing the source signal with the wavelet threshold function, perform inverse wavelet transform to obtain the reconstructed source signal, and the power data noise reduction process is completed; The wavelet threshold function expression generated in step 6 is: In the formula The wavelet coefficients are processed by the threshold function, sgn(·) is the sign function, α and τ are both adjustment factors, and α satisfies first-order continuous differentiability; The expression for the adjustment factor α is: Where α and τ are both adjustment factors, and the local threshold function represents thr j .

2. The power data denoising method based on optimal wavelet basis and improved wavelet threshold function according to claim 1, characterized in that, In step 2, the evaluation index includes noise power P. noise and noise power difference ΔP i The noise power P noise and noise power difference ΔP i The formulas are expressed as follows: Where N is the length of the signal sampling points, y i (n) is the denoised signal, i is the wavelet basis number, and x(n) = y0(n) is the clean signal.

3. The power data denoising method based on the optimal wavelet basis and improved wavelet threshold function according to claim 2, characterized in that, Step 2 includes the following steps; Step 21: Calculate the noise power of the same wavelet basis at multiple decomposition levels; Step 22: After sorting the noise power, calculate the noise power difference between adjacent noise powers; Step 23: Select the minimum noise power difference min(Δp) i ) and the smaller noise power involved in the corresponding calculations; Step 24: Select the decomposition layer with the lower noise power as the optimal decomposition layer; Step 25: Repeat steps 21 to 24 to determine the optimal decomposition layer for the remaining wavelet basis.

4. The power data denoising method based on optimal wavelet basis and improved wavelet threshold function according to claim 1, characterized in that, Step 3 includes the following steps: Step 31: Evaluate the performance evaluation index of multiple wavelet bases under the optimal decomposition level and obtain the performance index data; Step 32: Perform positive transformation and standardization on the performance index data to obtain adjusted performance index data; Step 33: Calculate the information entropy of the adjusted performance index data; Step 34: Calculate the weights of each performance evaluation index based on information entropy; Step 35: Evaluate the denoising performance based on the weights and determine the optimal wavelet basis.

5. The power data denoising method based on the optimal wavelet basis and improved wavelet threshold function according to claim 4, characterized in that, The performance evaluation metrics include the root mean square error and signal-to-noise ratio of the denoised data during the denoising process, as well as the cross-correlation between the denoised data and the original data.

6. The power data denoising method based on optimal wavelet basis and improved wavelet threshold function according to claim 1, characterized in that, The local threshold function described in step 5 represents thr j for: In the formula, j∈Z, N j Where m is the data length of the coefficients in each decomposition layer, and m is the threshold adjustment coefficient. Let w be the noise variance. j,k ={cD 1,k cD 2,k … cD j,k cA j,k } represents the k-th wavelet coefficient of the largest wavelet entropy subband in the j-th layer.

7. The power data denoising method based on optimal wavelet basis and improved wavelet threshold function according to claim 6, characterized in that, The formulas for the threshold adjustment coefficient m are as follows: in, w j,k It represents the k-th wavelet coefficient of the largest wavelet entropy subband in the j-th layer.

8. The power data denoising method based on the optimal wavelet basis and improved wavelet threshold function according to claim 7, characterized in that, In step 5, the peak-to-peak ratio of the approximation coefficients of the maximum decomposition layer is calculated, the peak-to-peak ratio of the detail coefficients that are closest to the peak-to-peak ratio of the approximation coefficients is determined, and the threshold of the detail coefficients in which the peak-to-peak ratio of the closest detail coefficients is located is used as the approximation coefficient threshold.

Images