Transformer leakage magnetic field signal measurement method based on improved SVMD combined wavelet threshold method
Through the improved combined method of SVMD and wavelet thresholding method, the transformer leakage magnetic field signal is denoised, which solves the problem of reducing diagnostic accuracy caused by noise interference, and achieves higher fault diagnosis accuracy and signal quality.
Patent Information
- Application Number
- CN202510255888.1
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2025-03-05
- Publication Date
- 2025-05-09
AI Technical Summary
The prior art has problems with noise and harmonic interference in transformer winding fault diagnosis, resulting in reduced diagnostic accuracy.
The combined method of improved SVMD (continuous variational modal decomposition) and wavelet threshold method is used to denoise the transformer leakage magnetic field signal. The SVMD parameters are determined by the NRBO optimization algorithm, and the remaining noise in the signal is further processed in combination with the wavelet threshold method.
Effectively remove noise interference, extract key features of leakage magnetic field signals, and improve the accuracy and signal quality of transformer winding fault diagnosis.
Smart Images
Figure CN119959833A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of transformer leakage magnetic field detection in an electric power system, and in particular relates to a transformer leakage magnetic field signal measurement method based on an improved SVMD combined with a wavelet threshold method. Background Art
[0002] With the gradual development of new domestic power systems and ultra-high voltage power grids, the operating environment of transformers has become more diverse and complex. During the operation of the transformer, the impact of short-circuit electromotive force on the winding may cause permanent deformation such as displacement, depression, and bulging of the winding, which seriously damages the inter-turn insulation of the winding and increases the probability of winding failure. At present, transformer fault detection methods are mainly divided into offline detection methods and online detection methods. Offline detection methods such as short-circuit reactance method and frequency response method are relatively mature in technology, but they need to be carried out during transformer maintenance, and cannot monitor and detect faults in real time; in the online detection method, the vibration method is difficult to form a clear criterion to measure the degree of fault, and the parameter identification method has problems such as low parameter calculation accuracy and unclear fault relationship. The above methods have shortcomings in transformer winding fault identification. Studies have shown that when there is a hidden fault in the transformer winding, the leakage magnetic field distribution around the winding will change and show a certain regularity, which provides a new research approach for diagnosing transformer winding faults by using the change of leakage magnetic distribution.
[0003] Although the transformer leakage magnetic field signal can reflect the operating status of the transformer to a large extent, these analyses are usually based on ideal conditions and do not fully consider the noise and harmonic interference that may exist in the actual measurement process. Due to the presence of these interference signals, the direct use of unprocessed leakage magnetic signals for fault diagnosis will significantly reduce the accuracy of diagnosis. Summary of the invention
[0004] The purpose of the present invention is to provide a transformer leakage magnetic field signal measurement method based on an improved SVMD combined with a wavelet threshold method. Efficient denoising of the actually measured leakage magnetic field signal is the key to improving the accuracy of fault diagnosis. The present invention combines the advantages of an improved continuous variational mode decomposition (SVMD) and a wavelet threshold method, can effectively remove noise interference, extract key features, and provide a reliable basis for accurately diagnosing transformer winding faults.
[0005] The technical solution of the present invention is as follows: A transformer leakage magnetic field signal measurement method based on an improved SVMD combined with a wavelet threshold method comprises the following steps:
[0006] Step 1: Determine the SVMD continuous variational mode decomposition parameters through NRBO;
[0007] Step 2: Based on step 1, perform SVMD continuous variation mode decomposition on the magnetic flux leakage signal measured by the magneto-optical sensor;
[0008] Step 3: Combined with wavelet threshold method, further process the remaining noise in the signal.
[0009] In step 1, NRSR calculates the gradient and second-order information of the fitness function. The fitness function uses the minimum fuzzy entropy as the optimization objective function and calculates it in each iteration. The fuzzy entropy as the fitness function can measure the fuzziness of the signal decomposition result. The minimized entropy value represents the optimal quality of the decomposition. In each iteration, NRBO adjusts α according to the current fitness function value through NRSR. max , thereby improving the accuracy of the decomposition results. Finally, through the iterative optimization of NRBO and the minimum fuzzy entropy fitness function, the optimal α is obtained. max value to ensure the best quality of SVMD decomposition results and improve the accuracy and robustness of signal processing.
[0010] In step 1, fuzzy entropy combines sample entropy and fuzzy membership function to describe the complexity of the signal. By introducing the fuzzy membership function, a more accurate signal description is provided. Assuming that the transformer winding leakage signal measured by the magneto-optical sensor is an E-dimensional time series {l(t)=l(1),l(2),…,l(E)}, after spatial reconstruction, the time series D(i) is obtained:
[0011] D(i)=[l(i),l(i+1),...l(i+a-1)]-l0(i) (1)
[0012]
[0013] Where: i = 1, 2, ..., E-a + 1, a is the embedding dimension; i, j are used to represent the index offset of the time series; l0(i) is the mean;
[0014] Then the distance d between two time series D(i) and D(j) is defined as follows, taking the maximum difference between the two elements;
[0015]
[0016] Where: d is the distance between time series D(i) and D(j); g is used to represent the index offset of the time series.
[0017] The fuzzy membership function can be used to calculate the similarity between sequences D(i) and D(j):
[0018]
[0019] Where: p is the similarity tolerance, which is generally 0.15 times the standard deviation of the original signal of the leakage magnetic field; i, j = 1, 2, ..., E-a + 1, and i ≠ j;
[0020]
[0021] Where: a (p) is the measure of fuzzy similarity of time series at scale a,
[0022] Then the expression of fuzzy entropy FuzzyEn(a,p,E) is:
[0023] FuzzyEn(a,p,E)=lnψ a (p)-lnψ a+1 (p) (6)
[0024] After calculating the fuzzy entropy, use this entropy value to guide the maximum penalty factor α max The optimization process, through the NRBO optimization algorithm, in each iteration, first according to the current α max Calculate the decomposition result, then calculate the corresponding fuzzy entropy value, and use this value as the fitness function to adjust α max In each iteration, NRBO updates α according to the current fuzzy entropy value and NRSR max The fuzzy entropy gradually decreases and finally converges to the minimum value.
[0025] In step 2, based on step 1, the leakage magnetic signal measured by the magneto-optical sensor is subjected to SVMD continuous variation mode decomposition: α is obtained by optimization. max The leakage magnetic signal l(t) measured by the magneto-optical sensor is subjected to SVMD continuous variation mode decomposition.
[0026] In step 2, the SVMD continuous variational mode decomposition divides l(t) into several mode functions. In each iteration, SVMD updates the mode function by minimizing the difference between l(t) and the mode function, and repeats the process until convergence. The final mode function is used to reconstruct the original signal.
[0027] In step 2, some constraints are added during the process of extracting the modal function to prevent the model from converging to the already extracted mode. The constraint model is as follows:
[0028]
[0029] Where: J1, J2, and J3 are three constraint criteria; α is the parameter that balances J1, J2, and J3, obtained by Lagrange multiplication, and its value must be less than α max ;u M is the Mth modal component of SVMD decomposition; ωM is the center frequency of the Mth mode; l r is the residual signal, that is, M Input signals other than
[0030] The augmented Lagrangian function is established by introducing a combination of a quadratic penalty coefficient and a Lagrangian multiplier λ, as follows:
[0031]
[0032] As with VMD, the minimization problem in formula (1) is iteratively solved by Parseval equation and alternating direction multiplier method, as follows:
[0033]
[0034] Where: is the spectrum of the mode; l(ω) is the spectrum of the signal; n is the number of iterations;
[0035] Formula (5) is used to update the Lagrange multiplier:
[0036]
[0037] During the iteration process, the center frequency and bandwidth of the eigenmode function are continuously updated until the following conditions are met:
[0038]
[0039] Where: is the nth update of the Mth mode; T is the dual rise time step; ε1 and ε2 are tolerances; σ 2 is the noise variance.
[0040] In step 3, the wavelet threshold method is combined to further process the residual noise in the signal, and the modal components and residual signals obtained by SVMD decomposition in step 2 are decomposed into r (t) is subjected to wavelet transform to obtain wavelet coefficients. Wavelet transform can effectively decompose the signal into different frequency bands. The soft threshold method is used for noise suppression, and its expression is as follows:
[0041]
[0042] Where: sgn is the sign function; β is the change scale; θ is the threshold parameter; w β,η is the original wavelet coefficient of the signal; w′ β,η is the wavelet coefficient after signal processing;
[0043] Among them, the threshold parameter θ is expressed as follows:
[0044]
[0045]
[0046] Where: μ is the noise standard deviation; N is the signal length; median is the median absolute deviation; w 1,η is the nth wavelet coefficient in the first layer.
[0047] The beneficial effect of the present invention is that: through the NRBO optimization algorithm, in each iteration, first according to the current α max Calculate the decomposition result, then calculate the corresponding fuzzy entropy value, and use this value as the fitness function to adjust α max In each iteration, NRBO updates α according to the current fuzzy entropy value and NRSR. max The fuzzy entropy is gradually reduced and finally converges to the minimum value. In this process, trap avoidance operation (TAO) helps the algorithm avoid local optimal solutions, thereby ensuring that the global optimal α can be found. max By continuously optimizing the fuzzy entropy, NRBO will eventually determine the best α max , which makes the modal decomposition of the signal optimal; after iteration and optimization of the SVMD algorithm, several modal components u are finally obtained M (t), these modal components represent the independent components of the leakage magnetic signal l(t) in different frequency ranges. Each modal function u M (t) After optimization and constraint processing, the signal characteristics can be accurately represented, which is helpful for more in-depth analysis and fault diagnosis of the leakage magnetic signal of the transformer winding. After the wavelet coefficient threshold processing, the processed coefficients are restored to the time domain signal using the inverse wavelet transform. After the wavelet threshold method processing, the final signal has higher accuracy and less noise than the original signal, which is suitable for further analysis and fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS
[0048] Figure 1 A flow chart of a transformer leakage magnetic field signal measurement method based on an improved SVMD combined with wavelet threshold method provided by the present invention;
[0049] Figure 2 Determining the SVMD parameter α for NRBO max flow chart. DETAILED DESCRIPTION
[0050] The present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0051] A transformer leakage magnetic field signal measurement method based on an improved SVMD combined with a wavelet threshold method comprises the following steps:
[0052] Step 1: Determine the SVMD continuous variational mode decomposition parameters through NRBO;
[0053] In the step 1, the SVMD continuous variational mode decomposition parameters are determined by NRBO. Continuous variational mode decomposition (SVMD) is a signal decomposition method that aims to extract different intrinsic mode functions (IMF) from complex signals. Through IMF decomposition, high-frequency components containing noise and low-frequency components containing useful signals can be separated. Removing high-frequency noise components and retaining important signal components help improve the quality of signal processing. Compared with traditional decomposition methods, SVMD performs well in processing nonlinear and non-stationary signals. It is an extension based on the variational mode decomposition (VMD) method. By introducing a serialized variational optimization strategy, different modes in the signal are gradually extracted.
[0054] In VMD theory, the decomposition result is mainly affected by the number of decompositions k and the quadratic penalty factor α. However, in SVMD theory, it is not necessary to know the total number of modes k in the signal in advance. Therefore, in SVMD, the most critical parameter is the maximum penalty factor α max .
[0055] When α max When α is set too high, the regularization penalty is too strong, which will lead to a large number of erroneous modes in the decomposition results. max When it is set too low, the regularization penalty is insufficient, which may cause multiple true modes to be mistakenly merged into one or several modes, which is the mode mixing problem.
[0056] In the traditional method, α is determined max Usually rely on cross-validation and trial and error. max , and observe the effect of decomposition, and then make adjustments based on the results. Although these methods are effective, the process is cumbersome and requires a lot of experimental debugging, and is easily affected by human factors. In response to the above problems, this paper uses NRBO to optimize α max .
[0057] NRBO is a search algorithm inspired by the Newton-Raphson method. It uses the Newton-Raphson search rule (NRSR) and the trap avoidance operator (TAO) and several sets of matrices to explore the entire search process to obtain the best results. NRSR uses the Newton-Raphson method to enhance the exploration ability of NRBO and speed up the convergence speed; TAO helps NRBO avoid local optimal problems.
[0058] Specifically, NRSR enhances the exploration capability of NRBO and accelerates the convergence speed by calculating the gradient and second-order information of the fitness function. The fitness function in this technology uses the minimum fuzzy entropy, which is used as the optimization objective function and calculated in each iteration. Fuzzy entropy as a fitness function can measure the fuzziness of the signal decomposition result, and minimizing the entropy value represents the optimal quality of the decomposition.
[0059] In each iteration, NRBO adjusts α according to the current fitness function value (minimum fuzzy entropy) through NRSR. max , thereby improving the accuracy of the decomposition results. max The resulting fitness function value is poor, NRBO will adjust α max , which improves the quality of modal decomposition. In order to avoid falling into the local optimum, the TAO operation helps NRBO to jump out of the local minimum and maintain the global search capability by adjusting the search path, so that α max Gradually optimize to the optimal value.
[0060] Finally, through the iterative optimization of NRBO and the minimum fuzzy entropy fitness function, the optimal α can be accurately obtained. max value to ensure the best quality of SVMD decomposition results and improve the accuracy and robustness of signal processing.
[0061] Fuzzy entropy combines sample entropy and fuzzy membership function to describe the complexity of the signal. By introducing the fuzzy membership function, fuzzy entropy can more comprehensively consider the uncertainty of the sample value, thereby providing a more accurate signal description. Assuming that the transformer winding leakage signal measured by the magneto-optical sensor is an E-dimensional time series {l(t) = l(1), l(2), ..., l(E)}, after spatial reconstruction, the time series D(i) is obtained:
[0062] D(i)=[l(i),l(i+1),…l(i+a-1)]-l0(i) (1)
[0063]
[0064] Where: i = 1, 2,…, E-a+1, a is the embedding dimension; i, j are used to represent the index offset of the time series; l0(i) is the mean.
[0065] Then the distance d between two time series D(i) and D(j) is defined as follows, taking the maximum difference between the elements of the two.
[0066]
[0067] Where: d is the distance between time series D(i) and D(j); g is used to represent the index offset of the time series.
[0068] The fuzzy membership function can be used to calculate the similarity between sequences D(i) and D(j):
[0069]
[0070] Where: p is the similarity tolerance, which is generally 0.15 times the standard deviation of the original signal of the leakage magnetic field;
[0071] i,j=1,2,…,E-a+1, and i≠j.
[0072]
[0073] Where: a (p) is a measure of the fuzzy similarity of time series at scale a.
[0074] Then the expression of fuzzy entropy FuzzyEn(a,p,E) is:
[0075] FuzzyEn(a,p,E)=lnψ a (p)-lnψ a+1 (p) (6)
[0076] After calculating the fuzzy entropy, the entropy value can be used to guide the maximum penalty factor α max The fuzzy entropy value measures the complexity and regularity of the signal after decomposition. The goal is to obtain a more accurate and ideal signal decomposition by minimizing the fuzzy entropy. Through the NRBO optimization algorithm, in each iteration, first according to the current α max Calculate the decomposition result, then calculate the corresponding fuzzy entropy value, and use this value as the fitness function to adjust α max In each iteration, NRBO updates α according to the current fuzzy entropy value and NRSR. max The fuzzy entropy is gradually reduced and finally converges to the minimum value. In this process, trap avoidance operation (TAO) helps the algorithm avoid local optimal solutions, thereby ensuring that the global optimal α can be found. max By continuously optimizing the fuzzy entropy, NRBO will eventually determine the best α max , so that the modal decomposition effect of the signal is optimal.
[0077] Step 2: Based on step 1, the leakage magnetic signal measured by the magneto-optical sensor is subjected to SVMD continuous variation mode decomposition; in summary, the SVMD parameter α is determined by the NRBO method. max The specific steps are as follows Figure 2 In step 2, based on step 1, the leakage magnetic signal measured by the magneto-optical sensor is subjected to SVMD continuous variation mode decomposition:
[0078] Based on step 1, we use optimization to get α max The magnetic flux leakage signal l(t) measured by the magneto-optical sensor is subjected to SVMD continuous variational mode decomposition. The SVMD algorithm divides l(t) into several mode functions through variational mode decomposition. In each iteration, SVMD updates the mode function by minimizing the difference between l(t) and the mode function. This process is repeated until convergence. The final mode function can be used to reconstruct the original signal.
[0079] Adding some constraints during the process of extracting the modal function can prevent the model from converging to the extracted modes. The constraint model is as follows:
[0080]
[0081] Where: J1, J2, and J3 are three constraint criteria; α is the parameter that balances J1, J2, and J3, obtained by Lagrange multiplication, and its value must be less than α max ;u M is the Mth modal component of SVMD decomposition; ω M is the center frequency of the Mth mode; l r is the residual signal, that is, M K is the total number of modal components, u M The Mth modal component.
[0082] To solve equation (1), the augmented Lagrangian function is established by introducing a combination of a quadratic penalty coefficient and a Lagrangian multiplier λ. It is as follows:
[0083]
[0084] As with VMD, the minimization problem in formula (1) is iteratively solved by Parseval equation and alternating direction multiplier method, as follows:
[0085]
[0086] Where: is the spectrum of the mode; l(ω) is the spectrum of the signal; n is the number of iterations. ω is the center frequency, and A is the intermediate variable.
[0087] Formula (5) is used to update the Lagrange multiplier:
[0088]
[0089] During the iteration process, the center frequency and bandwidth of the eigenmode function are continuously updated until the following conditions are met:
[0090]
[0091] Where: is the nth update of the Mth mode; T is the dual rise time step; ε1 and ε2 are tolerances; σ 2 is the noise variance.
[0092] After iteration and optimization of the SVMD algorithm, several modal components u are finally obtained. M (t), these modal components represent the independent components of the leakage magnetic signal l(t) in different frequency ranges. Each modal function u M (t) After optimization and constraint processing, the signal characteristics can be accurately represented, which is helpful for deeper analysis and fault diagnosis of the leakage magnetic signal of the transformer winding. However, after SVMD decomposition, there may still be some noise components that cannot be completely eliminated, especially in the low-frequency part. r (t) may contain unwanted noise. In order to further improve the quality of the signal, the wavelet threshold method will be introduced in step 3 to process the residual noise obtained after SVMD decomposition.
[0093] Step 3: Combined with wavelet threshold method, further processing the remaining noise in the signal. Combined with wavelet threshold method, further processing the remaining noise in the signal:
[0094] Although the use of NRBO to optimize the SVMD parameters can remove Gaussian white noise, some periodic interference or other non-Gaussian noise in a specific frequency range is not completely separated from the transformer leakage magnetic field signal. Therefore, by using the wavelet threshold method, the local features in the signal can be more effectively captured, the remaining noise can be processed, and the signal quality can be further improved.
[0095] The modal components and residual signal l obtained by SVMD decomposition in step 2 r (t) Perform wavelet transform to obtain wavelet coefficients. Wavelet transform can effectively decompose the signal into different frequency bands, so that the noise part of the signal can be separated more accurately. In order to better remove noise, the soft threshold method is used for noise suppression. Its expression is as follows:
[0096]
[0097] Where: sgn is the sign function; β is the change scale; θ is the threshold parameter; w β,η is the original wavelet coefficient of the signal; w′ β,η is the wavelet coefficient after signal processing.
[0098] Among them, the threshold parameter θ is expressed as follows:
[0099]
[0100] Where: μ is the noise standard deviation; N is the signal length; median is the median absolute deviation; w 1,η is the nth wavelet coefficient in the first layer.
[0101] After the wavelet coefficient threshold processing, the inverse wavelet transform is used to restore the processed coefficients to the time domain signal. This process can remove most of the noise components and effectively retain the important information in the signal, thereby improving the quality of the signal. After the wavelet threshold method is processed, the final signal has higher accuracy and less noise than the original signal, which is suitable for further analysis and fault diagnosis.
Claims
1. A transformer leakage magnetic field signal measurement method based on improved SVMD combined with wavelet threshold method, characterized in that: The following steps are involved: Step 1: Determine the SVMD continuous variational mode decomposition parameters through NRBO; Step 2: Based on step 1, perform SVMD continuous variation mode decomposition on the magnetic flux leakage signal measured by the magneto-optical sensor; Step 3: Combined with wavelet threshold method, further process the remaining noise in the signal.
2. The transformer leakage magnetic field signal measurement method based on the improved SVMD combined with wavelet threshold method as claimed in claim 1 is characterized in that: In step 1, NRSR calculates the gradient and second-order information of the fitness function. The fitness function uses the minimum fuzzy entropy as the optimization objective function and calculates it in each iteration. The fuzzy entropy as the fitness function can measure the fuzziness of the signal decomposition result. The minimized entropy value represents the optimal quality of the decomposition. In each iteration, NRBO adjusts α according to the current fitness function value through NRSR. max , thereby improving the accuracy of the decomposition results. Finally, through the iterative optimization of NRBO and the minimum fuzzy entropy fitness function, the optimal α is obtained. max value to ensure the best quality of SVMD decomposition results and improve the accuracy and robustness of signal processing.
3. The transformer leakage magnetic field signal measurement method based on the improved SVMD combined with wavelet threshold method as claimed in claim 2 is characterized in that: In step 1, fuzzy entropy combines sample entropy and fuzzy membership function to describe the complexity of the signal. By introducing the fuzzy membership function, a more accurate signal description is provided. Assuming that the transformer winding leakage signal measured by the magneto-optical sensor is an E-dimensional time series {l(t)=l(1),l(2),…,l(E)}, after spatial reconstruction, the time series D(i) is obtained: D(i)=[l(i),l(i+1),…l(i+a-1)]-l0(i) (1) Where: i = 1, 2, ..., E-a + 1, a is the embedding dimension; i, j are used to represent the index offset of the time series; l0(i) is the mean; Then the distance d between two time series D(i) and D(j) is defined as follows, taking the maximum difference between the two elements; Where: d is the distance between time series D(i) and D(j); g is used to represent the index offset of the time series. The fuzzy membership function can be used to calculate the similarity between sequences D(i) and D(j): Where: p is the similarity tolerance, which is generally 0.15 times the standard deviation of the original signal of the leakage magnetic field; i, j = 1, 2, ..., E-a + 1, and i ≠ j; Where: a (p) is the measure of fuzzy similarity of time series at scale a, Then the expression of fuzzy entropy FuzzyEn(a,p,E) is: FuzzyEn(a,p,E)=lnψ a (p)-lnψ a+1 (p) (6) After calculating the fuzzy entropy, use this entropy value to guide the maximum penalty factor α max The optimization process, through the NRBO optimization algorithm, in each iteration, first according to the current α max Calculate the decomposition result, then calculate the corresponding fuzzy entropy value, and use this value as the fitness function to adjust α max In each iteration, NRBO updates α according to the current fuzzy entropy value and NRSR max The fuzzy entropy gradually decreases and finally converges to the minimum value.
4. The transformer leakage magnetic field signal measurement method based on the improved SVMD combined with wavelet threshold method as claimed in claim 1, characterized in that: In step 2, based on step 1, the leakage magnetic signal measured by the magneto-optical sensor is subjected to SVMD continuous variation mode decomposition: α is obtained by optimization. max The leakage magnetic signal l(t) measured by the magneto-optical sensor is subjected to SVMD continuous variation mode decomposition.
5. The transformer leakage magnetic field signal measurement method based on the improved SVMD combined with wavelet threshold method as claimed in claim 4, characterized in that: In step 2, the SVMD continuous variational mode decomposition divides l(t) into several mode functions. In each iteration, SVMD updates the mode function by minimizing the difference between l(t) and the mode function, and repeats the process until convergence. The final mode function is used to reconstruct the original signal.
6. The transformer leakage magnetic field signal measurement method based on improved SVMD combined with wavelet threshold method as claimed in claim 5, characterized in that: In step 2, some constraints are added during the process of extracting the modal function to prevent the model from converging to the already extracted mode. The constraint model is as follows: Where: J1, J2, and J3 are three constraint criteria; α is the parameter that balances J1, J2, and J3, obtained by Lagrange multiplication, and its value must be less than α max ;u M is the Mth modal component of SVMD decomposition; ω M is the center frequency of the Mth mode; l r is the residual signal, that is, M Input signals other than The augmented Lagrangian function is established by introducing a combination of a quadratic penalty coefficient and a Lagrangian multiplier λ, as follows: As with VMD, the minimization problem in formula (1) is iteratively solved by Parseval equation and alternating direction multiplier method, as follows: Where: is the spectrum of the mode; l(ω) is the spectrum of the signal; n is the number of iterations; Formula (5) is used to update the Lagrange multiplier: During the iteration process, the center frequency and bandwidth of the eigenmode function are continuously updated until the following conditions are met: Where: is the nth update of the Mth mode; T is the dual rise time step; ε1 and ε2 are tolerances; σ 2 is the noise variance.
7. The transformer leakage magnetic field signal measurement method based on improved SVMD combined with wavelet threshold method as claimed in claim 1, characterized in that: In step 3, the wavelet threshold method is combined to further process the residual noise in the signal, and the modal components and residual signals obtained by SVMD decomposition in step 2 are decomposed into r (t) is subjected to wavelet transform to obtain wavelet coefficients. Wavelet transform can effectively decompose the signal into different frequency bands. The soft threshold method is used for noise suppression, and its expression is as follows: Where: sgn is the sign function; β is the change scale; θ is the threshold parameter; w β,η is the original wavelet coefficient of the signal; w′ β,η is the wavelet coefficient after signal processing; Among them, the threshold parameter θ is expressed as follows: Where: μ is the noise standard deviation; N is the signal length; median is the median absolute deviation; w 1,η is the nth wavelet coefficient in the first layer.