Noise suppression method suitable for cable terminal discharge multi-physical field signals
By optimizing the combination of VMD parameters [K,α] and wavelet threshold function, the stability and accuracy of noise suppression in cable terminal discharge signals are solved, achieving efficient signal decomposition and denoising, and improving diagnostic accuracy.
Patent Information
- Application Number
- CN202411100788.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-12
- Publication Date
- 2024-11-15
AI Technical Summary
In the existing technology for multi-physics field signal processing of cable terminal discharge, noise suppression methods suffer from hardware noise floor and environmental noise affecting diagnostic accuracy, and parameter selection is difficult, leading to unstable signal decomposition or information loss.
A signal processing method based on variational mode decomposition (VMD) with a subtraction average optimizer and wavelet threshold function is adopted. By optimizing the combination of VMD parameters [K,α], the correlation between the IMF component and the original signal is calculated, and the improved threshold function is used for effective signal screening and reconstruction.
It improves the stability and denoising effect of signal decomposition, reduces the impact of noise, enhances the accuracy and computational efficiency of signal diagnosis, and avoids the risk of local optima.
Smart Images

Figure CN118964871A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of cable terminal discharge signal processing, and in particular to a noise suppression method suitable for cable terminal discharge multi-physical field signals. Background Art
[0002] The cable terminal discharge includes multi-physical field signal monitoring quantities such as sound, magnetic field, temperature field, etc. At present, multi-physical field signal fusion is often used for fault analysis, and the accuracy of analysis is greatly improved compared with a single signal. However, various signal acquisition stages will bring a large amount of hardware background noise and environmental noise. Various signals need to be cleaned to reduce the noise carried by the original signal and cause misjudgment of the diagnosis result. At this stage, the commonly used methods for signal denoising mainly include wavelet transform, empirical mode decomposition, singular spectrum analysis, and Kalman filtering, as follows:
[0003] Wavelet transform retains the wavelet coefficients controlled by the desired signal, removes the wavelet coefficients controlled by noise, and reconstructs the signal using inverse transform to obtain the desired denoised signal. Wavelet transform has good time-frequency localization characteristics, can better capture the details and characteristics of the signal, and can effectively compress the signal, thereby reducing storage space and transmission bandwidth. However, it is difficult to select a suitable wavelet basis function, and different wavelet basis functions have different effects on signal decomposition and analysis.
[0004] Empirical mode decomposition (EMD) is an adaptive data decomposition method that can effectively process nonlinear and non-stationary signals. It does not require pre-setting of basis functions, but decomposes according to the local characteristics of the signal, so it is suitable for various types of signals. In addition, EMD does not require linear transformation of the signal, so it can preserve the original characteristics of the signal and avoid information loss. In addition, EMD does not need to rely on frequency domain analysis when processing signals, so it is suitable for signal analysis with prominent time domain characteristics. However, when EMD processes signals with large noise, there will be too many or too few intrinsic mode functions (IMFs), resulting in unstable decomposition results. Secondly, when EMD processes signals with sparse extreme points or large frequency jumps, the extraction of intrinsic mode functions (IMFs) may be inaccurate. In addition, the computational complexity of the EMD algorithm is high and the requirements for computing resources are large, so it may face the problem of low computational efficiency when processing large-scale data.
[0005] Singular spectrum analysis is a method for processing nonlinear time series data. By decomposing and reconstructing the trajectory matrix of the time series to be studied, different component sequences (long-term trend, seasonal trend, noise, etc.) in the time series are extracted, so that the time series can be analyzed or denoised and used for other tasks. Singular spectrum analysis is model-free, and does not require the time series to be stable, nor does it assume a parameter model. It can be widely used in various time series. However, it is difficult to select the threshold for distinguishing signal components from noise components. When the deviation is large, it is impossible to effectively distinguish whether the corresponding component is a signal component or a noise component based on the size of the singular value.
[0006] The key to Kalman filtering is to achieve state estimation by dynamically fusing prior information (prediction) and observation information (update). By considering the system's dynamic model, measurement noise, and uncertainty in state estimation, the algorithm can provide relatively accurate state estimation in the presence of noise and incomplete observation data. By cyclically processing a series of observation data, the prediction and update steps of the Kalman filter are implemented, and the state estimation results for each time step are output. However, the Kalman filter requires an accurate system dynamic model and noise statistics. For complex nonlinear systems, an extended version needs to be used for adaptation; and the Kalman filter assumes that the observation noise is Gaussian distributed and independent and identically distributed. For non-Gaussian distributed noise, other filtering algorithms may need to be used.
[0007] Variational mode decomposition (VMD) is an adaptive, completely non-recursive method of modal variation and signal processing. It determines the number of modal decompositions of a given sequence according to actual conditions, and adaptively matches the optimal center frequency and limited bandwidth of each mode to achieve effective separation of IMFs and frequency domain division of signals, and finally obtains the optimal solution to the variational problem. It has strong adaptability and high efficiency. When performing variational mode decomposition (VMD) on the cable terminal discharge signal, it is necessary to set the appropriate number of decomposition modes K and the quadratic penalty factor α. When the value of K is too large, distortion will occur, and when it is too small, it will be under-decomposed, resulting in modal aliasing; similarly, when the value of α is too large, the frequency band information will be lost, and when it is too small, it will cause frequency crossover. Therefore, it is necessary to determine the optimal combination parameters [K, α]. Summary of the invention
[0008] In view of the above-mentioned deficiencies in the prior art, the present invention provides a noise suppression method applicable to multi-physical field signals of cable terminal discharge.
[0009] The technical solution adopted by the present invention is specifically as follows:
[0010] A noise suppression method for cable terminal discharge multi-physics field signals, comprising:
[0011] Step 1: Perform VMD decomposition and combination parameter optimization based on the subtraction mean optimizer to obtain the optimal combination of VMD parameters [K, α].
[0012] Step 2: Based on the optimal combination of VMD parameters [K, α] obtained in step 1, the original cable terminal discharge signal is decomposed by VMD to obtain a series of IMF components. The effectiveness of the component is evaluated by calculating the correlation between the IMF component and the original cable terminal discharge signal. The effective signal is denoised based on the improved threshold function, and the final output signal is completed after signal reconstruction.
[0013] Furthermore, step 1 specifically includes:
[0014] Step 1.1: Randomly update population positions
[0015] The population is randomly initialized in d-dimensional space, and the expression is:
[0016]
[0017] x i,j =lb j +r(ub j -lb j ) (2)
[0018] Where: i∈{1,2,…,N}, j∈{1,2,…,m}; x i,j For individuals, lb j To find the upper boundary, ub j To find the lower bound, r is a random number between [0,1], and the objective function of the problem is evaluated based on each search agent:
[0019]
[0020] In the formula, is the vector representation of the evaluation value; F(X i ) is the X for calculating the objective function i value;
[0021] Step 1.2: “-v” search
[0022] In each iteration, the position of a search agent B is determined by subtracting a value v from the position of another search agent A. This search method is called a "-v" search, and its position update process is:
[0023]
[0024] Where: is a vector of dimension m; * represents the Hadamard product of two vectors; F(A) and F(B) are the values of the objective functions of search agent A and search agent B, respectively;
[0025] Step 1.3: Search for the new location of the proxy
[0026] Any Search Proxy X i The displacement in the search space is done by each search agent X i The new position of the search agent is updated as follows:
[0027]
[0028] Where: is the i-th search agent X i ; N is the total number of search agents, and A vector of dimension m, where the components are normally distributed with values in the interval [0,1];
[0029] Step 1.4: Generate new location
[0030] If the new position leads to an improvement in the objective function value, then accept X in step 1.3 i As the new position of the corresponding search agent, otherwise the position remains unchanged until N iterations are completed. The final result is the global optimal solution, which corresponds to the optimal combination of VMD parameters [K, α], where K is the number of decomposition modes and α is the quadratic penalty factor.
[0031] Furthermore, step 2 specifically includes:
[0032] Step 2.1: Variational Mode Decomposition
[0033] The optimal solution is obtained through iterative search, thereby determining the center frequency and bandwidth of each IMF component, and decomposing the signal into K IMF components under the condition that the sum of the bandwidths of all IMF components is minimized;
[0034] Step 2.1: Calculation of modal function validity
[0035] After obtaining the corresponding K IMF components, the correlation coefficient between each IMF component and the original signal is calculated. The correlation coefficient represents the degree of correlation between the IMF component and the original signal, and the expression is:
[0036]
[0037] Where: R iIt represents the correlation coefficient between the IMF component u(t) and the original signal x(t). The larger the value, the higher the correlation degree, and the lower the noise content in the corresponding IMF component. The IMF components are screened by the size of the correlation coefficient, and the correlation coefficient R is retained. i IMF components greater than the correlation coefficient threshold;
[0038] Step 2.3: Wavelet threshold function denoising
[0039] The expression of the wavelet threshold function is:
[0040]
[0041] In the formula, w x is the wavelet decomposition coefficient corresponding to the noisy signal after wavelet transform, W x is the estimated value of the wavelet coefficient after threshold processing, λ is the calculated threshold, β is the adjustment coefficient of different signals. For different signals, a better denoising effect can be obtained by adjusting the size of β; when β = 0, |w x |≥λ is a soft threshold function, and when β=1, it is a hard threshold function;
[0042] Step 2.4: Signal reconstruction
[0043] After extracting the effective IMF components with high correlation, the filtered IMF components are superimposed and reconstructed according to certain weights to obtain an output signal after VMD decomposition and wavelet threshold filtering. The output signal is the reconstructed signal after noise reduction.
[0044] Furthermore, step 2.1 specifically includes:
[0045] (1) Establish a constrained variational model:
[0046]
[0047] In the formula, u k (t) is the IMF component, f is the original signal, k is the number of iterations, K is the number of decomposed modes, δ(t) is the impulse function, ω is the center frequency of each component, and st represents the constraint condition;
[0048] (2) By introducing the Lagrange multiplier λ(t) and the quadratic penalty factor α, the constrained variational problem is transformed into an unconstrained problem, and the extended Lagrange formula is obtained:
[0049]
[0050] In the formula, <, > means finding the inner product;
[0051] (3) Iterative solution. The specific process is as follows:
[0052] ①Use the optimal penalty factor α and decomposition mode number K obtained in step 1, and set u k ,ω k , the initial
[0053] value, and set n = 0;
[0054] ② Update u cyclically starting from 1 k and ω k , until k = K:
[0055]
[0056] ③ Update the Lagrangian operator:
[0057]
[0058] Determine whether the convergence condition is met:
[0059]
[0060] If it is satisfied, K IMF components are output, otherwise n=n+1 and the cycle continues.
[0061] Furthermore, the correlation coefficient threshold in step 2.1 is set to 0.6.
[0062] The present invention adjusts the positions of group members in the solution space by calculating the subtraction mean between individuals, so that the group members can quickly locate the global optimal solution. The algorithm not only improves the convergence speed, but also enhances the ability of the algorithm to balance between global search and local development, improves the decomposition performance of VMD, and effectively avoids the risk of falling into the local optimal solution. BRIEF DESCRIPTION OF THE DRAWINGS
[0063] Figure 1 It is a flow chart of a noise suppression method applicable to multi-physical field signals of cable terminal discharge according to an embodiment of the present invention;
[0064] Figure 2 It is a flow chart of VMD decomposition and combination parameter optimization based on subtraction mean optimizer of the present invention;
[0065] Figure 3 It is a flow chart of signal reconstruction by combining VMD with wavelet threshold to perform signal denoising in the present invention. DETAILED DESCRIPTION
[0066] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0067] The embodiment of the present invention provides a noise suppression method suitable for multi-physical field signals of cable terminal discharge, which mainly includes two processes of VMD decomposition and combination parameter optimization and VMD combined with wavelet threshold denoising signal reconstruction. The parameter optimization process adopts a method based on subtraction average optimizer to obtain the optimal combination of VMD parameters [K, α]. Then, the original cable terminal discharge signal is decomposed by VMD to obtain a series of IMFs. Then, the correlation between the IMF component and the original signal is calculated to evaluate whether the component is valid (the embodiment of the present invention defines that the correlation coefficient is greater than 0.6 as a valid signal, otherwise it is discarded). Finally, the denoising process of the effective signal is completed based on the improved threshold function, and the signal reconstruction is completed. The final output signal is the output signal.
[0068] Please also read Figure 1-3 , a noise suppression method applicable to a multi-physical field signal of a cable terminal discharge according to an embodiment of the present invention comprises the following steps:
[0069] Step 1: Perform VMD decomposition and combination parameter optimization based on the subtraction mean optimizer to obtain the optimal combination of VMD parameters [K, α].
[0070] The subtraction mean optimizer adjusts the position of group members in the solution space by calculating the subtraction mean between individuals, so that it can quickly locate the global optimal solution. The implementation process is as follows (such as Figure 2 shown):
[0071] Step 1.1: Randomly update population positions
[0072] The population is randomly initialized in d-dimensional space, and the expression is:
[0073]
[0074] x i,j =lb j +r(ub j -lb j ) (2)
[0075] Where: i∈{1,2,...,N}, j∈{1,2,...,m}; x i,j For individuals, lb jTo find the upper boundary, ub j To find the lower bound, r is a random number between [0,1]. Each search agent is a candidate solution to the problem and proposes a value for the decision variable. Therefore, the objective function of the problem can be evaluated based on each search agent:
[0076]
[0077] In the formula, is the vector representation of the evaluation value; F(X i ) is the X for calculating the objective function i value;
[0078] Step 1.2: “-v” search
[0079] In each iteration, the position of a search agent B is determined by subtracting a value v from the position of another search agent A. This makes the new position of each search agent adjusted based on the difference with other agents, effectively avoiding the problem of the algorithm falling into the local optimal solution. This search method is called "-v" search. Its position update process is
[0080]
[0081] Where: is a vector of dimension m; * represents the Hadamard product of two vectors; F(A) and F(B) are the values of the objective functions of search agent A and search agent B, respectively.
[0082] Step 1.3: Search for the new location of the proxy
[0083] Any Search Proxy X i The displacement in the search space is done by each search agent X i The new position of the search agent is calculated by subtracting the arithmetic mean of v.
[0084]
[0085] Where: is the i-th search agent X i ; N is the total number of search agents, and A vector of dimension m where the components are normally distributed with values in the interval [0,1].
[0086] Step 1.4: Generate new location
[0087] If the new position leads to an improvement in the objective function value, then the X in step 1.3 can be accepted. iAs the new position of the corresponding search agent, otherwise keep the position unchanged. After completing N iterations, the final result is the global optimal solution, which is the optimal combination of VMD parameters [K, α], where K is the number of decomposition modes and α is the quadratic penalty factor.
[0088] Step 2: Based on the optimal combination of VMD parameters [K, α] obtained in step 1, the original cable terminal discharge signal is decomposed by VMD to obtain a series of IMF components. The effectiveness of the component is evaluated by calculating the correlation between the IMF component and the original cable terminal discharge signal. The effective signal is denoised based on the improved threshold function, and the final output signal is completed after signal reconstruction.
[0089] The specific implementation process of step 2 is as follows:
[0090] Step 2.1: Variational Mode Decomposition
[0091] VMD is an iterative time-domain signal decomposition method. Its essence is to establish a variational model and obtain its optimal solution through iterative search, thereby determining the center frequency and bandwidth of each intrinsic mode function (IMF) component, and decomposing the signal into K IMF components under the condition that the sum of the bandwidths of all IMFs is minimized. Step 2.1 specifically includes:
[0092] (1) Establish a constrained variational model:
[0093]
[0094] In the formula, u k (t) is the IMF component, f is the original signal, k is the number of iterations, K is the number of decomposed modes, δ(t) is the impulse function, ω is the center frequency of each component, and st represents the constraint condition.
[0095] (2) By introducing the Lagrange multiplier λ(t) and the quadratic penalty factor α, the constrained variational problem is transformed into an unconstrained problem, and the extended Lagrange formula is obtained:
[0096]
[0097] In the formula, <, > means finding the inner product.
[0098] (3) Iterative solution. The specific process is as follows:
[0099] ①Use the optimal penalty factor α and decomposition mode number K obtained in step 1, and set u k ,ω k , the initial value of λ, and let n=0.
[0100] ② Update u cyclically starting from 1k and ω k , until k = K
[0101]
[0102] ③Update the Lagrangian operator.
[0103]
[0104] Determine whether the convergence condition is met
[0105]
[0106] If it is satisfied, K IMFs are output, otherwise n=n+1 and the cycle continues.
[0107] Step 2.1: Calculation of modal function validity
[0108] After obtaining the corresponding K modal components, the correlation coefficient between each modal component and the original signal is calculated. The correlation coefficient represents the correlation between the IMF component and the original signal, and the expression is:
[0109]
[0110] Where: R i The correlation coefficient between the IMF component u(t) and the original signal x(t) is expressed. The larger the value, the higher the correlation, and the lower the noise content in the corresponding IMF component. The IMF components are screened by the size of the correlation coefficient, and the IMF components with higher correlation are retained, while the components with lower correlation or useless components are excluded. The correlation coefficient threshold set by the present invention is 0.6.
[0111] Step 2.3: Wavelet threshold function denoising
[0112] Wavelet threshold denoising is a nonlinear denoising method, the core of which is to select the appropriate decomposition mode number and threshold function. Threshold functions are mainly divided into two types: hard threshold and soft threshold. The hard threshold function is discontinuous at the threshold point, and the wavelet estimation coefficients obtained by the soft threshold function will deviate from the wavelet coefficients of the noisy signal, thereby affecting the denoising effect. The present invention adopts an improved threshold function, which is expressed as:
[0113]
[0114] In the formula, w x is the wavelet decomposition coefficient corresponding to the noisy signal after wavelet transform, W x is the estimated value of the wavelet coefficient after threshold processing, λ is the calculated threshold, and β is the adjustment coefficient of different signals. For different signals, a better denoising effect can be obtained by adjusting the size of β. When β=0, |wx |≥λ is a soft threshold function, and when β=1, it is a hard threshold function. By changing the values of the two parameters, the effective signal energy can be retained, thereby solving the constant error problem in the traditional soft threshold function.
[0115] Step 2.4: Signal reconstruction
[0116] After extracting the effective IMF components with high correlation, the filtered IMF components are superimposed and reconstructed according to certain weights to obtain the output signal after VMD decomposition and wavelet threshold filtering. This output signal is the reconstructed signal after noise reduction, which has better signal quality and reliability.
[0117] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any changes or substitutions that can be easily thought of by a person skilled in the art within the technical scope disclosed by the present invention should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention should be based on the protection scope of the claims.
Claims
1. A noise suppression method for multi-physical field signals of cable terminal discharge, characterized in that: The steps include: Step 1: Perform VMD decomposition and combination parameter optimization based on the subtraction mean optimizer to obtain the optimal combination of VMD parameters [K, α]. Step 2: Based on the optimal combination of VMD parameters [K, α] obtained in step 1, the original cable terminal discharge signal is decomposed by VMD to obtain a series of IMF components. The effectiveness of the component is evaluated by calculating the correlation between the IMF component and the original cable terminal discharge signal. The effective signal is denoised based on the improved threshold function, and the final output signal is completed after signal reconstruction.
2. The noise suppression method for cable terminal discharge multi-physics field signals according to claim 1, characterized in that: Step 1 specifically includes: Step 1.1: Randomly update population positions The population is randomly initialized in d-dimensional space, and the expression is: x i,j =lb j +r(ub j -lb j ) (2) Where: i∈{1,2,…,N}, j∈{1,2,…,m}; x i,j For individuals, lb j To find the upper boundary, ub j To find the lower bound, r is a random number between [0,1], and the objective function of the problem is evaluated based on each search agent: In the formula, is the vector representation of the evaluation value; F(X i ) is the X for calculating the objective function i value; Step 1.2: "-v" search In each iteration, the position of a search agent B is determined by subtracting a value v from the position of another search agent A. This search method is called "-v" search, and its position update process is: Where: is a vector of dimension m; * represents the Hadamard product of two vectors; F(A) and F(B) are the values of the objective functions of search agent A and search agent B, respectively; Step 1.3: Search for the new location of the proxy Any Search Proxy X i The displacement in the search space is done by each search agent X i The new position of the search agent is updated as follows: Where: is the i-th search agent X i ; N is the total number of search agents, and A vector of dimension m, where the components are normally distributed with values in the interval [0,1]; Step 1.4: Generate new location If the new position leads to an improvement in the objective function value, then accept X in step 1.3 i As the new position of the corresponding search agent, otherwise the position remains unchanged until N iterations are completed. The final result is the global optimal solution, which corresponds to the optimal combination of VMD parameters [K, α], where K is the number of decomposition modes and α is the quadratic penalty factor.
3. The noise suppression method for cable terminal discharge multi-physics field signals according to claim 1, characterized in that: Step 2 specifically includes: Step 2.1: Variational Mode Decomposition The optimal solution is obtained through iterative search, thereby determining the center frequency and bandwidth of each IMF component, and decomposing the signal into K IMF components under the condition that the sum of the bandwidths of all IMF components is minimized; Step 2.1: Calculation of modal function validity After obtaining the corresponding K IMF components, the correlation coefficient between each IMF component and the original signal is calculated. The correlation coefficient represents the degree of correlation between the IMF component and the original signal, and the expression is: Where: R i It represents the correlation coefficient between the IMF component u(t) and the original signal x(t). The larger the value, the higher the correlation degree, and the lower the noise content in the corresponding IMF component. The IMF components are screened by the size of the correlation coefficient, and the correlation coefficient R is retained. i IMF components greater than the correlation coefficient threshold; Step 2.3: Wavelet threshold function denoising The expression of the wavelet threshold function is: In the formula, w x is the wavelet decomposition coefficient corresponding to the noisy signal after wavelet transform, W x is the estimated value of the wavelet coefficient after threshold processing, λ is the calculated threshold, β is the adjustment coefficient of different signals. For different signals, a better denoising effect can be obtained by adjusting the size of β; when β = 0, |w x |≥λ is a soft threshold function, and when β=1, it is a hard threshold function; Step 2.4: Signal reconstruction After extracting the effective IMF components with high correlation, the filtered IMF components are superimposed and reconstructed according to certain weights to obtain an output signal after VMD decomposition and wavelet threshold filtering. The output signal is the reconstructed signal after noise reduction.
4. The noise suppression method for cable terminal discharge multi-physics field signals as claimed in claim 3, characterized in that: Step 2.1 specifically includes: (1) Establish a constrained variational model: In the formula, u k (t) is the IMF component, f is the original signal, k is the number of iterations, K is the number of decomposed modes, δ(t) is the impulse function, ω is the center frequency of each component, and st represents the constraint condition; (2) By introducing the Lagrange multiplier λ(t) and the quadratic penalty factor α, the constrained variational problem is transformed into an unconstrained problem, and the extended Lagrange formula is obtained: In the formula, <, > means finding the inner product; (3) Iterative solution. The specific process is as follows: ①Use the optimal penalty factor α and decomposition mode number K obtained in step 1, and set u k ,ω k , the initial value of λ, and let n = 0; ② Update u cyclically starting from 1 k and ω k , until k = K: ③ Update the Lagrangian operator: Determine whether the convergence condition is met: If it is satisfied, K IMF components are output, otherwise n=n+1 and the cycle continues.
5. The noise suppression method for cable terminal discharge multi-physics field signals as claimed in claim 3, characterized in that: The correlation coefficient threshold described in step 2.1 was set to 0.6.
Citation Information
Patent Citations
Image denoising method for improving wavelet threshold and threshold function
CN117593210A
Cited By
Insulator partial discharge positioning method
CN120214516A
