A Decomposition Method for Complex Electromagnetic Interference Signals with High Time-Frequency Aggregation Degree

Abstract

The present invention discloses a method for decomposing complex electromagnetic interference signals with high time-frequency concentration, and proposes a time-frequency decomposition method for complex electromagnetic interference signals based on variational mode decomposition technology, generalized cross-correlation entropy criterion, correlation coefficient and time-frequency energy concentration target. The proposed method solves the problem of selecting the mode number K and the penalty factor α, and can obtain the optimal and most reasonable mode number and penalty factor without losing or adding sub-modes, so as to make the time-frequency concentration of the decomposition reach the optimal. The method for decomposing complex electromagnetic interference signals with high time-frequency concentration proposed by the present invention helps to study the electromagnetic interference problems of smart grid equipment, especially provides a feasible and excellent method for the decomposition and calculation of electromagnetic interference signals with wide frequency bands, dense frequency points and high complexity, and provides solutions and reference values for electromagnetic interference protection and resistance.

Description

Technical Field

[0001] The present invention belongs to the technical field of electromagnetic compatibility of smart grid equipment, and particularly relates to a method for decomposing complex electromagnetic interference signals with high time-frequency concentration degree. Background Art

[0002] New types of smart grid equipment are an important part of future smart grid construction. During the process of research and development and popularization, faults caused by electromagnetic interference account for the vast majority. Moreover, many electromagnetic interference signals are caused by the superposition of multiple factors, and the signals are usually noisy and broadband, which may lead to the loss or aliasing of characteristic frequency bands, and it is difficult to achieve results using traditional analysis methods. Therefore, it is necessary to first study the characteristics of transient electromagnetic interference in the power system suffered by smart grid equipment, master its time-frequency spectrum distribution, and then carry out protection according to the characteristics of electromagnetic interference signals to improve the reliability of new equipment when encountering electromagnetic interference, which has important theoretical and engineering value.

[0003] The electromagnetic interference problems in new equipment mainly exist in the following aspects currently:

[0004] (1) Power equipment represented by technologies such as primary-secondary integration and new DC breaking generates electromagnetic interference problems with new characteristics;

[0005] (2) The problem of mutual transfer of interference between AC and DC caused by flexible DC systems appears for the first time;

[0006] (3) The electromagnetic compatibility problems of a series of measurement equipment derived from new sensing technologies when encountering strong electromagnetic transients in the power system.

[0007] Therefore, the typical electromagnetic interference signals generated by the above problems have significant differences in time-domain and frequency-domain characteristics compared with traditional electromagnetic interference signals. They mainly exhibit the characteristics of high amplitude, multiple frequency points, and broadband. There is usually time-domain crossover between signals emitted by the same interference source or signals from different interference sources, and it is necessary to use time-frequency analysis technology for precise discrimination and extraction. Summary of the Invention

[0008] The purpose of the present invention is to provide a method for decomposing complex electromagnetic interference signals with high time-frequency concentration degree. By analyzing signal characteristics, using the variational mode decomposition method, extracting the optimal decomposition parameters of the signal, and then performing variational mode decomposition and time-frequency spectrum superposition, the precise time-frequency spectrum distribution of the electromagnetic interference signal is obtained, and its characteristic frequency is accurately controlled.

[0009] The technical solution adopted by the present invention is: a method for decomposing complex electromagnetic interference signals with high time-frequency concentration degree, which specifically includes the following steps:

[0010] Step 1: Conduct characteristic test verification on the target electromagnetic interference signal to determine whether its signal characteristics conform to non-Gaussian distribution, that is, the kurtosis is not equal to 3 and the skewness is not equal to 0. If it conforms, proceed to Step 2;

[0011] Step 2: Decompose the target electromagnetic interference signal, that is, perform variational mode decomposition calculation on it. First, take the initial values for the parameters K and α preset for variational mode decomposition (usually K = 2, α = 1000), perform variational mode decomposition corresponding to K and α, decompose to obtain K sub-modal data and the spectral data of the K sub-modalities, construct a similarity matrix using the spectral data of the K sub-modalities obtained by variational mode decomposition, calculate each element of the similarity matrix using the generalized cross-correlation entropy criterion, and use the particle swarm optimization algorithm to optimize the internal parameters of the generalized cross-correlation entropy during the calculation to obtain a set of calculation results of the generalized cross-correlation entropy criterion;

[0012] Step 3: If the calculation result in Step 2 is equal to the optimal criterion result of the generalized cross-correlation entropy, that is, equal to 1, it indicates that the optimal K and α have been found, and proceed to execute Step 4. If it does not meet the optimal criterion result, that is, is not equal to 1, then continue to superimpose the values of K and α, increase K by 1, and increase α by 1000 (the values by which K and α are increased each time here can also be selected according to the purpose, with K increased by 1 / 2 / 3 each time and α increased by 1000 / 2000 / 5000 each time), and continue to execute Step 2 until the calculation result of Step 2 is equal to the optimal generalized cross-correlation entropy criterion result;

[0013] Step 4: Perform variational mode decomposition with the optimal K and α values obtained in Step 3, perform generalized S-transform decomposition on the obtained K sub-modal data, and superimpose the K groups of generalized S-transform decomposition results to obtain the time-frequency spectrum distribution with the highest time-frequency concentration degree, that is, the optimal time-frequency spectrum distribution of the target electromagnetic interference signal, and observe the frequency and time distribution of the target electromagnetic interference signal.

[0014] The characteristics of the present invention also lie in that

[0015] The characteristic test verification of the target electromagnetic interference signal performed in Step 1 is as follows: Assume that the problem to be solved is the interference signal f, and the signal is tested using the Jarque-Bera method. The Jarque-Bera method JB is a method for testing whether data conforms to Gaussian distribution through kurtosis and skewness. Skewness is used to describe the symmetry degree of the data distribution characteristics. If the value is greater than 0, it has a right-skewed characteristic; if the value is less than 0, it has a left-skewed characteristic; when the value is equal to 0, it indicates that the symmetry degree is consistent with the Gaussian distribution. Kurtosis is used to describe the steepness of the data distribution. If the value is greater than 3, the distribution is steep; otherwise, the distribution is gentle. When the value is equal to 3, it indicates that the steepness is consistent with the Gaussian distribution.

[0016] In summary, in the test results, if the skewness is not equal to 0 and the kurtosis is not equal to 3, it can be determined that the signal does not conform to the Gaussian distribution. The more it does not conform to the Gaussian distribution, the more complex the signal is, and the higher the decomposition value is, which can be used in the method proposed by the present invention.

[0017] The statistic JB of the interference signal f can be expressed as:

[0018]

[0019] Where:

[0020] μ —— The mean of f;

[0021] n —— The data length of the interference signal;

[0022] E[·] —— Mathematical expectation;

[0023] σ —— The standard deviation of f;

[0024] K —— The kurtosis of f;

[0025] S —— The skewness of f, S < 0 indicates right skew, S > 0 indicates left skew.

[0026] The variational mode decomposition method in step 2 is a new signal processing method with a multi-resolution non-recursive variational structure that combines Wiener filtering, Hilbert transform, and alternating direction multiplier method. Its effect can be briefly explained as: decomposing an initial signal into multiple sub-signals, these sub-signals have their own time-frequency characteristics, and the algebraic sum is still the initial signal.

[0027] Its mechanism is: constructing the original electromagnetic interference signal f into a constrained variational problem as shown in Equation (2):

[0028]

[0029] In the formula, {u K}:={u1,...,u k} and {w K}:={w1,...,w K} respectively represent the sets of all sub-modalities and the corresponding center frequencies, is the partial derivative with respect to time t, δ(t) is the unit impulse response function, j is the imaginary unit, * represents convolution, f represents the input signal, and K is the number of sub-modalities for decomposition.

[0030] Introducing the quadratic penalty factor α and the Lagrange multiplier λ, Equation (2) is transformed into Equation (3)

[0031]

[0032] In the formula, {u K}:={u1,...,uk} and {w K}: = {w1,..., w K} represent the sets of all sub - modalities and the corresponding center frequencies respectively. is the partial derivative with respect to time t, δ(t) is the unit impulse response function, j is the imaginary unit, * represents convolution, f represents the input signal, K is the number of decomposed sub - modalities, α is the quadratic penalty factor, λ is the Lagrange multiplier, f(t) is the original electromagnetic interference signal, and λ(t) is the formula with the Lagrange multiplier term.

[0033] The saddle point of the Lagrangian equation in Equation (3) is the optimal solution of its variational problem. The alternating direction multiplier method is used to alternately and iteratively update the sub - modality u k and the center frequency w k and the Lagrange multiplier λ, and the iteration is completed relatively quickly to obtain the corresponding solution.

[0034] The generalized cross - correlation entropy criterion in Step 2 is described as follows:

[0035] The generalized cross - correlation entropy has an adaptive non - parametric property and can learn more information from numerous data, showing its superior performance when dealing with non - Gaussian signals. It can be applied to similarity measurement. The following gives the expression of the correlation entropy between random variables X and Y:

[0036] V θ (X, Y) = E[k θ (X, Y)] = ∫∫k θ (x, y)f XY (x, y)dxdy (4)

[0037] Where:

[0038] E[·] —— Mathematical expectation;

[0039] k θ (·, ·) —— Represents the kernel function with kernel width θ;

[0040] f XY (·, ·) —— Joint probability density function.

[0041] In practice, f XY (·, ·) is usually unknown, and generally, the data collected by sensors are mostly some discrete and finite amounts of data Therefore, using sample estimation, the expression of the correlation entropy of discrete variables can be obtained:

[0042]

[0043] Where k θ is the kernel function of the correlation entropy. When the kernel function k of the correlation entropy θWhen (·,·) selects the generalized Gaussian density function and is used as the learning criterion for adaptive system training, it is called the generalized cross-correlation entropy criterion. The expression of the generalized Gaussian density is:

[0044]

[0045] In the formula:

[0046] Γ(·) —— Gamma function;

[0047] β>0 —— Proportional bandwidth;

[0048] α>0 —— Shape parameter;

[0049] λ —— Kernel parameter;

[0050] γ α,β —— Normalization constant.

[0051] Then the generalized cross-correlation entropy criterion can be expressed by the following formula:

[0052]

[0053] Where N represents the data length, and the absolute value of the difference between x i and y i represents the difference measure between samples. When it is infinitely close to 0, the calculated evaluation criterion maxJ is larger. Ideally, the optimal value of the generalized cross-correlation entropy criterion is 1.

[0054] According to Equation (7), x i and y i represent two different groups of discrete signals respectively. Let x i -y i =Δd t , and use Δd t to measure the correlation between the main frequencies of K sub-modal signals. Then Equation (7) can be rewritten as

[0055]

[0056] In the formula, Δd t is the measurement factor, γ α,λ is the normalization constant, λ is the kernel parameter, and α is the shape parameter.

[0057] According to the properties of the generalized cross-correlation entropy criterion, the smaller the absolute value of Δd t , the larger its maxJ, and the generalized cross-correlation entropy criterion has an optimal value.

[0058] α and λ in Equation (8) are used as internal parameters in the calculation process. In step 2, the particle swarm optimization algorithm is used here to optimize the two-dimensional parameters.

[0059] The construction method of the similarity matrix in Step 2 is as follows:

[0060] When performing spectral analysis on the variational mode decomposition signal, usually each mode IMF i (i = 1, 2, 3...K) can obtain the corresponding spectral curve, which generally contains 1 - 3 spectral peaks (the number depends on the decomposition performance). The K sub - modes IMF obtained by variational mode decomposition of the data with the same length N i have the length of N, so the length of their spectral curves is also N. Therefore, the correlation between each pair of sub - modes can be mapped to Δd in the generalized cross - correlation entropy criterion in Equation (6). t For mapping.

[0061] The correlation between each sub - mode is recorded as matrix C

[0062]

[0063] In the formula, C ij represents the correlation between the i - th and j - th sub - modes (i, j = 1, 2...K). Since the correlation of each sub - mode itself is meaningless and this matrix is strictly symmetric, matrix (9) can be rewritten as (10).

[0064]

[0065] Matrix C1 is essentially an upper triangular matrix without diagonal elements. Compared with Equation (9), it retains its meaning and can reduce the computational complexity in the follow - up.

[0066] Let C1 = Δd t , so Equation (8) can be written as

[0067]

[0068] In the formula, K is the set number of decomposition layers, Δd t is the measurement factor, γ α,β is the normalization constant, λ is the kernel parameter, α is the shape parameter, N is the number of rows of the upper triangular matrix of matrix C1, and i and j respectively represent the row and column of the elements of matrix C1.

[0069] In the present invention, the Spearman correlation coefficient (SCC) is selected as the calculation criterion here. This correlation coefficient can be effectively used for measuring the similarity between two sets of data x and y, and its expression is as shown in Equation (12). The common performance measurements are shown in Table 4.

[0070]

[0071] In the formula, N represents the data length, xi represents each element in data x, represents the average value of data x, and yi represents each element in data y. represents the average value of data y.

[0072] Table 4 SCC Evaluation Criteria

[0073]

[0074] Accordingly, Δd in Equation (11) ij can be denoted as SCC<IMFi, IMFj>. Substituting it into Equation (11) gives Equation (13), and the optimal solution of this equation can be used to constrain the parameters of variational mode decomposition.

[0075]

[0076] The generalized S-transform in Step 4 is a commonly used operation method in time-frequency signal analysis. Using this method, the time-frequency distribution characteristics of the signal can be obtained, and it has good time-frequency resolution performance. Assuming that the target electromagnetic interference signal component is f, the generalized S-transform is defined as Equation (14).

[0077]

[0078] where f(t) is the target electromagnetic interference signal, t is the integration variable, v is the frequency, τ is the time shift factor, w(t - τ, v, λ) is the Gaussian window function, and λ is the adjustment factor. The introduction of λ can control the change speed of v, making the time-frequency analysis more adaptable and flexible. To objectively reflect the effect of the method proposed in the present invention, λ is uniformly taken as 0.5 here. After all the sub-modalities obtained by variational mode decomposition are processed using the above-mentioned generalized S-transform, the final result can be obtained by performing the addition of matrices of the same order.

[0079] The beneficial effects of the present invention are:

[0080] A method for decomposing complex electromagnetic interference signals with high time-frequency concentration proposed by the present invention, aiming at the current situation of increasingly complex electromagnetic interference, wide frequency band, and dense frequency points that are difficult to distinguish, proposes an adaptive variational mode decomposition method to accurately decompose electromagnetic interference signals and obtain an accurate time-frequency spectrum distribution, thereby revealing the signal law and providing a reference and basis for the design of protecting complex electromagnetic interference in engineering applications.

[0081] Introduce the new signal processing method of variational mode decomposition, combine it with the generalized cross-correlation entropy criterion, and creatively invent a correlation matrix constructed from the results of variational mode decomposition. Combine the matrix elements with the generalized cross-correlation entropy to overcome the problem that both the optimal K and α are unknown during variational mode decomposition, and also greatly improve the time-frequency concentration of the decomposition results, providing a feasible solution method for solving such electromagnetic interference signal problems. Description of the Drawings

[0082] Figure 1 is the electromagnetic interference signal pattern in Embodiment 1 of the present invention;

[0083] Figure 2 is the specific implementation flowchart of the present invention;

[0084] Figure 3 is the time-frequency distribution result of the final decomposition of the electromagnetic interference signal in Embodiment 1 of the present invention;

[0085] Figure 4 is the electromagnetic interference signal pattern in Embodiment 2 of the present invention;

[0086] Figure 5 is the time-frequency distribution result of the final decomposition of the electromagnetic interference signal in Embodiment 2 of the present invention. Specific Embodiment

[0087] The following further clarifies the present invention in conjunction with specific embodiments. It should be understood that these embodiments are only used to illustrate the present invention and not to limit the scope of the present invention. After reading the present invention, various equivalent forms of modification by those skilled in the art fall within the scope defined by the appended claims of this application.

[0088] The method for decomposing complex electromagnetic interference signals with high time-frequency concentration degree of the present invention has the specific process as Figure 2 shown, and the specific operation steps are as follows:

[0089] Embodiment 1

[0090] Step 1: Perform feature testing and verification on a target electromagnetic interference signal with a certain data length of 1000 (as Figure 1 shown). The signal length is 5 microseconds and is obtained by superimposing 6 sub-modalities with different frequencies (the specific information of these 6 sub-modalities is shown in Table 1). First, perform the Jarque-Bera test on the target electromagnetic interference signal, and obtain its kurtosis of 14.1427 and skewness of -1.5429, belonging to non-Gaussian distribution, and the method of the present invention can be used for decomposition;

[0091] Step 2: Initialize the electromagnetic interference signal to be decomposed, import the variational mode decomposition calculation, first take the initial values of K and α for the variational mode decomposition, the initial value of K is 2, the initial value of α is 1000, perform the variational mode decomposition under the corresponding K and α, construct a similarity matrix for the K sub-modal spectrum data obtained by decomposition, use the particle swarm optimization algorithm to optimize the generalized cross-correlation entropy criterion, calculate the generalized cross-correlation entropy criterion for each element of the similarity matrix, and obtain the calculation result under the generalized cross-correlation entropy criterion.

[0092] Step 3: Judge the calculation result. If it does not meet the optimal requirement of the criterion, that is, it is not equal to 1, then accumulate the values of K and α, K + 1, α + 1000, and continue to execute Step 2 until the result of the generalized cross-correlation entropy criterion is calculated as 1, obtaining a set of optimal K and α values, and then proceed to Step 4;

[0093] Step 4: Perform variational mode decomposition on the optimal K and α values obtained in Step 3 (in this example, K is 6 and α is 26000), perform generalized S-transform decomposition on the obtained 6 sub-modalities, and superimpose the 6 groups of decomposition results to obtain the time-frequency spectrum distribution with the highest time-frequency concentration degree, that is, the optimal time-frequency spectrum distribution of the target electromagnetic interference signal, and observe the frequency and time distribution of the target electromagnetic interference signal as Figure 3 shown (the abscissa is time, the ordinate is the frequency spectrum, and the color depth represents the amplitude size).

[0094] To evaluate the superiority and effectiveness of the method described in the present invention, seven other similar time-frequency decomposition methods are used to decompose this electromagnetic interference signal. These seven methods are variational mode decomposition (variational mode decomposition), empirical mode decomposition (variational mode decomposition), local mean decomposition (LMD), intrinsic time-scale decomposition (ITD), continuous variational mode decomposition (S variational mode decomposition), maximum correlation entropy-variational mode decomposition (MCC-variational mode decomposition), and mixed correlation entropy-variational mode decomposition (MC-variational mode decomposition). To measure the advantages and disadvantages of various methods, three indicators are used here to measure and compare the decomposition results, including Rayleigh entropy Ra, time-frequency energy concentration degree CM, and norm ratio RN. Among them, when the value of Ra is smaller and the values of CM and RN are larger, it indicates that the decomposition effect is better, indicating that the time-frequency concentration degree of the decomposition result is higher. The specific results are shown in Table 2. The method proposed in the present invention is optimal in all three indicators under the same conditions, and the number of sub-modalities obtained by decomposition is 6, which is consistent with the superposition of 6 sub-modalities with different frequency components of this signal. The results obtained by other decomposition methods do not match the actual number of sub-modalities, proving that the decomposition effect of the present invention has obvious advantages.

[0095] Table 1 Situation of 6 sub-modalities in Example 1

[0096]

[0097] Table 2 Comparison of the effects of the present invention and other methods (Example 1)

[0098]

[0099]

[0100] Example 2

[0101] Step 1: Conduct feature testing and verification on the electromagnetic interference signal 2 in a real engineering project with a data length of 2000 (such as Figure 4 shown). First, perform the Jarque - Bera test on it, and obtain that its kurtosis is 12.462 and skewness is 3.69, belonging to non - Gaussian distribution, and the method of the present invention can be used for decomposition;

[0102] Step 2: Initialize the electromagnetic interference signal to be decomposed, import the variational mode decomposition calculation, first take the initial values of K and α for the variational mode decomposition. The initial value of K is 2, and the initial value of α is 1000. Execute the variational mode decomposition under the corresponding K and α, construct a similarity matrix for the K sub - modal spectrum data obtained by decomposition, use the particle swarm optimization algorithm to optimize the generalized cross - correlation entropy criterion, calculate the generalized cross - correlation entropy criterion for each element of the similarity matrix, and obtain the calculation result under the generalized cross - correlation entropy criterion.

[0103] Step 3: Judge the calculation result. If it does not meet the optimal requirement of the criterion, that is, it is not equal to 1, then accumulate the values of K and α, K + 1, α+1000, and continue to execute Step 2 until the generalized cross - correlation entropy criterion result is 1, obtaining a set of optimal K and α values, and proceed to Step 4;

[0104] Step 4: Perform variational mode decomposition with the optimal K and α values obtained in Step 3 (in this example, K is 4 and α is 3000), perform generalized S - transform decomposition on the 4 obtained sub - modes, and superimpose the 4 groups of decomposition results to obtain the time - frequency spectrum distribution with the highest time - frequency concentration degree, that is, the best time - frequency spectrum distribution of the target electromagnetic interference signal. Observe the frequency and time distribution of the target electromagnetic interference signal as shown in Figure 5 shown (the abscissa is time, the ordinate is the spectrum, and the color depth represents the amplitude size).

[0105] Similarly, the comparison effect with other methods is shown in Table 3. The evaluation index results of the same as Example 1 show that the method proposed by the present invention has obvious advantages under the same conditions.

[0106] Table 3 Comparison of the effects of the present invention and other methods (Example 2)

[0107]

[0108]

Claims

1. A method for decomposing complex electromagnetic interference signals with high time-frequency aggregation, characterized in that, Specifically, it includes the following steps: Step 1: Conduct characteristic test verification on the target electromagnetic interference signal to determine whether its signal characteristics conform to non-Gaussian distribution, that is, the kurtosis is not equal to 3 and the skewness is not equal to 0. If it conforms, proceed to Step 2; Step 2: Perform variational mode decomposition on the target electromagnetic interference signal. First, take the initial values for the parameters K and α required for variational mode decomposition. Let K = 2 and α = 1000. Perform variational mode decomposition corresponding to K and α, and decompose to obtain K sub-modal data and the spectral data of K sub-modalities. Construct a similarity matrix using the spectral data of the K sub-modalities obtained by variational mode decomposition. Calculate each element of the similarity matrix using the generalized cross-correlation entropy criterion. When calculating, use the particle swarm optimization algorithm to optimize the internal parameters of the generalized cross-correlation entropy to obtain a set of calculation results of the generalized cross-correlation entropy criterion; Step 3: If the calculation result in Step 2 is equal to the optimal criterion result of the generalized cross-correlation entropy, that is, equal to 1, it means that the optimal K and α have been found, and output to execute Step 4; If it does not meet the optimal criterion result, that is, it is not equal to 1, then continue to superimpose the values of K and α. Increase K by 1 and increase α by 1000, and continue to execute Step 2 until the calculation result of Step 2 is equal to the optimal generalized cross-correlation entropy criterion result; Step 4: Perform variational mode decomposition with the optimal K and α values obtained in Step 3. Perform generalized S-transform decomposition on the obtained K sub-modal data, and superimpose the K sets of generalized S-transform decomposition results to obtain the time-frequency spectrum distribution with the highest time-frequency concentration degree, that is, the optimal time-frequency spectrum distribution of the target electromagnetic interference signal, and observe the frequency and time distribution of the target electromagnetic interference signal.

2. The method for decomposing complex electromagnetic interference signals with high time-frequency concentration according to claim 1, wherein Specifically, the characteristic test verification of the target electromagnetic interference signal in Step 1 is as follows: The problem to be solved is the interference signal f. Use the Jarque-Bera method JB to test the signal. The Jarque-Bera method is a method to test whether the data conforms to Gaussian distribution through kurtosis and skewness; skewness is used to describe the symmetry degree of the data distribution. If the value is greater than 0, it has a right-skewed characteristic. If the value is less than 0, it has a left-skewed characteristic. When the value is equal to 0, it indicates that it is consistent with the symmetry degree of Gaussian distribution; kurtosis is used to describe the steepness of the data distribution. If the value is greater than 3, the distribution is steep, otherwise the distribution is flat. When the value is equal to 3, it indicates that it is consistent with the steepness of Gaussian distribution; In short, in the test result, if the skewness is not equal to 0 and the kurtosis is not equal to 3, it can be judged that the signal does not conform to Gaussian distribution. The more it does not conform to Gaussian distribution, the more complex the signal is and the higher the decomposition value is, and it can be used for further decomposition: Among them, the statistic JB of the interference signal f can be expressed as: Where: μ —— The mean of f; n —— The data length of the interference signal; E[·] —— Mathematical expectation; σ —— The standard deviation of f; K —— The kurtosis of f; S —— The skewness of f, S < 0 indicates right skew, and S > 0 indicates left skew.

3. A method for decomposing complex electromagnetic interference signals with high time-frequency concentration degree according to claim 1, characterized in that The specific process of the variational mode decomposition in Step 2 is as follows: Decompose an initial signal into multiple sub-signals. These sub-signals have their own time-frequency characteristics, and their algebraic sum is still the initial signal. The mechanism is: construct the original electromagnetic interference signal f into a constrained variational problem as shown in Equation (2): where {u K}:={u1,...,u k} and {w K}:={w1,...,w K} represent the sets of all sub - modes and the corresponding center frequencies respectively, is the partial derivative with respect to time t, δ(t) is the unit impulse response function, j is the imaginary unit, * represents convolution, f represents the input signal, and K is the number of sub - modes for decomposition; Introduce the secondary penalty factor α and the Lagrange multiplier λ, and transform Equation (2) into Equation (3). Where α is the quadratic penalty factor, λ is the Lagrange multiplier, f(t) is the original electromagnetic interference signal, and λ(t) is the formula containing the Lagrange multiplier term; the alternating direction method of multipliers is used to alternately and iteratively update the sub-mode u k and the center frequency w k and the Lagrange multiplier λ. After the iteration is completed, the corresponding solution is obtained.

4. A method for decomposing complex electromagnetic interference signals with high time-frequency concentration degree according to claim 1, characterized in that The generalized cross-correlation entropy criterion in Step 2 is as follows: The following gives the expression of the generalized cross-correlation entropy between random variables X and Y: V θ (X, Y) = E[k θ (X, Y)] = ∫∫k θ (x, y)f XY (x, y)dxdy (4) In the formula: E[·] —— Mathematical expectation; k θ (·,·) —— represents a kernel function with a kernel width of θ; f XY (·,·)——Joint probability density function; f XY (·,·) is a set of discrete and finite data Therefore, using sample estimation, the expression of the correlation entropy of discrete variables can be obtained: where k θ is the kernel function of the correlation entropy; when the kernel function k of the correlation entropy θ (·,·), the generalized Gaussian density function is selected, and the expression of the generalized Gaussian density is: In the formula: Γ(·) —— Gamma function; β>0 —— Proportional bandwidth; α>0 —— Shape parameter; λ —— Kernel parameter; γ α,β —— Normalization constant; Then the generalized cross-correlation entropy criterion can be expressed by the following formula: where N represents the data length, x i and y i The absolute value of the difference represents the difference measure between samples. When it approaches 0 infinitely, the calculated evaluation criterion maxJ is larger. Ideally, the optimal value of the generalized cross-correlation entropy criterion is 1; According to Equation (7), x i and y i respectively represent two different groups of discrete signals. Let x i - y i = Δd t . Using Δd t to measure the correlation between the main frequencies of K sub-modal signals, then Equation (7) can be rewritten as where Δd t is a measurement factor, γ α,λ is a normalization constant, λ is a kernel parameter, and α is a shape parameter; According to the properties of the generalized cross-correlation entropy criterion, the smaller the absolute value of Δd t , the larger its maxJ, and the generalized cross-correlation entropy criterion has an optimal value.

5. A method for decomposing complex electromagnetic interference signals with high time-frequency concentration according to claim 1, characterized in that The method for constructing the similarity matrix in Step 2 is as follows: When performing spectral analysis on the variational mode decomposition signal, usually each mode IMF i , i = 1, 2, 3... K; the corresponding spectral curve can be obtained. The K sub-modes IMF i obtained by variational mode decomposition of the target electromagnetic interference signal with a length of N are all of length N, so the length of its spectral curve is also N. Therefore, the similarity between each pair of sub-modes and Δd t in the generalized cross-correlation entropy criterion in Equation (6) can be mapped: The similarity between each sub-modal is counted as matrix C where C ij represents the correlation between the i-th and j-th sub-modalities, i, j = 1, 2... K. Since the similarity of each sub-modality itself is meaningless and the matrix is strictly symmetric, the matrix (9) can be rewritten as matrix C1, as shown in Equation (10): Matrix C1 is essentially an upper triangular matrix without diagonal elements. Compared with Equation (9), its meaning is retained and the computational complexity can be reduced subsequently; Let C1 = Δd t , so Equation (8) can be written as where K is the set number of decomposition layers, Δd t is the metric factor, γ α,β is the normalization constant, λ is the kernel parameter, α is the shape parameter, N is the number of rows of the upper triangular matrix of matrix C1, and i and j respectively represent the row and column of the elements of matrix C1.

6. A method for decomposing complex electromagnetic interference signals with high time-frequency concentration according to claim 2, characterized in that, Step 4 is specifically as follows: Let the target electromagnetic interference signal component be f, and define the generalized S transform as Equation (14): Among them, f(t) is the target electromagnetic interference signal, w(t-τ,v,λ) is the Gaussian window function, t is the integration variable, v is the frequency, τ is the time shift factor, λ is the adjustment factor, and the introduction of λ can control the change speed of v, making the adaptability and flexibility of time-frequency analysis stronger. Here, λ is uniformly taken as 0.5; after all sub-modalities obtained by variational mode decomposition are processed using the above generalized S transform, the final result can be obtained by performing the addition of matrices of the same order.

7. A method for decomposing complex electromagnetic interference signals with high time-frequency concentration according to claim 2, characterized in that The values of K and α increased each time in Step 3 can also be selected according to the purpose. K is increased by 1 or 2 or 3 each time, and α is increased by 1000 or 2000 or 5000 each time.

Images