Residual current noise reduction method based on adaptive MVMD

Abstract

The invention discloses a residual current noise reduction method based on adaptive MVMD, and the method comprises the steps: firstly carrying out the robust empirical mode decomposition (REMD) of a residual current signal, obtaining a plurality of intrinsic mode components IMF and a residual component Res, and calculating the correlation coefficient of each IMF component; determining the modal decomposition number K of the MVMD according to the number of the IMFs greater than the correlation intensity threshold, performing multivariate variational modal decomposition MVMD on the original residual current signal, and selecting the IMFs with the correlation intensity threshold to perform signal reconstruction; and the reconstructed current signal is subjected to singular value decomposition (SVD) secondary noise reduction to obtain a final noise-reduced signal, so that the influence of resonance noise caused by distributed photovoltaic access on the action of the residual current signal detection and protection device is reduced. The method solves the problems that a traditional noise reduction technology is poor in noise reduction effect, and a noise reduction algorithm based on an artificial intelligence optimization algorithm is long in optimization time and large in calculation amount, and facilitates the popularization and application of the algorithm.

Description

Technical Field

[0001] The present invention relates to the technical field of signal processing, and in particular to a residual current noise reduction method based on adaptive MVMD. Background Art

[0002] Grid-connected inverters for renewable energy sources, such as wind and photovoltaic power generation, can introduce harmonic pollution into distribution networks. Furthermore, the presence of numerous nonlinear loads in distribution networks further complicates harmonics, impacting the accuracy of residual current and voltage data extraction and the reliability of residual current assurance devices. Among existing noise reduction methods, wavelet transforms (WTs) possess time-frequency localization capabilities and can match signal characteristics by selecting different basis functions (such as Daubechies and Morlet). However, these methods require pre-setting of the basis functions, resulting in poor adaptability and energy leakage and frequency aliasing. Empirical mode decomposition (EMD) does not require a basis function and adaptively decomposes nonlinear and nonstationary signals, intuitively generating IMF components. However, these methods lack rigorous mathematical proofs, suffer from severe modal aliasing and endpoint effects, are sensitive to noise, and exhibit poor robustness. Variational mode decomposition (VMD) decomposes signals based on variational principles and avoids modal aliasing by optimizing the bandwidth of the framework. This results in more stable decomposition and robustness to noise. However, the number of modes K and the penalty parameter α must be manually pre-set, resulting in a computational complexity higher than that of EMD. Among them, to address the problem that decompositions such as VMD and MVMD require the manual pre-setting of the modal number K, most existing methods use intelligent optimization algorithms such as particle swarm optimization (PSO), sparrow search algorithm (SSA), and grey wolf optimization algorithm (GWO) to search for the optimal parameter K through multiple iterations. However, such optimization algorithms have high requirements on computing power, require multiple iterations, take a long time to calculate, and may fall into local optimal solutions, thus limiting their usage scenarios. Summary of the Invention

[0003] 1. Technical problems to be solved:

[0004] In response to the above technical problems, the present invention provides a residual current denoising method based on adaptive MVMD. First, the number of effective IMF components in the signal is determined through REMD pre-decomposition, which solves the problem of selecting the number K of parameter modal decomposition in the MVMD signal decomposition process. After MVMD decomposition, the effective IMF is screened and the signal is reconstructed through the Pearson correlation coefficient method. Finally, the reconstructed signal is subjected to secondary denoising through SVD to obtain the final denoised signal.

[0005] 2. Technical solution:

[0006] A residual current noise reduction method based on adaptive MVMD is characterized by comprising the following steps:

[0007] Step 1: Obtain the residual current acquisition signal in the low-voltage distribution network containing distributed photovoltaics The residual current acquisition signal is obtained as follows:

[0008]

[0009] In the above formula: is the repeated grounding point current; It is the three-phase normal ground leakage current; is the common mode leakage current of the photovoltaic power supply;

[0010] Step 2: The residual current signal is pre-decomposed using robust empirical mode decomposition (REMD) to obtain multiple intrinsic mode components (IMFs) and a residual component (Res). The Pearson correlation coefficient between each intrinsic mode component (IMF) and the residual current signal is calculated, and the multiple intrinsic mode components (IMFs) obtained by pre-decomposition are divided into signal components and noise components according to the Pearson correlation coefficient.

[0011] Step 3: Obtain the number of intrinsic mode component IMF signal components in the REMD pre-decomposition result, and use this number as the parameter modal decomposition number K of multivariate variational modal decomposition MVMD; perform multivariate variational modal decomposition MVMD on the residual current acquisition signal to obtain multiple intrinsic mode components IMF; then obtain the Pearson correlation coefficient between each intrinsic mode component IMF obtained by MVMD decomposition and the residual current acquisition signal, and divide each intrinsic mode component IMF obtained by MVMD decomposition into signal component and noise component according to the size of the Pearson correlation coefficient, and reconstruct all its signal components to obtain a preliminary noise reduction signal

[0012] Step 4: Reconstruct the initial noise reduction signal Then, the singular value decomposition (SVD) is used to perform secondary noise reduction, the singular values ​​reflecting the noise signal are set to zero, and the final noise-reduced residual current signal is reconstructed.

[0013] Furthermore, in step 2, during the robust empirical mode decomposition (REMD) process of the residual current signal, the REMD screening stop conditions include:

[0014] REMD defines the judgment condition of the screening stop condition SSC as f(k):

[0015]

[0016] f(k)=e RMS +E k

[0017] Where: g k (n) is the component of the initial signal obtained after k iterations; n is the signal length; e RMS g k RMS of (n); Ek g k Excess kurtosis of (n);

[0018] If the number of zero points and extreme points is equal, or the difference between the two is less than 1, the screening stops; if the threshold conditions f(k-2)<f(k-1) and f(k-1)<f(k) are met, the screening stops and the (k-2)th decomposition result is output, otherwise the decomposition continues until the maximum number of iterations is reached;

[0019] The robust empirical mode decomposition REMD process is shown as follows:

[0020]

[0021] R RS =G i [n]

[0022]

[0023] In the above formula: x[n] is the initial signal; G i [n] is the new initial signal; R RS is the signal residual that cannot be decomposed; i is the number of IMFs obtained after REMD decomposition.

[0024] Furthermore, in step 3, determining the number of modal decompositions K specifically includes:

[0025] The calculation described in step 2 is to divide the multiple intrinsic mode components IMF obtained by pre-decomposition into signal components and noise components according to the size of the Pearson correlation coefficient. Specifically, the Pearson correlation coefficient between the multiple intrinsic mode components IMF obtained by REMD decomposition and the residual current acquisition signal is calculated; a first correlation strength threshold is preset, and the number of components corresponding to the multiple intrinsic mode components IMF whose Pearson correlation coefficients are greater than the first correlation strength threshold is obtained. This number is used as the modal decomposition number K of MVMD.

[0026] Furthermore, as described in step three; the residual current acquisition signal is decomposed by multivariate variational modal decomposition (MVMD) to obtain multiple intrinsic modal components (IMF). This is an optimization problem under constraints. The Lagrangian operator λ and the second-order penalty factor α are introduced to transform the MVMD into an unconstrained variational problem. The augmented Langrange function is constructed, and the penalty operator and the alternating direction multiplier method are used to continuously update the optimization problem, and the modal u is continuously updated. k,c (t), center frequency ω k and the Lagrangian factor λ c (t), until the final IMF component that satisfies the constraints is obtained.

[0027] Furthermore, the MVMD decomposition process in step 3 specifically includes:

[0028] S31: Initialize the intrinsic mode function Center frequency Lagrange multiplication operator λ 1 , let the number of cycles n = 0; the number of decomposition modes k = 1;

[0029] S32: Let n=n+1, and execute the iterative loop;

[0030] S33: According to the following formula, the eigenmode function Center frequency ω k To update:

[0031]

[0032] The subscript c in the above formula represents the corresponding number of channels, and the total number of channels is C; “←” represents update;

[0033] S34: k=k+1, repeat step S33 until k=K;

[0034] S35: Update the Lagrangian factor λ according to the following formula:

[0035]

[0036] In the above formula, τ represents the time step; xc represents the signal corresponding to channel c;

[0037] S36: Repeat S32 to S35 until the convergence accuracy ε is met, terminate the iteration, and obtain K intrinsic mode components IMF.

[0038] Furthermore, in step 3, the K intrinsic mode components IMF obtained by MVMD decomposition are respectively calculated with the Pearson correlation coefficient between them and the residual current acquisition signal, and the IMF with a strength greater than the preset second correlation strength threshold is selected for signal reconstruction to obtain the preliminary noise-reduced residual current signal.

[0039] Furthermore, the first correlation strength threshold and the second correlation strength threshold are equal, both being 0.1.

[0040] Furthermore, the residual current noise reduction system includes an original signal acquisition module, a signal preprocessing module, a noise reduction module, and a secondary noise reduction module;

[0041] The original signal acquisition module is used to obtain the residual current acquisition signal of the low-voltage distribution network;

[0042] The signal preprocessing module performs REMD pre-decomposition on the residual current signal, obtains the number of effective intrinsic mode components (IMFs) through the Pearson correlation coefficient method, and optimizes the number of parameter modal decompositions (K) of MVMD;

[0043] The noise reduction module determines the number of modal decompositions K in the MVMD decomposition based on the number of valid IMFs in the signal preprocessing module, performs MVMD decomposition on the residual current acquisition signal, screens the valid IMFs using the Pearson correlation coefficient method, and reconstructs the signal.

[0044] The secondary noise reduction module performs SVD secondary noise reduction on the signal reconstructed by MVMD decomposition, screens valid singular values ​​according to the Hankel matrix, and obtains the final noise-reduced and reconstructed residual current signal through inverse transformation.

[0045] Furthermore, the invention is applied to a computer device comprising a memory, a processor and a stored computer program.

[0046] 3.Beneficial effects:

[0047] (1) The present invention proposes a residual current noise reduction method based on adaptive MVMD. Aiming at the situation where a large number of distributed power sources and nonlinear loads are connected in the distribution network, which leads to a large amount of noise interference introduced into the measured residual current signal, it proposes to use REMD pre-decomposition to obtain the parameter modal decomposition number K in MVMD decomposition, so as to solve the difficulty of manually setting the accurate parameter K.

[0048] (2) Compared with the noise reduction algorithm based on artificial intelligence optimization algorithm, the residual current noise reduction method based on adaptive MVMD proposed in the present invention does not require multiple searches to find the optimal parameters. The number of parameter modal decompositions K in the MVMD decomposition can be obtained through only one REMD pre-decomposition, which reduces the amount of calculation in the algorithm process and can greatly reduce the time required for signal noise reduction.

[0049] (3) The present invention proposes a residual current noise reduction method based on adaptive MVMD, which obtains K intrinsic mode components IMF through MVMD decomposition. By calculating their Pearson correlation coefficients, the correlation coefficient values ​​greater than the threshold setting are valid IMF components, and those lower than the threshold setting are noise components. The effective IMF components are reconstructed to achieve the purpose of reducing noise interference.

[0050] (4) The present invention proposes a residual current denoising method based on adaptive MVMD. After the initial denoising is performed based on MVMD decomposition, the reconstructed signal is subjected to secondary denoising by SVD. The denoising effect is good. Compared with traditional wavelet transform WT, empirical mode decomposition EMD, variational mode decomposition VMD, etc., the correlation with the original signal is as high as over 99%. BRIEF DESCRIPTION OF THE DRAWINGS

[0051] Figure 1 It is the overall flow chart of the present invention;

[0052] Figure 2 This is the IMF result diagram of REMD decomposition;

[0053] Figure 3 This is the IMF result diagram of MVMD decomposition;

[0054] Figure 4 The figure shows the comparison between the final noise-reduced signal and the noise-free signal. DETAILED DESCRIPTION

[0055] The present invention will be described in detail below with reference to the accompanying drawings.

[0056] As attached Figure 1 A residual current noise reduction method based on adaptive MVMD is shown, characterized in that it includes the following steps:

[0057] Step 1: Obtain the residual current acquisition signal in the low-voltage distribution network containing distributed photovoltaics The residual current acquisition signal is obtained as follows:

[0058]

[0059] In the above formula: is the repeated grounding point current; It is the three-phase normal ground leakage current; is the common mode leakage current of the photovoltaic power supply;

[0060] Step 2: The residual current signal is pre-decomposed using robust empirical mode decomposition (REMD) to obtain multiple intrinsic mode components (IMFs) and a residual component (Res). The Pearson correlation coefficient between each intrinsic mode component (IMF) and the residual current signal is calculated, and the multiple intrinsic mode components (IMFs) obtained by pre-decomposition are divided into signal components and noise components according to the Pearson correlation coefficient.

[0061] Step 3: Obtain the number of intrinsic mode component IMF signal components in the REMD pre-decomposition result, and use this number as the parameter modal decomposition number K of multivariate variational modal decomposition MVMD; perform multivariate variational modal decomposition MVMD on the residual current acquisition signal to obtain multiple intrinsic mode components IMF; then obtain the Pearson correlation coefficient between each intrinsic mode component IMF obtained by MVMD decomposition and the residual current acquisition signal, and divide each intrinsic mode component IMF obtained by MVMD decomposition into signal component and noise component according to the size of the Pearson correlation coefficient, and reconstruct all its signal components to obtain a preliminary noise reduction signal

[0062] Step 4: Reconstruct the initial noise reduction signal Then, the singular value decomposition (SVD) is used to perform secondary noise reduction, the singular values ​​reflecting the noise signal are set to zero, and the final noise-reduced residual current signal is reconstructed.

[0063] Furthermore, in step 2, during the robust empirical mode decomposition (REMD) process of the residual current signal, the REMD screening stop conditions include:

[0064] REMD defines the judgment condition of the screening stop condition SSC as f(k):

[0065]

[0066]

[0067] f(k)=e RMS +E k

[0068] Where: g k (n) is the component of the initial signal obtained after k iterations; n is the signal length; e RMS g k RMS of (n); E k g k Excess kurtosis of (n);

[0069] If the number of zero points and extreme points is equal, or the difference between the two is less than 1, the screening stops; if the threshold conditions f(k-2)<f(k-1) and f(k-1)<f(k) are met, the screening stops and the (k-2)th decomposition result is output, otherwise the decomposition continues until the maximum number of iterations is reached;

[0070] The robust empirical mode decomposition REMD process is shown as follows:

[0071]

[0072] R RS =G i [n]

[0073]

[0074] In the above formula: x[n] is the initial signal; G i [n] is the new initial signal; R RS is the signal residual that cannot be decomposed; i is the number of IMFs obtained after REMD decomposition.

[0075] Furthermore, in step 3, determining the number of modal decompositions K specifically includes:

[0076] The calculation described in step 2 is to divide the multiple intrinsic mode components IMF obtained by pre-decomposition into signal components and noise components according to the size of the Pearson correlation coefficient. Specifically, the Pearson correlation coefficient between the multiple intrinsic mode components IMF obtained by REMD decomposition and the residual current acquisition signal is calculated; a first correlation strength threshold is preset, and the number of components corresponding to the multiple intrinsic mode components IMF whose Pearson correlation coefficients are greater than the first correlation strength threshold is obtained. This number is used as the modal decomposition number K of MVMD.

[0077] Furthermore, as described in step three; the residual current acquisition signal is decomposed by multivariate variational modal decomposition (MVMD) to obtain multiple intrinsic modal components (IMF). This is an optimization problem under constraints. The Lagrangian operator λ and the second-order penalty factor α are introduced to transform the MVMD into an unconstrained variational problem. The augmented Langrange function is constructed, and the penalty operator and the alternating direction multiplier method are used to continuously update the optimization problem, and the modal u is continuously updated. k,c (t), center frequency ω k and the Lagrangian factor λ c (t), until the final IMF component that satisfies the constraints is obtained.

[0078] Furthermore, the MVMD decomposition process in step 3 specifically includes:

[0079] S31: Initialize the intrinsic mode function Center frequency Lagrange multiplication operator λ 1 , let the number of cycles n = 0; the number of decomposition modes k = 1;

[0080] S32: Let n=n+1, and execute the iterative loop;

[0081] S33: According to the following formula, the eigenmode function Center frequency ω k To update:

[0082]

[0083] The subscript c in the above formula represents the corresponding number of channels, and the total number of channels is C; “←” represents update;

[0084] S34: k=k+1, repeat step S33 until k=K;

[0085] S35: Update the Lagrangian factor λ according to the following formula:

[0086]

[0087] In the above formula, τ represents the time step; xc represents the signal corresponding to channel c;

[0088] S36: Repeat S32 to S35 until the convergence accuracy ε is met, terminate the iteration, and obtain K intrinsic mode components IMF.

[0089] Furthermore, in step 3, the K intrinsic mode components IMF obtained by MVMD decomposition are respectively calculated with the Pearson correlation coefficient between them and the residual current acquisition signal, and the IMF with a strength greater than the preset second correlation strength threshold is selected for signal reconstruction to obtain the preliminary noise-reduced residual current signal.

[0090] Furthermore, the first correlation strength threshold and the second correlation strength threshold are equal, both being 0.1.

[0091] Furthermore, the residual current noise reduction system includes an original signal acquisition module, a signal preprocessing module, a noise reduction module, and a secondary noise reduction module;

[0092] The original signal acquisition module is used to obtain the residual current acquisition signal of the low-voltage distribution network;

[0093] The signal preprocessing module performs REMD pre-decomposition on the residual current signal, obtains the number of effective intrinsic mode components (IMFs) through the Pearson correlation coefficient method, and optimizes the number of parameter modal decompositions (K) of MVMD;

[0094] The noise reduction module determines the number of modal decompositions K in the MVMD decomposition based on the number of valid IMFs in the signal preprocessing module, performs MVMD decomposition on the residual current acquisition signal, screens the valid IMFs using the Pearson correlation coefficient method, and reconstructs the signal.

[0095] The secondary noise reduction module performs SVD secondary noise reduction on the signal reconstructed by MVMD decomposition, screens valid singular values ​​according to the Hankel matrix, and obtains the final noise-reduced and reconstructed residual current signal through inverse transformation.

[0096] Furthermore, the invention is applied to a computer device comprising a memory, a processor and a stored computer program.

[0097] Verification example:

[0098] The analog signal x(t) used in this verification example is shown in the following equation. This residual current signal contains the fundamental wave, the third, fifth, and seventh odd harmonics, the second and sixth even harmonics, and a 110Hz interharmonic. n(t) is 20dB Gaussian white noise. The sampling frequency is set to 10kHz, the sampling duration is 0.3s, and the number of sampling points is 3000.

[0099] x(t)=3sin(2π×50t+0.78)+0.45sin(2π×100t+0.35)+0.19sin(2π×110t+0.55)+

[0100] 0.52sin(2π×150t+0.69)+0.41sin(2π×250t+1.01)+0.33sin(2π×300t+0.38)+

[0101] 0.28sin(2π×350t+0.41)+n(t)

[0102] Using REMD for pre-decomposition, the screening stop judgment condition is defined as f(k), which should be as small as possible. In this embodiment, the screening stop SSC process of REMD is modified as follows:

[0103] (1) The number of zero points and extreme points is equal, or the difference between the two numbers is less than 1;

[0104] (2) When the threshold conditions f(k-2)<f(k-1) and f(k-1)<f(k) are met, the screening stops and the (k-2)th decomposition result is output, otherwise the decomposition continues until the maximum number of iterations is reached.

[0105] The eight intrinsic mode components IMF obtained by REMD decomposition are as follows Figure 2 As shown, the Pearson correlation coefficient was calculated for each of them, and the results are shown in Table 1.

[0106] Table 1 IMF component correlation coefficient values

[0107]

[0108]

[0109] In this embodiment, the first correlation strength threshold and the second correlation strength threshold are both preset to 0.1.

[0110] As shown in Table 1, IMF1 is a high-frequency noise component, and the correlation coefficients of IMF2 to IMF5 all exceed the preset value of 0.1, which can be regarded as effective IMF components. Therefore, the decomposition parameter of MVMD is determined to be 4.

[0111] In step 3, the specific steps of MVMD are as follows:

[0112] (1) Use Hilbert transform to calculate the analytical vector U corresponding to each IMF component U(t) + (t):

[0113]

[0114] Where: H is the Hilbert operator; δ(t) is the impulse function; * represents convolution.

[0115] (2) Move the frequency band of each IMF component to the corresponding baseband:

[0116]

[0117] (3) Calculate the analytical vector U + (t) The L2 norm of the gradient obtains the estimated bandwidth of the IMF component, and the multivariable function is:

[0118]

[0119] Where: is the partial derivative operation at time t; ω k is the center frequency of each IMF component.

[0120] Then The spectrum shift of the corresponding channel ω k And find the Frobenius norm, then f is expressed as:

[0121]

[0122] Where: is the analytical signal corresponding to mode k in channel c.

[0123] The constrained optimization problem of the constructed MVMD decomposition is as follows:

[0124]

[0125] Where: {u k,c (t)} and {ω k} are the sets of multivariate signals and corresponding signal center frequencies in channel c respectively.

[0126] According to the above formula, MVMD is decomposed into an optimization problem under constraints. Therefore, the Lagrangian operator λ and the second-order penalty factor α are introduced to transform MVMD into an unconstrained variational problem. The corresponding augmented Langrange function is:

[0127]

[0128] Where: <·,·> represents the inner product.

[0129] In order to simplify it into multiple sub-optimization problems, the penalty operator and alternating direction multiplication method are used to continuously update the complex optimization problem of the above formula. as well as get:

[0130]

[0131] Where: “←” represents update; n is the number of iterations; τ is the time step.

[0132] Convert the formula AAA into:

[0133]

[0134] Each IMF component is converted into the frequency domain through Fourier transform:

[0135]

[0136] Solution to the optimization problem The update in the frequency domain is expressed as:

[0137]

[0138] According to Plancherel's theorem, the center frequency The update is expressed as:

[0139]

[0140] Assuming the first-order derivative of the above formula is 0, we have:

[0141]

[0142] The convergence accuracy ε criterion is:

[0143]

[0144] In this embodiment, the decomposition parameter results of MVMD are as follows: Figure 3 As shown in the figure, MVMD decomposes the residual current signal into three IMF components and one residual component Res. The correlation coefficient values ​​of each IMF component and the residual current acquisition signal are calculated as shown in Table 2. As can be seen from Table 2, the correlation coefficient values ​​of IMF1 to IMF3 components are all greater than 0.1, and the residual Res is a high-frequency noise component.

[0145] Table 2 IMF component correlation coefficient values

[0146]

[0147] In step 4, the denoising method based on singular value decomposition (SVD) first obtains the singular values ​​σ of the Hankel matrix and sorts them by size to obtain:

[0148]

[0149] where σ p+1And the subsequent singular value values ​​gradually approach 0. According to the singular value data, the constructor is constructed and the second-order derivative is obtained. If

[0150]

[0151] This indicates that the matrix singular values ​​​​mutate before and after the critical point p. The singular values ​​reflecting the effective signal are retained, and the smaller singular values ​​reflecting the noise signal are set to zero. Then the inverse transform is performed to obtain the signal after noise reduction and reconstruction. The result is as follows Figure 4 As shown in the figure, the blue line represents the original signal without noise, and the red dashed line represents the reconstructed signal after noise reduction. It can be seen that the residual current waveform reconstructed in this embodiment is largely consistent with the original signal, preserving the original residual current signal characteristics and achieving good noise reduction effect.

[0152] Although the present invention has been disclosed above in terms of preferred embodiments, they are not intended to limit the present invention. Anyone skilled in the art can make various changes or modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be based on the scope of protection defined by the claims of this application.

Claims

1. A residual current noise reduction method based on adaptive MVMD, characterized by: The following steps are involved: Step 1: Obtain the residual current acquisition signal in the low-voltage distribution network containing distributed photovoltaics The residual current acquisition signal is obtained as follows: In the above formula: is the repeated grounding point current; It is the three-phase normal ground leakage current; is the common mode leakage current of the photovoltaic power supply; Step 2: The residual current signal is pre-decomposed using robust empirical mode decomposition (REMD) to obtain multiple intrinsic mode components (IMFs) and a residual component (Res). The Pearson correlation coefficient between each intrinsic mode component (IMF) and the residual current signal is calculated, and the multiple intrinsic mode components (IMFs) obtained by pre-decomposition are divided into signal components and noise components according to the Pearson correlation coefficient. Step 3: Obtain the number of intrinsic mode component IMF signal components in the REMD pre-decomposition result, and use this number as the parameter modal decomposition number K of multivariate variational modal decomposition MVMD; perform multivariate variational modal decomposition MVMD on the residual current acquisition signal to obtain multiple intrinsic mode components IMF; then obtain the Pearson correlation coefficient between each intrinsic mode component IMF obtained by MVMD decomposition and the residual current acquisition signal, and divide each intrinsic mode component IMF obtained by MVMD decomposition into signal component and noise component according to the size of the Pearson correlation coefficient, and reconstruct all its signal components to obtain a preliminary noise reduction signal Step 4: Reconstruct the initial noise reduction signal Then, the singular value decomposition (SVD) is used to perform secondary noise reduction, the singular values ​​reflecting the noise signal are set to zero, and the final noise-reduced residual current signal is reconstructed.

2. The method for reducing residual current noise based on adaptive MVMD according to claim 1, characterized in that: In step 2, during the robust empirical mode decomposition (REMD) process of the residual current signal, the REMD screening stop conditions include: REMD defines the judgment condition of the screening stop condition SSC as f(k): Where: g k (n) is the component of the initial signal obtained after k iterations; n is the signal length; e RMS g k RMS of (n); E k g k Excess kurtosis of (n); If the number of zero points and extreme points is equal, or the difference between the two is less than 1, the screening stops; if the threshold conditions f(k-2)<f(k-1) and f(k-1)<f(k) are met, the screening stops and the (k-2)th decomposition result is output, otherwise the decomposition continues until the maximum number of iterations is reached; The robust empirical mode decomposition REMD process is shown as follows: In the above formula: x[n] is the initial signal; G i [n] is the new initial signal; R RS is the signal residual that cannot be decomposed; i is the number of IMFs obtained after REMD decomposition.

3. The method for reducing residual current noise based on adaptive MVMD according to claim 2, characterized in that: In step 3, determining the number of modal decompositions K specifically includes: The multiple intrinsic mode components IMFs obtained by pre-decomposition as described in step 2 are divided into signal components and noise components according to the size of the Pearson correlation coefficient. Specifically, the Pearson correlation coefficients between the multiple intrinsic mode components IMFs obtained by REMD decomposition and the residual current acquisition signal are calculated; a first correlation strength threshold is preset, and the number of components corresponding to the multiple intrinsic mode components IMFs whose Pearson correlation coefficients are greater than the first correlation strength threshold is obtained. This number is used as the modal decomposition number K of MVMD in step 3.

4. The method for reducing residual current noise based on adaptive MVMD according to claim 3, characterized in that: As described in step three; perform multivariate variational modal decomposition (MVMD) on the residual current acquisition signal to obtain multiple intrinsic modal components (IMF). This is an optimization problem under constraints. The Lagrangian operator λ and the second-order penalty factor α are introduced to transform the MVMD into an unconstrained variational problem. The augmented Langrange function is constructed, and the penalty operator and the alternating direction multiplier method are used to continuously update the optimization problem, and the modal u is continuously updated. k,c (t), center frequency ω k and the Lagrangian factor λ c (t), until the final IMF component that satisfies the constraints is obtained.

5. The method for reducing residual current noise based on adaptive MVMD according to claim 4, characterized in that: The MVMD decomposition process in step 3 specifically includes: S31: Initialize the intrinsic mode function Center frequency Lagrange multiplication operator λ 1 , let the number of cycles n = 0; the number of decomposition modes k = 1; S32: Let n=n+1, and execute the iterative loop; S33: According to the following formula, the eigenmode function Center frequency ω k To update: The subscript c in the above formula represents the corresponding number of channels, and the total number of channels is C; "←" represents update; S34: k=k+1, repeat step S33 until k=K; S35: Update the Lagrangian factor λ according to the following formula: In the above formula, τ represents the time step; x c represents the signal corresponding to channel c; S36: Repeat S32 to S35 until the convergence accuracy ε is met, terminate the iteration, and obtain K intrinsic mode components IMF.

6. The method for reducing residual current noise based on adaptive MVMD according to claim 5, characterized in that: In step 3, the K intrinsic mode components IMF obtained by MVMD decomposition are respectively calculated with the Pearson correlation coefficient between them and the residual current acquisition signal, and the IMF with a strength greater than the preset second correlation strength threshold is selected for signal reconstruction to obtain the preliminary noise-reduced residual current signal.

7. The method for reducing residual current noise based on adaptive MVMD according to claim 6, characterized in that: The first correlation strength threshold and the second correlation strength threshold are equal to each other, both being 0.

1.

8. The method for reducing residual current noise based on adaptive MVMD according to claim 1, characterized in that: Its residual current noise reduction system includes an original signal acquisition module, a signal preprocessing module, a noise reduction module, and a secondary noise reduction module; the original signal acquisition module is used to obtain the residual current acquisition signal of the low-voltage distribution network; The signal preprocessing module performs REMD pre-decomposition on the residual current signal, obtains the number of effective intrinsic mode components (IMFs) through the Pearson correlation coefficient method, and optimizes the number of parameter modal decompositions (K) of MVMD; The noise reduction module determines the number of modal decompositions K in the MVMD decomposition based on the number of valid IMFs in the signal preprocessing module, performs MVMD decomposition on the residual current acquisition signal, screens the valid IMFs using the Pearson correlation coefficient method, and reconstructs the signal. The secondary denoising module performs SVD secondary denoising on the signal reconstructed by MVMD decomposition, screens the effective singular values ​​according to the Hankel matrix, and obtains the final denoised and reconstructed residual current signal through inverse transformation.

9. The method for reducing residual current noise based on adaptive MVMD according to claim 1, characterized in that: The invention is applied to a computer device, which includes a memory, a processor and a stored computer program.

Images