Transformer sound signal processing method and system based on improved ICA algorithm

Abstract

The invention discloses a transformer sound signal processing method and system based on an improved ICA algorithm, and the method comprises the steps: inputting a noise-containing transformer sound signal, and obtaining a mixed to-be-processed signal matrix; the method comprises the following steps: decomposing an underdetermined mixed to-be-processed audio signal matrix based on an EEMD algorithm improved based on a minimum lower limit frequency to obtain IMF components and carrying out preliminary denoising, estimating a source signal dimension and a dimension-raised sound signal matrix dimension based on a voiceprint dimension identification model of a GRU network, judging whether the dimension-raised sound signal matrix is not underdetermined, if the dimension-raised sound signal matrix is not underdetermined, extracting a residual error, and if the dimension-raised sound signal matrix is not underdetermined, extracting the residual error; and if not, screening the dimension raising signal group based on a correlation maximum principle, and improving an ICA algorithm component blind source signal based on N-MNE. The sound signal processing effect of the transformer can be effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of power systems, and in particular relates to a transformer acoustic signal processing method and system based on an improved ICA algorithm. Background Art

[0002] Transformers are key equipment in high-voltage, long-distance power transmission and transformation systems in substations. At the same time, transformers are also one of the main noise sources in substations. During operation, the acoustic signals generated by mechanical vibrations of transformer windings, cores, oil tanks and other structures contain rich state characteristics. Transformer faults are closely related to voiceprints, and the abnormality of transformer body acoustic signals is highly correlated with its own faults. Moreover, compared with power-off detection such as transition resistance method, voiceprint recognition has become a research hotspot for scholars at home and abroad in recent years due to its real-time, convenience and economy. Carrying out transformer voiceprint detection and fault diagnosis is of great significance to ensure its safe operation. The microphone array is designed to collect transformer acoustic signals. The transformer working condition detection and diagnosis can be realized by analyzing the acoustic signals. However, in the process of acoustic signal acquisition of power transformers, there are often a large number of interfering noise signals such as ambient noise, corona discharge, and bird calls, which makes it impossible to obtain pure transformer operating state acoustic signals efficiently and quickly, resulting in frequent misjudgments of the fault state monitoring system.

[0003] Blind source separation (BSS) is a technique that recovers source signals only by observing signals without prior knowledge of passive signals and aliasing coefficients. Due to its advancedness and superiority, it has become a research hotspot for scholars at home and abroad in recent years. When the number of microphone array elements is less than the number of noise source signals, the BSS problem is called underdetermined blind source separation (UBSS). Due to its wider application, UBSS has become a research hotspot in the field of signal processing internationally. As a common algorithm for solving the blind source separation problem, the FastICA (fast independent component analysis) algorithm has good separation accuracy under non-underdetermined conditions. However, in actual conditions, the types of noise sources are generally random and complex (up to dozens of types), and the number of noise sensors is often underdetermined. How to meet the reversibility constraints of the mixing process and the uniqueness constraints of the separated signal of the FastICA algorithm in the blind source separation of transformer voiceprints has become a key issue in solving the underdetermined blind source separation of transformer voiceprints. Summary of the invention

[0004] The present invention provides a transformer acoustic signal processing method and system based on an improved ICA algorithm, which are used to solve the technical problem of underdetermined blind source separation of transformer voiceprint.

[0005] In a first aspect, the present invention provides

[0006] In a second aspect, the present invention provides

[0007] In a third aspect, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can perform the steps of the transformer acoustic signal processing method based on the improved ICA algorithm according to any embodiment of the present invention.

[0008] In a fourth aspect, the present invention further provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor executes the steps of the transformer acoustic signal processing method based on the improved ICA algorithm according to any embodiment of the present invention.

[0009] The transformer acoustic signal processing method and system based on the improved ICA algorithm of the present application inputs a noisy transformer acoustic signal to obtain a mixed signal matrix to be processed; based on the minimum lower frequency improved EEMD algorithm, the underdetermined mixed audio signal matrix to be processed is decomposed to obtain IMF components and perform preliminary denoising; based on the GRU network-based voiceprint dimension recognition model, the source signal dimension is estimated, the dimension of the acoustic signal matrix is ​​increased in dimension, and it is judged whether the increased-dimensional signal matrix is ​​not underdetermined; if it is underdetermined, the residual is extracted; if it is not underdetermined, the increased-dimensional signal group is screened based on the maximum correlation principle; based on the N-MNE improved ICA algorithm component blind source signal, the processing effect of the transformer acoustic signal can be effectively improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0011] Figure 1 A flowchart of a transformer acoustic signal processing method based on an improved ICA algorithm provided by an embodiment of the present invention;

[0012] Figure 2 A structural block diagram of a transformer acoustic signal processing system based on an improved ICA algorithm provided by an embodiment of the present invention;

[0013] Figure 3 It is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0014] 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.

[0015] See also Figure 1 , which shows a flow chart of a transformer acoustic signal processing method based on an improved ICA algorithm of the present application.

[0016] like Figure 1 As shown, the transformer acoustic signal processing method based on the improved ICA algorithm specifically includes the following steps:

[0017] Step S101, constructing a mixed signal matrix to be processed according to the acquired noisy transformer signal.

[0018] Step S102: Decompose the mixed signal matrix to be processed by using an improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, and perform preliminary denoising on the at least one IMF component to obtain at least one target IMF component.

[0019] In this step, the short-time Fourier transform power spectrum analysis is performed on the noisy transformer sound signal collected by the i-th microphone to find the minimum frequency in the noisy transformer sound signal as the cutoff frequency f of its corresponding signal decomposition. L , and use it as the lower limit of signal decomposition.

[0020] Adding standard Gaussian white noise to the noisy transformer signal, a mixed noise signal is obtained, which is expressed as:

[0021] M i (t) = m i (t)+n w (t),

[0022] In the formula, m i (t) is the i-th noisy transformer acoustic signal, n w (t) is standard Gaussian white noise, M i (t) is the i-th mixed noise signal;

[0023] Perform EEMD decomposition on the mixed noise signal to obtain the IMF components and residuals, expressed as:

[0024]

[0025] In the formula, I i，j(t) is the jth order IMF component obtained by decomposing the i-th noisy transformer acoustic signal, r i (t) is the residual obtained by EEMD decomposition of the i-th noisy transformer acoustic signal, and J is the order of the IMF component group;

[0026] For the decomposed I i，j (t) Perform short-time Fourier transform power spectrum analysis to obtain I i，j (t) target frequency f max , the target frequency f max With the cut-off frequency f L In comparison, if the target frequency f max Less than the cut-off frequency f L , the decomposition ends, otherwise, the decomposition continues, and finally the N-order IMF component of the i-th noisy transformer acoustic signal after EEMD decomposition is obtained;

[0027] After averaging each IMF component and offsetting the influence of Gaussian white noise, the overall average of each order IMF component after decomposition of the noisy transformer acoustic signal is obtained. The expression is:

[0028]

[0029] In the formula, C k，j To calculate the overall average of the j-th order IMF components of K-times decomposition, C i，j is the j-th order IMF component of the i-th noisy transformer acoustic signal, M is the number of noisy transformer acoustic signals;

[0030] Calculate the residual r(t) obtained after EEMD decomposition of the noisy transformer acoustic signal. The expression is:

[0031]

[0032] The correlation coefficients of each order IMF component and the corresponding noisy transformer acoustic signal are calculated respectively, and the expression is:

[0033]

[0034] In the formula, ρ k is the correlation coefficient between the k-th order IMF component and the corresponding transformer acoustic signal containing noise, C k (t) is the kth order IMF component, C abs is the average value of the k-th order IMF component, x k (t) is the transformer sound signal, x abs is the average value of the transformer acoustic signal, K′ is the order of the IMF component;

[0035] Set the threshold μ, retain the IMF components with correlation coefficient values ​​greater than the threshold μ, and obtain at least one target IMF component. The expression for calculating the threshold μ is:

[0036]

[0037] In the formula, is k The average value of .

[0038] It should be noted that after the mixed signal matrix to be processed is decomposed by the improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, a voiceprint order recognition model based on the GRU network is constructed, and the dimension of the noisy transformer signal is identified according to the voiceprint order recognition model, and it is determined whether the dimension of the noisy transformer signal is greater than the dimension of a certain IMF component, wherein the expression for calculating the dimension of the noisy transformer signal is:

[0039] S e =GRU(m i (t)),

[0040] In the formula, m i (t) is the i-th noisy transformer sound signal, GRU is the gated recurrent unit, S e is the dimension of the noisy transformer signal;

[0041] If it is greater than the dimension of a certain IMF component, the residual corresponding to the certain IMF component is decomposed again based on the improved EEMD algorithm;

[0042] If it is not greater than the dimension of a certain IMF component, the certain IMF component is subjected to preliminary denoising processing.

[0043] Specifically, through the transformer voiceprint simulation laboratory, the on-site transformer noise-free fault voiceprint, single interference source fault voiceprint, two interference source fault voiceprint and more than two noise interference source fault voiceprint are simulated, and the voiceprint dimension identification model database is constructed based on this.

[0044] Since transformer acoustic signals have strong temporal characteristics, we consider using a recurrent neural network that is good at analyzing the temporal relationship between discrete data sequences. The GRU network is a simplified long short-term memory (LSTM) network. The input gate and forget gate in the LSTM network are merged into an update gate to reduce the number of parameters and alleviate overfitting. While improving computational efficiency, it maintains a prediction effect close to that of LSTM. The specific process of building a voiceprint dimension recognition model includes:

[0045] The voiceprint dimension recognition model based on the GRU network is trained using the voiceprint dimension recognition model database. During the model training stage, the voiceprint dimension recognition model and the voiceprint dimension prediction model are built respectively, and the training hyperparameters are set, including batch size, learning rate, and number of iterations, to verify the recognition accuracy of the model and the prediction error of the prediction model.

[0046] The reset gate gating signal r of the GRU network t The expression of ′ is:

[0047] r′ t =Sig(W r X t +U r h t-1 ),

[0048] Where W r To reset the gate connection weight matrix, X t Input for the current time, h t-1 is the hidden gate output at time t-1, U r is the connection weight matrix between the reset gate and the hidden state of the previous time step, and the features related to the reset gate are extracted from the hidden state of the previous time step. Sig is the Sigmoid function;

[0049] The update gate gating signal Z of the GRU network t The expression is:

[0050] Z t =Sig{W Z [h t-1 , x t ]},

[0051] Where W Z Connect the weight matrix for the update gate;

[0052] c t Represents the memory information at time t, and the expression is:

[0053] c t =tanh(W c X t +U c (r t ′⊙h t-1 )),

[0054] Where W c and U c All represent learnable weight parameters, ⊙ represents dot product;

[0055] Among them, h t is the hidden gate output at time t, expressed as:

[0056] ht =z t h t-1 +(1-z t )c t .

[0057] Step S103, based on the principle of maximum correlation, select the top S with the greatest correlation with the noisy transformer signal. e The target IMF components of order constitute the raised-order signal group D u , where S e is the dimension of the noisy transformer signal.

[0058] In this step, the expression for calculating the correlation is:

[0059]

[0060] In the formula, cov(·) is the covariance function, σ(·) is the standard deviation function, is the correlation coefficient, m(t) is the noisy transformer signal, C i，j (t) is the j-th order IMF component of the i-th noisy transformer acoustic signal;

[0061] The dimensionality-enhanced signal group is equivalent to the original underdetermined voiceprint signal group obtained after linear transformation. Since there is no nonlinear process, using the IMF signal group as input for blind source separation will not destroy the linearity requirement of the FastICA algorithm, and the dimensionality-enhanced signal group solves the reversibility constraint of the mixed signal and the uniqueness constraint of the separated signal, solving the key problem in the underdetermined blind source separation of transformer voiceprint signals.

[0062] Furthermore, the step-up signal group D u The expression is:

[0063]

[0064] Where C1(t) is the first IMF component with the largest correlation of the noisy transformer signal, and C2(t) is the second IMF component with the largest correlation of the noisy transformer signal. The first Sth signal with the largest correlation to the noisy transformer e IMF components.

[0065] Step S104: based on the preset N-MNE-ICA model, the up-order signal group D u Perform blind source separation to obtain blind source signal.

[0066] In this step, since the traditional negative entropy calculation method is too difficult and the accuracy of the empirical formula is not high enough, it is necessary to improve it, that is, the independent component analysis algorithm (N-MNE-ICA) based on Newton Limit-Maximum Negative Entropy (N-MNE). The signal group after dimensionality increase is input into the ICA algorithm module based on N-MNE improvement to reduce the computational complexity of negative entropy, improve the convergence speed, and increase the separation accuracy.

[0067] The maximum entropy principle assumes that no unknown assumptions are made about information, and that unknown events are considered to be equally probable events. Substituting the maximum entropy principle into the negative entropy formula, we can obtain the maximum approximate negative entropy, which is expressed as:

[0068]

[0069] In the formula, J max (y) is the maximum approximate negative entropy, N and E are the total number of samples and mathematical expectation, respectively, k i is the weight factor, which is a positive constant, G i It is a non-quadratic function, and both y and v are variables with mean 0 and variance 1;

[0070] In order to obtain y = B T The maximum approximate negative entropy value of x is obtained by E{G i (y)} makes the non-Gaussianity of y reach the maximum, and the expression of the constraint condition is:

[0071]

[0072] Where B is the convergence threshold function, B T is the transpose of the convergence threshold function B, χ is the input signal, J G is the input signal, G′ i is the derivative of a non-quadratic function, J G is a non-Gaussian measure function based on the input signal, and β′ is a Lagrange multiplier operator used for constrained optimization problems.

[0073] The maximum entropy theory and the negative entropy formula have proven that there must be an optimal solution that can make the constraint equation valid. Assuming that the optimal solution is B0, solving the constraint conditions, we get:

[0074]

[0075] In the formula, is the transpose of the optimal solution;

[0076] According to Newton's limit theorem, the tangent of a curve can be used to approximate the curve, so it is also approximate to solve the root of the tangent. If the accuracy requirement is not met, continue to execute Newton's method until convergence. The general formula of Newton's limit method is:

[0077]

[0078] In the formula, x n is the estimated value of the nth iteration, x n+1 is the estimated value of the n+1th iteration, f(x n ) is the Newton limit function, f′(x n ) is the derivative of Newton's limiting function.

[0079] The iterative formula of Newton's limit method is:

[0080]

[0081] In the formula, B * is the iterative function, G" i is the second-order derivative of the non-quadratic function, and x is the input signal;

[0082] The application of Newton's limit method not only simplifies complex calculations such as solving probability density functions, but also reduces the error accumulation that may be caused by high-order power calculations in empirical formulas. This method transforms the originally complex statistical mathematical structure into a relatively simple iterative root-finding structure, thereby improving the accuracy and efficiency of the calculation.

[0083] In order to facilitate the subsequent blind source separation calculation, the de-averaging process should be performed before the calculation. After de-averaging, the data is distributed near the coordinate axis, and the function obtained by random initialization can approach the target function faster. The de-averaging expression is:

[0084]

[0085] Where Z is the transformer voiceprint signal sample after de-averaging, α is the eigenvector of the transformer voiceprint signal sample X, and λ is the eigenvalue of X.

[0086] Set up the iteration function B * , the expression is:

[0087] B * =E{ZG′ i (B T Z)}-E{G″ i (B T Z)}B,

[0088] Define the convergence threshold function B, the expression is:

[0089]

[0090] It should be noted that, based on the preset N-MNE-ICA model, the up-order signal group D u In the process of blind source separation, the signal similarity coefficient is used as an evaluation index to evaluate the separation result. The expression for calculating the signal similarity coefficient is:

[0091]

[0092] In the formula, ξ ij is the signal similarity coefficient between the i-th signal in the α group and the j-th signal in the β group, ξ ij = [0, 1], when ξ ij =1, it is considered that the i-th signal is completely consistent with the j-th signal, ξ ij =0, on the contrary, α i (t) is the i-th signal in group α, β j (t) is the jth signal in the β group, ξ(α i , β j ) is α i (t) and β j (t), N′ is the upper limit of the acoustic signal time t.

[0093] In summary, the method of the present application inputs a noisy transformer acoustic signal to obtain a mixed signal matrix to be processed; based on the minimum lower frequency improved EEMD algorithm, the underdetermined mixed audio signal matrix to be processed is decomposed to obtain IMF components and perform preliminary denoising, the source signal dimension is estimated based on the voiceprint dimension recognition model of the GRU network, the dimension of the acoustic signal matrix is ​​increased in dimension, and it is judged whether the increased-dimensional signal matrix is ​​not underdetermined, if it is underdetermined, the residual is extracted, if it is not underdetermined, the increased-dimensional signal group is screened based on the maximum correlation principle, and the blind source signal of the component of the ICA algorithm is improved based on the N-MNE, which can effectively improve the processing effect of the transformer acoustic signal.

[0094] See also Figure 2 , which shows a structural block diagram of a transformer acoustic signal processing system based on an improved ICA algorithm of the present application.

[0095] like Figure 2 As shown, the transformer acoustic signal processing system 200 includes a construction module 210 , a decomposition module 220 , a screening module 230 and a separation module 240 .

[0096] Among them, the construction module 210 is configured to construct a mixed signal matrix to be processed according to the acquired noisy transformer signal; the decomposition module 220 is configured to decompose the mixed signal matrix to be processed by using an improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, and to perform preliminary denoising on the at least one IMF component to obtain at least one target IMF component; the screening module 230 is configured to screen the front S with the greatest correlation with the noisy transformer signal based on the maximum correlation principle. e The target IMF components of order constitute the raised-order signal group D u , where S e is the dimension of the noisy transformer signal; the separation module 240 is configured to separate the raised-order signal group D based on the preset N-MNE-ICA model u Perform blind source separation to obtain blind source signal.

[0097] It should be understood that Figure 2 Modules and references documented in Figure 1 Therefore, the operations and features described above for the method and the corresponding technical effects are also applicable to Figure 2 The modules in it will not be described in detail here.

[0098] In some other embodiments, the embodiments of the present invention further provide a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor is caused to execute the transformer acoustic signal processing method based on the improved ICA algorithm in any of the above method embodiments;

[0099] As an implementation mode, the computer-readable storage medium of the present invention stores computer-executable instructions, and the computer-executable instructions are configured as follows:

[0100] A mixed signal matrix to be processed is constructed according to the obtained noisy transformer signal;

[0101] Decomposing the mixed signal matrix to be processed by using an improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, and performing preliminary denoising on the at least one IMF component to obtain at least one target IMF component;

[0102] Based on the maximum correlation principle, the front S with the greatest correlation with the noisy transformer signal is selected. e The target IMF components of order constitute the raised-order signal group D u , where S e is the dimension of the noisy transformer signal;

[0103] Based on the preset N-MNE-ICA model, the up-order signal group D uPerform blind source separation to obtain blind source signal.

[0104] The computer-readable storage medium may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application required by at least one function; the data storage area may store data created according to the use of the transformer acoustic signal processing system based on the improved ICA algorithm, etc. In addition, the computer-readable storage medium may include a high-speed random access memory, and may also include a memory, such as at least one disk storage device, a flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include a memory remotely arranged relative to the processor, and these remote memories may be connected to the transformer acoustic signal processing system based on the improved ICA algorithm via a network. Examples of the above-mentioned network include, but are not limited to, the Internet, an intranet, a local area network, a mobile communication network, and combinations thereof.

[0105] Figure 3 is a schematic diagram of the structure of an electronic device provided by an embodiment of the present invention, such as Figure 3 As shown, the device includes: a processor 310 and a memory 320. The electronic device may also include: an input device 330 and an output device 340. The processor 310, the memory 320, the input device 330 and the output device 340 may be connected via a bus or other means. Figure 3 In the example, the bus connection is used. The memory 320 is the above-mentioned computer-readable storage medium. The processor 310 executes various functional applications and data processing of the server by running the non-volatile software programs, instructions and modules stored in the memory 320, that is, the transformer sound signal processing method based on the improved ICA algorithm of the above-mentioned method embodiment is realized. The input device 330 can receive input digital or character information, and generate key signal input related to user settings and function control of the transformer sound signal processing system based on the improved ICA algorithm. The output device 340 may include display devices such as display screens.

[0106] The electronic device can execute the method provided by the embodiment of the present invention, and has the functional modules and beneficial effects corresponding to the execution method. For technical details not described in detail in this embodiment, please refer to the method provided by the embodiment of the present invention.

[0107] As an implementation mode, the electronic device is applied to a transformer acoustic signal processing system based on an improved ICA algorithm, and is used for a client, and includes: at least one processor; and a memory connected to the at least one processor in communication; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor so that the at least one processor can:

[0108] A mixed signal matrix to be processed is constructed according to the obtained noisy transformer signal;

[0109] Decomposing the mixed signal matrix to be processed by using an improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, and performing preliminary denoising on the at least one IMF component to obtain at least one target IMF component;

[0110] Based on the maximum correlation principle, the front S with the greatest correlation with the noisy transformer signal is selected. e The target IMF components of order constitute the ascending order signal group D u , where S e is the dimension of the noisy transformer signal;

[0111] Based on the preset N-MNE-ICA model, the up-order signal group D u Perform blind source separation to obtain blind source signal.

[0112] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods of each embodiment or some parts of the embodiment.

[0113] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A transformer acoustic signal processing method based on an improved ICA algorithm, characterized in that: include: A mixed signal matrix to be processed is constructed according to the obtained noisy transformer signal; Decomposing the mixed signal matrix to be processed by using an improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, and performing preliminary denoising on the at least one IMF component to obtain at least one target IMF component; Based on the maximum correlation principle, the front S with the greatest correlation with the noisy transformer signal is selected. e The target IMF components of order constitute the ascending order signal group D u , where S e is the dimension of the noisy transformer signal; Based on the preset N-MNE-ICA model, the up-order signal group D u Perform blind source separation to obtain blind source signal.

2. The transformer acoustic signal processing method based on the improved ICA algorithm according to claim 1 is characterized in that: The step of using the improved EEMD algorithm based on the minimum lower frequency limit to decompose the mixed signal matrix to obtain at least one IMF component includes: Adding standard Gaussian white noise to the noisy transformer signal, a mixed noise signal is obtained, which is expressed as: M i (t)=m i (t)+n w (t), In the formula, m i (t) is the i-th noisy transformer acoustic signal, n w (t) is standard Gaussian white noise, M i (t) is the i-th mixed noise signal; Perform EEMD decomposition on the mixed noise signal to obtain the IMF components and residuals, expressed as: In the formula, I i，j (t) is the jth order IMF component obtained by decomposing the i-th noisy transformer acoustic signal, r i (t) is the residual obtained by EEMD decomposition of the i-th noisy transformer acoustic signal, and J is the order of the IMF component group; For the decomposed I i，j (t) Perform short-time Fourier transform power spectrum analysis to obtain I i，j (t) target frequency f max , the target frequency f max With the cut-off frequency f L In comparison, if the target frequency f max Less than the cut-off frequency f L , the decomposition ends, otherwise, the decomposition continues, and finally the N-order IMF component of the i-th noisy transformer acoustic signal after EEMD decomposition is obtained; After averaging each IMF component and offsetting the influence of Gaussian white noise, the overall average of each order IMF component after decomposition of the noisy transformer acoustic signal is obtained. The expression is: In the formula, C k，j To calculate the overall average of the j-th order IMF components of K-times decomposition, C i，j is the j-th order IMF component of the i-th noisy transformer acoustic signal, M is the number of noisy transformer acoustic signals; Calculate the residual r(t) obtained after EEMD decomposition of the noisy transformer acoustic signal. The expression is:

3. The transformer acoustic signal processing method based on the improved ICA algorithm according to claim 2 is characterized in that: The performing preliminary denoising on the at least one IMF component to obtain at least one target IMF component comprises: The correlation coefficients of each order IMF component and the corresponding noisy transformer acoustic signal are calculated respectively, and the expression is: Where ρx is the correlation coefficient between the k-th order IMF component and the corresponding noisy transformer acoustic signal, C k (t) is the kth order IMF component, C abs is the average value of the k-th order IMF component, x k (t) is the transformer sound signal, x abs is the average value of the transformer acoustic signal, K′ is the order of the IMF component; Set the threshold μ, retain the IMF components with correlation coefficient values ​​greater than the threshold μ, and obtain at least one target IMF component. The expression for calculating the threshold μ is: In the formula, is k The average value of .

4. The transformer acoustic signal processing method based on the improved ICA algorithm according to claim 1 is characterized in that: After decomposing the mixed signal matrix to be processed by using the improved EEMD algorithm based on the minimum lower limit frequency to obtain at least one IMF component, the method further includes: A voiceprint order recognition model based on the GRU network is constructed, and the dimension of the noisy transformer signal is identified according to the voiceprint order recognition model, and it is determined whether the dimension of the noisy transformer signal is greater than the dimension of a certain IMF component, wherein the expression for calculating the dimension of the noisy transformer signal is: S e =GRU(m i (t)), In the formula, m i (t) is the i-th noisy transformer sound signal, GRU is the gated recurrent unit, S e is the dimension of the noisy transformer signal; If it is greater than the dimension of a certain IMF component, the residual corresponding to the certain IMF component is decomposed again based on the improved EEMD algorithm; If it is not greater than the dimension of a certain IMF component, the certain IMF component is subjected to preliminary denoising processing.

5. The transformer acoustic signal processing method based on the improved ICA algorithm according to claim 1 is characterized in that: The step-up signal group D u The expression is: Where C1(t) is the first IMF component with the largest correlation of the noisy transformer signal, and C2(t) is the second IMF component with the largest correlation of the noisy transformer signal. The first Sth signal with the largest correlation to the noisy transformer e IMF components.

6. The transformer acoustic signal processing method based on the improved ICA algorithm according to claim 1 is characterized in that: in, Based on the preset N-MNE-ICA model, the up-order signal group D u In the process of blind source separation, the signal similarity coefficient is used as an evaluation index to evaluate the separation result. The expression for calculating the signal similarity coefficient is: In the formula, ξ ij is the signal similarity coefficient between the i-th signal in the α group and the j-th signal in the β group, ξ ij = [0, 1], when ξ ij =1, it is considered that the i-th signal is completely consistent with the j-th signal, ξ ij =0, on the contrary, α i (t) is the i-th signal in group α, β j (t) is the jth signal in the β group, ξ(α i , β j ) is α i (t) and β j (t), N′ is the upper limit of the acoustic signal time t.

7. A transformer acoustic signal processing system based on an improved ICA algorithm, characterized in that: include: A construction module configured to construct a mixed signal matrix to be processed according to the acquired noisy transformer signal; A decomposition module is configured to decompose the mixed signal matrix to be processed by adopting an improved EEMD algorithm based on a minimum lower limit frequency to obtain at least one IMF component, and to perform preliminary denoising on the at least one IMF component to obtain at least one target IMF component; A screening module is configured to screen the front S with the greatest correlation with the noisy transformer signal based on the principle of maximum correlation. e The target IMF components of order constitute the ascending order signal group D u , where S e is the dimension of the noisy transformer signal; A separation module is configured to separate the raised-order signal group D based on a preset N-MNE-ICA model. u Perform blind source separation to obtain blind source signal.

8. An electronic device, characterized in that: include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the method described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the program is executed by a processor, the method according to any one of claims 1 to 6 is implemented.

Images