Transformer fault voiceprint diagnosis sample enhancement method and system and storage medium

Abstract

The invention provides a transformer fault voiceprint diagnosis sample enhancement method, a transformer fault voiceprint diagnosis sample enhancement system and a storage medium. The method comprises the following steps: constructing an acoustic wave equation of vibration of an iron core and a winding in medium propagation and a medium interface boundary condition; performing sound field simulation on the fault model according to the acoustic wave equation and the medium interface boundary condition to obtain a fault simulation sound field, and determining a physical simulation data set according to the fault simulation sound field; carrying out sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and carrying out transfer learning on a real fault sample to obtain a transfer dictionary; and performing sample enhancement according to the sparse dictionary, the sparse coefficient matrix and the migration dictionary to obtain a transformer fault voiceprint enhancement sample. According to the embodiment of the invention, the sample enhancement is carried out through the sparse dictionary, the sparse coefficient matrix and the migration dictionary, and the sample enhancement effect of the transformer fault sample can be effectively realized.

Description

Technical Field

[0001] The present invention relates to the technical field of fault diagnosis, and in particular to a method, system and storage medium for transformer fault voiceprint diagnosis sample enhancement. Background Art

[0002] Fault detection technology for oil-immersed transformers has long relied on methods such as oil chromatography and partial discharge detection, which suffer from limitations such as delayed response and complex operation. In recent years, fault monitoring based on voiceprint analysis has become a research hotspot due to its non-invasive and real-time capabilities. It analyzes transformer vibration sound waves to identify internal mechanical faults (such as winding deformation and core loosening).

[0003] In the existing transformer vibration acoustic wave fault identification process, due to the scarcity of transformer fault samples, the data-driven fault identification model is prone to overfitting. Therefore, how to enhance the transformer fault samples has become an urgent problem to be solved. Summary of the Invention

[0004] The purpose of the embodiments of the present invention is to provide a transformer fault voiceprint diagnosis sample enhancement method, system and storage medium to solve the problem of how to enhance transformer fault samples.

[0005] The embodiment of the present invention is implemented as follows: a method for enhancing transformer fault voiceprint diagnosis samples, the method comprising: Establishing a transformer core winding vibration mechanism model, and performing electromagnetic force correction on the transformer core winding vibration mechanism model to obtain a transformer core winding vibration model; Performing fault simulation on the transformer core winding vibration model to obtain a fault model, and constructing an acoustic wave equation and a medium interface boundary condition for the vibration of the core and winding in the medium propagation; Performing an acoustic field simulation on the fault model according to the acoustic wave equation and the medium interface boundary condition to obtain a fault simulation sound field, and determining a physical simulation data set according to the fault simulation sound field; Performing sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and performing transfer learning on real fault samples to obtain a transfer dictionary; Sample enhancement is performed according to the sparse dictionary, the sparse coefficient matrix, and the migration dictionary to obtain a transformer fault voiceprint enhancement sample.

[0006] Preferably, the magnetostriction coefficient of the core in the transformer core winding vibration mechanism model is expressed as: in, represents the normal magnetostriction, represents the saturation magnetostriction, represents the core saturation flux, Indicates the number of turns of the high voltage coil, represents the winding voltage amplitude, represents the effective area of ​​magnetic flux, Indicates the frequency of alternating current, t represents the time variable; The electromagnetic force on the winding coil in the transformer core winding vibration mechanism model is expressed as: in, represents the radius of the winding coil, Indicates the current amplitude loaded in the winding, Represents the proportionality coefficient between leakage flux density and current, Represents the electromagnetic force on the coil; The electromagnetic force formula in the transformer core winding vibration model is expressed as: in, Indicates the correction proportional coefficient of winding deformation, It represents the electromagnetic force acting on the coil in the transformer core winding vibration model.

[0007] Preferably, the fault model includes a winding deformation fault model, a core loosening fault model and a winding displacement fault model; The winding deformation fault model is expressed as: in, represents the deformation sensitivity factor, Indicates the relative deformation of the winding radius; The core loosening fault model is expressed as: in, Indicates the magnetostriction after loosening, represents the core loose coupling coefficient, Indicates the gap between the laminations, Indicates the thickness of silicon steel sheet; The winding displacement fault model is expressed as: in, represents the cross-sectional area of ​​the core, represents the elastic modulus of silicon steel, represents the strain-stress conversion coefficient, Indicates the core mass, represents the stiffness coefficient, is the core vibration displacement.

[0008] Preferably, the acoustic wave equation is expressed as: in, Represents the winding, Indicates insulating oil, Indicates the mechanical housing, Indicates the The sound pressure in the medium, Indicates the The speed of sound in a medium, Indicates the The viscosity coefficient of the medium, Indicates the The density of the medium, represents the Laplace operator, represents the derivative; The medium interface boundary conditions include a solid-liquid interface sound pressure continuity condition, a normal particle velocity continuity condition, and a liquid-solid interface boundary condition that is an acoustic impedance constraint; The solid-liquid interface sound pressure continuity condition is expressed as: in, represents the sound pressure of the incident sound wave, represents the sound pressure of the reflected sound wave, Indicates the sound pressure of the transmitted sound wave; The normal particle velocity continuity condition is expressed as: in, represents the total sound pressure of the first medium, represents the total sound pressure of the second medium, represents the normal direction of the first medium pointing to the first medium; The liquid-solid interface boundary condition is an acoustic impedance constraint expressed as: in, represents the angular frequency of the sound wave, Represents the acoustic impedance of the outer surface of the mechanical housing, Indicates the total sound pressure of the third medium.

[0009] Preferably, determining a physical simulation data set according to the fault simulation sound field includes: The acoustic field reconstruction model is established based on the equivalent source method, and the surface of the transformer tank is discretized into equivalent sound source areas. Acquiring a measurement point signal in the fault simulation sound field, and calculating a correlation coefficient between the equivalent sound source signal and the measurement point signal; A target acquisition point in the measurement point signal is determined according to the correlation coefficient, and data of the target acquisition point in the fault simulation sound field is acquired to obtain the physical simulation data set.

[0010] Preferably, the formula used to calculate the correlation coefficient between the equivalent sound source signal and the measurement point signal includes: in, represents the anti-zero correction term, Indicates the frequency amplitude of the equivalent sound source signal, Indicates the frequency amplitude of the measurement point signal, Indicates the n The equivalent sound source signal and the m The correlation coefficient between the signals of the measurement points is Indicates the m The signal of each measurement point at all frequencies f The mean on Indicates the n Equivalent sound source signals at all frequency points f The mean on Indicates the total number of frequency points.

[0011] Preferably, the physical simulation data set is sparsely represented to obtain a sparse dictionary and a sparse coefficient matrix, and transfer learning is performed on real fault samples to obtain a transfer dictionary, including: performing DC removal and bandpass filtering on the physical simulation data set, and performing wavelet packet decomposition on the physical simulation data set after the bandpass filtering to obtain sub-band coefficients; splicing the sub-band coefficients to obtain an initial dictionary, and iteratively optimizing the initial dictionary to obtain the sparse dictionary and the sparse coefficient matrix; Performing sparse reconstruction on the real fault sample according to the sparse dictionary to obtain sample coefficients, and calculating reconstruction residuals according to the sample coefficients; Performing credible labeling on the real fault sample according to the reconstructed residual to obtain a credible labeled sample, and determining a sparse coefficient set according to the credible labeled sample; The sparse dictionary is used as an initial value of the dictionary, and the initial value of the dictionary is iteratively updated according to the credible labeled samples and the sparse coefficient set to obtain the migration dictionary.

[0012] Preferably, performing sample enhancement according to the sparse dictionary, the sparse coefficient matrix and the migration dictionary to obtain a transformer fault voiceprint enhancement sample includes: Generating candidate enhancement samples according to the sparse coefficient matrix and the migration dictionary, and calculating the residual of the candidate enhancement samples under the sparse dictionary to obtain an enhancement residual; The candidate enhanced samples are screened according to the enhanced residual to obtain a screened sample, and the screened sample is mixed with the credible labeled sample to obtain the transformer fault voiceprint enhanced sample.

[0013] Another object of an embodiment of the present invention is to provide a transformer fault voiceprint diagnosis sample enhancement system, the system comprising: A model building module is used to establish a transformer core winding vibration mechanism model and perform electromagnetic force correction on the transformer core winding vibration mechanism model to obtain a transformer core winding vibration model; A fault simulation module is used to perform fault simulation on the transformer core winding vibration model to obtain a fault model and construct an acoustic wave equation and a medium interface boundary condition for the vibration of the core and winding in the medium propagation; a fault simulation module, configured to perform an acoustic field simulation on the fault model according to the acoustic wave equation and the medium interface boundary condition to obtain a fault simulation sound field, and determine a physical simulation data set according to the fault simulation sound field; A sparse migration module is used to perform sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and to perform migration learning on real fault samples to obtain a migration dictionary; The sample enhancement module is used to perform sample enhancement according to the sparse dictionary, the sparse coefficient matrix and the migration dictionary to obtain a transformer fault voiceprint enhancement sample.

[0014] The embodiments of the present invention can effectively simulate the fault state of the transformer by performing fault simulation on the vibration model of the transformer core winding. By constructing the acoustic wave equation of the vibration of the core and winding in the medium propagation and the medium interface boundary conditions, the sound field simulation of the fault model is effectively guaranteed. Based on the fault simulation sound field, the physical simulation data set can be effectively determined. By performing sparse representation on the physical simulation data set, a sparse dictionary and a sparse coefficient matrix can be effectively obtained. By performing transfer learning on real fault samples, a transfer dictionary can be effectively obtained. By performing sample enhancement through the sparse dictionary, sparse coefficient matrix and transfer dictionary, the sample enhancement effect of the transformer fault sample can be effectively achieved. BRIEF DESCRIPTION OF THE DRAWINGS

[0015] Figure 1 This is a flow chart of a transformer fault voiceprint diagnosis sample enhancement method provided by the first embodiment of the present invention; Figure 2This is a schematic diagram of a specific implementation of the transformer fault voiceprint diagnosis sample enhancement method provided by the first embodiment of the present invention; Figure 3 2 is a schematic diagram of the structure of a transformer fault voiceprint diagnosis sample enhancement system provided by the second embodiment of the present invention; Figure 4 2 is a schematic diagram of the structure of a transformer fault voiceprint diagnosis sample enhancement system provided by the third embodiment of the present invention; Figure 5 It is a structural diagram of a terminal device provided by the fifth embodiment of the present invention. DETAILED DESCRIPTION

[0016] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0017] In order to illustrate the technical solution of the present invention, specific embodiments are provided below.

[0018] Example 1 See also Figures 1 to 2 , is a flow chart of a transformer fault voiceprint diagnosis sample enhancement method provided by the first embodiment of the present invention. The transformer fault voiceprint diagnosis sample enhancement method can be applied to any device or system. The transformer fault voiceprint diagnosis sample enhancement method includes the following steps: Step S10, establishing a transformer core winding vibration mechanism model, and performing electromagnetic force correction on the transformer core winding vibration mechanism model to obtain a transformer core winding vibration model; Among them, a transformer core winding vibration mechanism model is established in the multi-physics field simulation software. Considering the influence of winding deformation on leakage flux, the transformer core winding vibration mechanism model is corrected by electromagnetic force to obtain the transformer core winding vibration model.

[0019] In this step, the transformer's core is made of laminated silicon steel sheets. When the alternating magnetic flux passes through the core, the silicon steel sheets will produce periodic expansion and contraction deformation due to the magnetostrictive effect. The electromagnetic force between the core leg and the iron yoke, and between the winding and the core, will change periodically with the change of the alternating current. The magnetostriction coefficient of the core in the transformer core winding vibration mechanism model is expressed as: in, represents the normal magnetostriction, represents the saturation magnetostriction, represents the core saturation flux, Indicates the number of turns of the high voltage coil, represents the winding voltage amplitude, represents the effective area of ​​magnetic flux, Indicates the frequency of alternating current, t represents the time variable; Under the influence of various factors such as electromagnetic force and mechanical force, the geometric shape, size or position of the transformer winding changes. In the transformer core winding vibration mechanism model, the electromagnetic force acting on the winding coil is expressed as: in, represents the radius of the winding coil, Indicates the current amplitude loaded in the winding, Represents the proportionality coefficient between leakage flux density and current, Represents the electromagnetic force acting on the coil.

[0020] The electromagnetic force formula in the transformer core winding vibration model is expressed as: in, Indicates the correction proportional coefficient of winding deformation, It represents the electromagnetic force acting on the coil in the transformer core winding vibration model.

[0021] Step S20, performing fault simulation on the transformer core winding vibration model to obtain a fault model, and constructing an acoustic wave equation for the vibration of the core and winding in the medium propagation and a medium interface boundary condition; Among them, based on the transformer core winding vibration model, the winding deformation fault is introduced to build a winding deformation fault model, the core loosening fault is introduced to build a core loosening fault model, the winding displacement fault is introduced to build a winding displacement fault model, and the acoustic wave equation of the vibration propagation of the core and winding in the "solid-liquid-solid" multi-media and the medium interface boundary conditions are established; Optionally, the fault model includes a winding deformation fault model, a core loosening fault model and a winding displacement fault model; When the transformer winding undergoes mechanical deformation, its geometric structural parameters (such as coil radius, turn spacing, etc.) change, resulting in significant changes in the leakage magnetic field distribution characteristics and electromagnetic force amplitude-frequency characteristics. The winding deformation fault model is expressed as: in, represents the deformation sensitivity factor, Indicates the relative deformation of the winding radius; The looseness between the core laminations leads to the weakening of magnetostrictive strain constraint. The core looseness fault model is expressed as: in, Indicates the magnetostriction after loosening, represents the core loose coupling coefficient, Indicates the gap between the laminations, Indicates the thickness of silicon steel sheet; The winding displacement fault model is expressed as: in, represents the cross-sectional area of ​​the core, represents the elastic modulus of silicon steel, represents the strain-stress conversion coefficient, Indicates the core mass, represents the stiffness coefficient, is the core vibration displacement.

[0022] In this step, the sound source of the transformer winding and core vibration is transmitted through the oil and mechanical casing to the air. Its propagation medium is a "solid-liquid-solid" multi-medium transmission. The attenuation, impedance, and refraction of the soundprint vary when propagating in different media. Establishing the acoustic wave method and boundary conditions at the medium interface is the key to accurately analyzing the fault soundprint of the oil-immersed transformer. The acoustic wave equation is expressed as: in, Represents the winding, Indicates insulating oil, Indicates the mechanical housing, Indicates the The sound pressure in the medium, Indicates the The speed of sound in a medium, Indicates the The viscosity coefficient of the medium, Indicates the The density of the medium, represents the Laplace operator, represents the derivative; The medium interface boundary conditions include a solid-liquid interface sound pressure continuity condition, a normal particle velocity continuity condition, and a liquid-solid interface boundary condition that is an acoustic impedance constraint; The solid-liquid interface sound pressure continuity condition is expressed as: in, represents the sound pressure of the incident sound wave, represents the sound pressure of the reflected sound wave, Indicates the sound pressure of the transmitted sound wave; The normal particle velocity continuity condition is expressed as: in, represents the total sound pressure of the first medium, represents the total sound pressure of the second medium, represents the normal direction of the first medium pointing to the first medium; The liquid-solid interface boundary condition is an acoustic impedance constraint expressed as: in, represents the angular frequency of the sound wave, Represents the acoustic impedance of the outer surface of the mechanical housing, Indicates the total sound pressure of the third medium.

[0023] Step S30, performing an acoustic field simulation on the fault model according to the acoustic wave equation and the medium interface boundary condition to obtain a fault simulation sound field, and determining a physical simulation data set according to the fault simulation sound field; Among them, the finite element simulation software is used to simulate the winding deformation fault model, winding looseness fault model, and winding displacement fault model under the boundary conditions of the acoustic wave equation and the medium interface, and the winding deformation fault sound field, winding looseness fault sound field, and winding displacement fault sound field are generated. The physical simulation data set is determined based on the winding deformation fault sound field, winding looseness fault sound field, and winding displacement fault sound field.

[0024] Optionally, determining a physical simulation data set according to the fault simulation sound field includes: The acoustic field reconstruction model is established based on the equivalent source method, and the surface of the transformer tank is discretized into equivalent sound source areas. Acquiring a measurement point signal in the fault simulation sound field, and calculating a correlation coefficient between the equivalent sound source signal and the measurement point signal; Determining a target acquisition point in the measurement point signal according to the correlation coefficient, and acquiring data of the target acquisition point in the fault simulation sound field to obtain the physical simulation data set; Among them, the maximum correlation measurement point is selected in the sound field of the winding deformation fault, and the first sub-timing diagram is output; the maximum correlation measurement point is selected in the sound field of the winding looseness fault, and the second sub-timing diagram is output; the maximum correlation measurement point is selected in the sound field of the winding displacement fault, and the third sub-timing diagram is output. The first sub-timing diagram, the second sub-timing diagram, and the third sub-timing diagram are combined to obtain the physical simulation data set.

[0025] Furthermore, the formula used to calculate the correlation coefficient between the equivalent sound source signal and the measurement point signal includes: in, represents the anti-zero correction term, Indicates the frequency amplitude of the equivalent sound source signal, Indicates the frequency amplitude of the measurement point signal, Indicates the n The equivalent sound source signal and the m The correlation coefficient between the signals of the measurement points is Indicates the m The signal of each measurement point at all frequencies f The mean on Indicates the n Equivalent sound source signals at all frequency points f The mean on Indicates the total number of frequency points.

[0026] Step S40, performing sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and performing transfer learning on real fault samples to obtain a transfer dictionary; Optionally, performing sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and performing transfer learning on real fault samples to obtain a transfer dictionary, including: performing DC removal and bandpass filtering on the physical simulation data set, and performing wavelet packet decomposition on the physical simulation data set after the bandpass filtering to obtain sub-band coefficients; Among them, the physical simulation data set is preprocessed by removing DC and bandpass filtering to ensure that the length of each sample vector in the physical simulation data set is the same; in, represents the physical simulation dataset, represents the first sample vector in the physical simulation dataset, Indicates the length of the sample vector.

[0027] splicing the sub-band coefficients to obtain an initial dictionary, and iteratively optimizing the initial dictionary to obtain the sparse dictionary and the sparse coefficient matrix; Among them, the K-Singular Value Decomposition (K-SVD) algorithm is used to iteratively optimize the initial dictionary to solve the sparse representation problem: in, represents a sparse dictionary, represents a sparse coefficient matrix, represents the number of non-zero elements in the vector, represents the constrained sparsity, represents the first sparse coefficients, represents the variable value that minimizes the objective function. represents the combined weight of the sparse coding vector reflecting the fault components, Indicates limiting conditions. represents the square root of the sum of the squares of all elements of the Frobenius matrix.

[0028] The iterative process of K-SVD involves first fixing a sparse dictionary for each sample vector and solving for sparse coefficients through orthogonal matching pursuit (OMP). Each column of the sparse dictionary is then optimized column by column, and each column is updated by performing SVD decomposition on the local residual matrix until the iteration converges. After the iteration is complete, the final sparse dictionary and its sparse coefficient moments are output. Preferably, in order to eliminate the energy difference between different sparse coefficients, each column vector of the sparse coefficient matrix is ​​normalized according to the norm: Performing sparse reconstruction on the real fault sample according to the sparse dictionary to obtain sample coefficients, and calculating reconstruction residuals according to the sample coefficients; The real fault samples are preprocessed to ensure the consistency of sample length between the real fault samples and the physical simulation dataset. The sparse dictionary is used to perform sparse reconstruction on each target domain sample in the real fault samples: in, represents a real fault sample, represents the length of the target domain sample, represents the first target domain sample, represents the jth sample coefficient, represents the norm constraint threshold, represents the jth target domain sample.

[0029] Optionally, the formula used to calculate the reconstruction residual according to the sample coefficient includes: in, represents the reconstruction residual; Performing credible labeling on the real fault sample according to the reconstructed residual to obtain a credible labeled sample, and determining a sparse coefficient set according to the credible labeled sample; The reconstructed residual is compared with the residual threshold, and the target domain samples whose reconstructed residual is less than or equal to the residual threshold are credibly labeled to obtain credible labeled samples, and the sparse coefficients corresponding to the credible labeled samples are obtained to obtain a sparse coefficient set; Using the sparse dictionary as a dictionary initial value, and iteratively updating the dictionary initial value according to the credible labeled sample and the sparse coefficient set to obtain the migration dictionary; Among them, the idea of ​​incremental update is adopted, and the sparse dictionary is used as the initial value of the migration dictionary. The trusted annotation samples are gradually integrated through "online dictionary learning" to obtain the final migration dictionary. Specifically, let: If the number of iterations is less than or equal to the number of credible labeled samples, the credible labeled samples of the current iteration number and their sparse coefficients are selected. 、 、 Represent the coefficient matrix, residual matrix and migration dictionary in the initial state respectively; If the number of iterations is greater than or equal to the number of credible labeled samples, any pair of true fault samples and sample coefficients is recycled or randomly selected; In the In the iteration, the cumulative matrix is ​​updated first: in, and Respectively represent i The accumulated coefficient matrix and the accumulated residual matrix in the iterations, and Respectively represent i -1 The accumulated coefficient matrix and the accumulated residual matrix in the iteration, Indicates the i The sample coefficients, Indicates the i target domain samples, Represents matrix transpose; Then solve the dictionary update subproblem: in, Represents the updated dictionary, represents the trace of the matrix, D represents the dictionary matrix, and C represents the dictionary constraint set.

[0030] The updated dictionary for the current credible annotation sample Fine-tune the sparse coefficients as follows: If the new sparse coefficient Can significantly reduce the reconstruction error, then replace it with the original sparse coefficient , and continue to use it in the next iteration; otherwise, keep the original coefficient.

[0031] Repeat the iteration until the preset number of iterations is reached, and then output the final migration dictionary ,at this time, It not only retains the basic structure of the source domain simulation samples, but also integrates a small number of real sample features in the target domain, which can be used to generate a large number of candidate enhanced samples.

[0032] Step S50, performing sample enhancement according to the sparse dictionary, the sparse coefficient matrix and the migration dictionary to obtain a transformer fault voiceprint enhancement sample; Optionally, performing sample enhancement according to the sparse dictionary, the sparse coefficient matrix, and the migration dictionary to obtain a transformer fault voiceprint enhancement sample includes: Generating candidate enhancement samples according to the sparse coefficient matrix and the migration dictionary, and calculating the residual of the candidate enhancement samples under the sparse dictionary to obtain an enhancement residual; Among them, for each source domain sample corresponding to the sparse coefficient , reconstructed directly under the migration dictionary: in, Represents candidate enhanced samples; this process is called "interaction one", that is, only sparse coefficients and transfer dictionary are used to generate; To further ensure the quality of the generated samples, the residual under the sparse dictionary can be calculated for each candidate enhanced sample: in, represents the enhanced residual.

[0033] Screening the candidate enhanced samples according to the enhanced residual to obtain a screened sample, and mixing the screened sample with the trusted labeled sample to obtain the transformer fault voiceprint enhanced sample; The enhanced residual is compared with the preset residual, and the candidate enhanced samples whose enhanced residual is greater than the preset residual are deleted, and the remaining candidate enhanced samples are determined as screening samples.

[0034] In this embodiment, by performing fault simulation on the transformer core winding vibration model, the fault state of the transformer can be effectively simulated. By constructing the acoustic wave equation of the vibration of the core and winding in the medium propagation and the medium interface boundary conditions, the sound field simulation of the fault model is effectively guaranteed. Based on the fault simulation sound field, the physical simulation data set can be effectively determined. By performing sparse representation on the physical simulation data set, a sparse dictionary and a sparse coefficient matrix can be effectively obtained. By performing transfer learning on real fault samples, a transfer dictionary can be effectively obtained. By performing sample enhancement through the sparse dictionary, sparse coefficient matrix and transfer dictionary, the sample enhancement effect of the transformer fault sample can be effectively achieved.

[0035] Example 2 See also Figure 3 , is a schematic diagram of the structure of a transformer fault voiceprint diagnosis sample enhancement system 100 provided in a second embodiment of the present invention, comprising: The model building module 10 is used to establish a transformer core winding vibration mechanism model and perform electromagnetic force correction on the transformer core winding vibration mechanism model to obtain the transformer core winding vibration model.

[0036] The fault simulation module 11 is used to perform fault simulation on the transformer core winding vibration model to obtain a fault model and construct an acoustic wave equation and a medium interface boundary condition for the vibration of the core and winding in the medium propagation.

[0037] The fault simulation module 12 is configured to perform sound field simulation on the fault model according to the acoustic wave equation and the medium interface boundary conditions to obtain a fault simulation sound field, and determine a physical simulation data set according to the fault simulation sound field.

[0038] The sparse migration module 13 is used to perform sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and perform migration learning on real fault samples to obtain a migration dictionary.

[0039] The sample enhancement module 14 is configured to perform sample enhancement according to the sparse dictionary, the sparse coefficient matrix and the migration dictionary to obtain a transformer fault voiceprint enhancement sample.

[0040] In this embodiment, by performing fault simulation on the transformer core winding vibration model, the fault state of the transformer can be effectively simulated. By constructing the acoustic wave equation of the vibration of the core and winding in the medium propagation and the medium interface boundary conditions, the sound field simulation of the fault model is effectively guaranteed. Based on the fault simulation sound field, the physical simulation data set can be effectively determined. By performing sparse representation on the physical simulation data set, a sparse dictionary and a sparse coefficient matrix can be effectively obtained. By performing transfer learning on real fault samples, a transfer dictionary can be effectively obtained. By performing sample enhancement through the sparse dictionary, sparse coefficient matrix and transfer dictionary, the sample enhancement effect of the transformer fault sample can be effectively achieved.

[0041] Example 3 See also Figure 4 , is a flow chart of a transformer fault voiceprint diagnosis sample enhancement method provided by a third embodiment of the present invention. The transformer fault voiceprint diagnosis sample enhancement method can be applied to any device or system. The transformer fault voiceprint diagnosis sample enhancement method includes the following steps: Step S1: Establish a geometric model of the transformer core in multi-physics simulation software. Based on the principles of electromagnetics and mechanical dynamics, establish a model that describes the mechanical vibration of the transformer core components and windings under the action of magnetostriction and electromagnetic forces; Optionally, in step S1: The transformer's core is made of laminated silicon steel sheets. When alternating magnetic flux passes through the core, the sheets undergo periodic expansion and contraction due to the magnetostrictive effect. The electromagnetic forces between the core legs and yoke, and between the windings and the core, change periodically with the alternating current.

[0042] Step S2: Based on the transformer model established in step S1, introduce winding deformation fault and build winding deformation fault model, introduce core loose fault and build core loose fault model, introduce winding displacement fault and build winding displacement fault model, Step S3: Establishing the acoustic wave equation and boundary conditions for the vibration propagation of the core and winding in the "solid-liquid-solid" multi-medium; Step S4: Using finite element simulation software to simulate the sound field of the fault and obtain a soundprint timing diagram at the measuring point; Optionally, in step S4: Step S4.1, simulating the fault using finite element simulation software under the acoustic wave equation and boundary conditions to simulate the winding deformation fault model, the winding looseness fault model, and the winding displacement fault model to generate the winding deformation fault sound field, the winding looseness fault sound field, and the winding displacement fault sound field; Step S4.2: Obtain the timing diagram of the voiceprint of the measurement point Discretize the transformer tank surface into N equivalent sound source areas; An acoustic field reconstruction model was established based on the equivalent source method. The transformer tank surface was discretized into N equivalent sound source regions. The equivalent sound source signal and the signal from the candidate measurement points were subjected to windowed Fourier transform, and the spectrum matrix was normalized. The maximum correlation measurement point was selected based on the Pearson correlation matrix. The maximum correlation measurement point is selected in the sound field of the winding deformation fault, the soundprint timing diagram is output, and the sound pressure time domain signal is extracted to generate the soundprint timing diagram; the maximum correlation measurement point is selected in the sound field of the winding looseness fault, the soundprint timing diagram is output, and the sound pressure time domain signal is extracted to generate the soundprint timing diagram; the maximum correlation measurement point is selected in the sound field of the winding displacement fault, the soundprint timing diagram is output, and the sound pressure time domain signal is extracted to generate the voiceprint timing diagram.

[0043] Step S5: Fault sound sample transfer enhancement based on sparse dictionaries of the source and target domains: Use source domain knowledge to build a basic dictionary, carefully select reliable samples from the target domain as a bridge, dynamically adjust the feature representation to adapt it to the target domain, and finally use the adjusted feature representation to interactively synthesize a large number of transformer fault voiceprint enhancement samples; Optionally, in step S5, the fault sound sample migration enhancement of the transformer multi-physics field fault sample generation data (source domain data) and the actual transformer collected data (target domain data) specifically includes the following steps: Step S5.1: Source domain sparse dictionary learning and sparse coefficient acquisition First, the clean fault sound sample set generated by simulation is preprocessed by removing DC and performing bandpass filtering to ensure that each sample vector has the same length. Wavelet packet decomposition is then used to perform multi-scale and multi-band decomposition of each sample vector. Sub-band coefficients are generated and concatenated to construct an initial dictionary, which is then iteratively optimized using the K-SVD algorithm.

[0044] Sparse coefficients are obtained through orthogonal matching pursuit (OMP). Each column of the dictionary is then optimized column by column. Each column of atoms is updated by performing SVD decomposition on the local residual matrix until the iteration converges. After the iteration is complete, the final source domain dictionary and its sparse coefficient matrix are output.

[0045] Step S5.2: Screening of trusted labeled samples in the target domain First, a small number of labeled fault sound samples in the target domain are collected and preprocessed in the same way to keep the sample length consistent with the source domain. Then, a sparse dictionary is used to perform sparse reconstruction on each target domain sample and the corresponding reconstruction residual is calculated: Performing credible labeling on the real fault sample according to the reconstructed residual to obtain a credible labeled sample, and determining a sparse coefficient set according to the credible labeled sample; The sparse dictionary is used as an initial value of the dictionary, and the initial value of the dictionary is iteratively updated according to the credible labeled samples and the sparse coefficient set to obtain the migration dictionary.

[0046] Step S5.3: Online iterative update of migration dictionary Adopting the idea of ​​incremental update, the sparse dictionary is used as the initial value of the migration dictionary. Through "online dictionary learning", a small number of credible annotated samples in the target domain are gradually integrated to obtain the final migration dictionary.

[0047] Step S5.4: Interactive sample generation and enhancement in dual target domains Using the source domain sparse coefficient matrix obtained in step S5.1 and the transfer dictionary obtained in step S5.3, a set of pseudo target domain samples can be generated. For each source domain sample, the sparse coefficient vector corresponding to the transfer dictionary is reconstructed to obtain a candidate enhanced sample. This process is called "interaction one", which means that only the source domain coefficients and the transfer dictionary are used. Subsequently, the true and credible labeled samples screened in step S5.2 are mixed with the candidate enhanced samples to construct the enhanced transformer fault voiceprint enhanced sample: This mixing process, called "Interaction II," balances the diversity and authenticity of target domain samples by combining a small number of real samples with a large number of generated samples. To further ensure the quality of generated samples, the residual of each candidate enhancement sample under a sparse dictionary is calculated. Based on the calculated residual, the candidate enhancement samples are screened. The remaining high-quality candidate enhancement samples are then remixed with the real, credible samples to obtain the transformer fault voiceprint enhancement sample.

[0048] Example 4 This is a flowchart of a transformer fault voiceprint diagnosis sample enhancement method provided by the fourth embodiment of the present invention. The transformer fault voiceprint diagnosis sample enhancement method can be applied to any device or system. The transformer fault voiceprint diagnosis sample enhancement method includes the following steps: Step 1. Modeling the vibration mechanism of the core and winding Step 1.1 Calibration of magnetostrictive parameters of the core The saturation magnetostriction, saturation magnetic flux and elastic modulus of silicon steel sheets were measured experimentally; Calculate the corrected magnetostriction based on the gap between the core laminations and the thickness of the silicon steel sheets; Step 1.2: Winding deformation parameter calibration Apply different current amplitudes and measure the winding radius deformation; Fit the deformation sensitivity factor to determine the corrected leakage flux proportional coefficient; Step 2. Multi-physics coupling model construction and simulation Step 2.1: Electromagnetic Field Simulation Setup Import the transformer 3D geometry model into COMSOL and define the winding material properties (conductivity, magnetic permeability); Set the winding current excitation (amplitude I = 100 A, frequency f = 50 Hz) and solve the leakage magnetic field distribution; Step 2.2 Solution of coupling between force field and acoustic field Map electromagnetic force data to the solid mechanics module to define the elastic modulus and damping coefficient of the core and winding; Set the acoustic module medium parameters (structural parts = 7850 kg / m³, insulating oil = 850 kg / m³, shell = 7800 kg / m³); The vibration displacement and sound pressure distribution are solved simultaneously, with the time step set to 1 μs and the total time length being 0.1 s.

[0049] Step 3. Verify simulation results Step 3.1: Extracting sound pressure signals at measurement points Set up 10 measuring points on the surface of the transformer casing to output the sound pressure time domain signal; Perform short-time Fourier transform on the sound pressure time domain signal to generate a voiceprint time-frequency graph.

[0050] Step 3.2: Error Analysis Compare the measured sound pressure data with the simulation results and calculate the average relative error (required to be <5%). If the error exceeds the limit, adjust the mesh density or time step and solve again.

[0051] 2. Cross-domain migration enhancement method for scarce fault voiceprint samples Step 4.1: Fault parameter combination design The winding deformation degree and core looseness level are preset to generate nine fault scenarios. A simulation is run for each scenario, outputting the sound pressure time domain signal and corresponding label.

[0052] Step 4.2: Equivalent sound source screening The tank surface is discretized into 50 equivalent sound source areas; Calculate the Pearson correlation coefficient between each sound source and the alternative measurement points, and select the top five measurement points with the highest correlation as data collection points.

[0053] Step 5. Sparse dictionary transfer learning Step 5.1: Initial dictionary training The simulation data set is divided into frames (frame length 512 points, overlap rate 50%); Use K-SVD algorithm to train the initial dictionary, and the sparsity is set to 10; Extract the shared sparse coefficient matrix.

[0054] Step 5.2: Migrate dictionary optimization Load the actual collected samples and align them with the simulation data after normalization; Fixed the shared sparse coefficient matrix, optimized the target dictionary by gradient descent, and iterated until the loss function converged (usually 100-200 rounds).

[0055] Step 5.3: Enhanced sample generation Reconstruct the signal using the migration dictionary and the shared sparse coefficient matrix; The measured environmental noise (the signal-to-noise ratio is set to 20 dB) is injected into the reconstructed signal to generate an enhanced sample set.

[0056] Step 6. Cross-domain sample verification Step 6.1: Visualize feature distribution Extract MFCC features from simulated data, enhanced data and measured data; Use t-SNE to reduce the dimension to 2D space and verify the overlap of feature distribution (target: cosine similarity between enhanced data and measured data>0.85).

[0057] Step 6.2: Cross-domain identification test Train a support vector machine (SVM) classifier using augmented data for the source domain and 50 measured samples for the target domain; Calculate cross-domain recognition accuracy (goal: improve by 15%~20% compared to pure simulation data).

[0058] Example 5 Figure 5 This is a block diagram of a terminal device 2 provided in the fifth embodiment of the present application. Figure 4 As shown, the terminal device 2 of this embodiment includes: a processor 20, a memory 21, and a computer program 22 stored in the memory 21 and executable on the processor 20, such as a program for a transformer fault voiceprint diagnostic sample enhancement method. When the processor 20 executes the computer program 22, the steps of each of the above-mentioned transformer fault voiceprint diagnostic sample enhancement methods are implemented.

[0059] Exemplarily, the computer program 22 may be divided into one or more modules, which are stored in the memory 21 and executed by the processor 20 to implement the present application. The one or more modules may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution process of the computer program 22 in the terminal device 2. The terminal device may include, but is not limited to, a processor 20 and a memory 21.

[0060] The processor 20 may be a central processing unit (CPU), other general-purpose processors, digital signal processors (DSP), application-specific integrated circuits (ASIC), field-programmable gate arrays (FPGA), other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. The general-purpose processor may be a microprocessor or any conventional processor.

[0061] The memory 21 can be an internal storage unit of the terminal device 2, such as a hard drive or memory of the terminal device 2. The memory 21 can also be an external storage device of the terminal device 2, such as a plug-in hard drive, a Smart Media Card (SMC), a Secure Digital (SD) card, a flash memory card, etc. equipped with the terminal device 2. Furthermore, the memory 21 can include both an internal storage unit of the terminal device 2 and an external storage device. The memory 21 is used to store the computer program and other programs and data required by the terminal device. The memory 21 can also be used to temporarily store data that has been output or is about to be output.

[0062] In addition, the functional modules in the various embodiments of the present application may be integrated into a single processing unit, or each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0063] If the integrated module is implemented as a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. The computer-readable storage medium can be either non-volatile or volatile. Based on this understanding, the present application can implement all or part of the process steps in the above-mentioned method embodiments by instructing the relevant hardware through a computer program. The computer program can be stored in a computer-readable storage medium. When executed by a processor, the computer program can implement the steps of each of the above-mentioned method embodiments. The computer program includes computer program code, which can be in source code form, object code form, executable file, or some intermediate form. Computer-readable storage media can include: any entity or device capable of carrying computer program code, recording medium, USB flash drive, mobile hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signal, telecommunication signal, and software distribution medium. It should be noted that the content contained in computer-readable storage media can be appropriately increased or decreased according to the requirements of legislation and patent practices in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practices, computer-readable storage media do not include electrical carrier signals and telecommunications signals.

[0064] The above-described embodiments are only used to illustrate the technical solutions of the present application, rather than to limit them. Although the present application 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. 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 various embodiments of the present application, and should all be included in the scope of protection of the present application.

Claims

1. A transformer fault voiceprint diagnosis sample enhancement method, characterized in that: The method comprises: Establishing a transformer core winding vibration mechanism model, and performing electromagnetic force correction on the transformer core winding vibration mechanism model to obtain a transformer core winding vibration model; Performing fault simulation on the transformer core winding vibration model to obtain a fault model, and constructing an acoustic wave equation and a medium interface boundary condition for the vibration of the core and winding in the medium propagation; Performing an acoustic field simulation on the fault model according to the acoustic wave equation and the medium interface boundary condition to obtain a fault simulation sound field, and determining a physical simulation data set according to the fault simulation sound field; Performing sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and performing transfer learning on real fault samples to obtain a transfer dictionary; Sample enhancement is performed according to the sparse dictionary, the sparse coefficient matrix, and the migration dictionary to obtain a transformer fault voiceprint enhancement sample.

2. The transformer fault voiceprint diagnosis sample enhancement method according to claim 1, characterized in that: The magnetostriction coefficient of the core in the transformer core winding vibration mechanism model is expressed as: in, represents the normal magnetostriction, represents the saturation magnetostriction, represents the core saturation flux, Indicates the number of turns of the high voltage coil, represents the winding voltage amplitude, represents the effective area of ​​magnetic flux, Indicates the frequency of alternating current, t represents the time variable; The electromagnetic force on the winding coil in the transformer core winding vibration mechanism model is expressed as: in, represents the radius of the winding coil, Indicates the current amplitude loaded in the winding, Represents the proportionality coefficient between leakage flux density and current, Represents the electromagnetic force on the coil; The electromagnetic force formula in the transformer core winding vibration model is expressed as: in, Indicates the correction proportional coefficient of winding deformation, It represents the electromagnetic force acting on the coil in the transformer core winding vibration model.

3. The transformer fault voiceprint diagnosis sample enhancement method according to claim 2, characterized in that: The fault models include a winding deformation fault model, a core loosening fault model and a winding displacement fault model; The winding deformation fault model is expressed as: in, represents the deformation sensitivity factor, Indicates the relative deformation of the winding radius; The core loosening fault model is expressed as: in, Indicates the magnetostriction after loosening, represents the core loose coupling coefficient, Indicates the gap between the laminations, Indicates the thickness of silicon steel sheet; The winding displacement fault model is expressed as: in, represents the cross-sectional area of ​​the core, represents the elastic modulus of silicon steel, represents the strain-stress conversion coefficient, Indicates the core mass, represents the stiffness coefficient, is the core vibration displacement.

4. The transformer fault voiceprint diagnosis sample enhancement method according to claim 1, characterized in that: The acoustic wave equation is expressed as: in, Represents the winding, Indicates insulating oil, Indicates the mechanical housing, Indicates the The sound pressure in the medium, Indicates the The speed of sound in a medium, Indicates the The viscosity coefficient of the medium, Indicates the The density of the medium, represents the Laplace operator, represents the derivative; The medium interface boundary conditions include a solid-liquid interface sound pressure continuity condition, a normal particle velocity continuity condition, and a liquid-solid interface boundary condition that is an acoustic impedance constraint; The solid-liquid interface sound pressure continuity condition is expressed as: in, represents the sound pressure of the incident sound wave, represents the sound pressure of the reflected sound wave, Indicates the sound pressure of the transmitted sound wave; The normal particle velocity continuity condition is expressed as: in, represents the total sound pressure of the first medium, represents the total sound pressure of the second medium, represents the normal direction of the first medium pointing to the first medium; The liquid-solid interface boundary condition is an acoustic impedance constraint expressed as: in, represents the angular frequency of the sound wave, Represents the acoustic impedance of the outer surface of the mechanical housing, Indicates the total sound pressure of the third medium.

5. The transformer fault voiceprint diagnosis sample enhancement method according to claim 1, characterized in that: Determining a physical simulation data set according to the fault simulation sound field includes: The acoustic field reconstruction model is established based on the equivalent source method, and the surface of the transformer tank is discretized into equivalent sound source areas. Acquiring a measurement point signal in the fault simulation sound field, and calculating a correlation coefficient between the equivalent sound source signal and the measurement point signal; A target acquisition point in the measurement point signal is determined according to the correlation coefficient, and data of the target acquisition point in the fault simulation sound field is acquired to obtain the physical simulation data set.

6. The transformer fault voiceprint diagnosis sample enhancement method according to claim 5, characterized in that: The formula used to calculate the correlation coefficient between the equivalent sound source signal and the measurement point signal includes: in, represents the anti-zero correction term, Indicates the frequency amplitude of the equivalent sound source signal, Indicates the frequency amplitude of the measurement point signal, Indicates the n The equivalent sound source signal and the m The correlation coefficient between the signals of the measurement points is Indicates the m The signal of each measurement point at all frequencies f The mean on Indicates the n Equivalent sound source signals at all frequency points f The mean on Indicates the total number of frequency points.

7. The transformer fault voiceprint diagnosis sample enhancement method according to claim 1, characterized in that: The physical simulation data set is sparsely represented to obtain a sparse dictionary and a sparse coefficient matrix, and transfer learning is performed on real fault samples to obtain a transfer dictionary, including: performing DC removal and bandpass filtering on the physical simulation data set, and performing wavelet packet decomposition on the physical simulation data set after the bandpass filtering to obtain sub-band coefficients; splicing the sub-band coefficients to obtain an initial dictionary, and iteratively optimizing the initial dictionary to obtain the sparse dictionary and the sparse coefficient matrix; Performing sparse reconstruction on the real fault sample according to the sparse dictionary to obtain sample coefficients, and calculating reconstruction residuals according to the sample coefficients; Performing credible labeling on the real fault sample according to the reconstructed residual to obtain a credible labeled sample, and determining a sparse coefficient set according to the credible labeled sample; The sparse dictionary is used as an initial value of the dictionary, and the initial value of the dictionary is iteratively updated according to the credible labeled samples and the sparse coefficient set to obtain the migration dictionary.

8. The transformer fault voiceprint diagnosis sample enhancement method according to claim 7, characterized in that: Performing sample enhancement according to the sparse dictionary, the sparse coefficient matrix, and the migration dictionary to obtain a transformer fault voiceprint enhancement sample includes: Generating candidate enhancement samples according to the sparse coefficient matrix and the migration dictionary, and calculating the residual of the candidate enhancement samples under the sparse dictionary to obtain an enhancement residual; The candidate enhanced samples are screened according to the enhanced residual to obtain a screened sample, and the screened sample is mixed with the credible labeled sample to obtain the transformer fault voiceprint enhanced sample.

9. A transformer fault voiceprint diagnosis sample enhancement system, characterized in that: The system comprises: A model building module is used to establish a transformer core winding vibration mechanism model and perform electromagnetic force correction on the transformer core winding vibration mechanism model to obtain a transformer core winding vibration model; A fault simulation module is used to perform fault simulation on the transformer core winding vibration model to obtain a fault model and construct an acoustic wave equation and a medium interface boundary condition for the vibration of the core and winding in the medium propagation; a fault simulation module, configured to perform an acoustic field simulation on the fault model according to the acoustic wave equation and the medium interface boundary condition to obtain a fault simulation sound field, and determine a physical simulation data set according to the fault simulation sound field; A sparse migration module is used to perform sparse representation on the physical simulation data set to obtain a sparse dictionary and a sparse coefficient matrix, and to perform migration learning on real fault samples to obtain a migration dictionary; The sample enhancement module is used to perform sample enhancement according to the sparse dictionary, the sparse coefficient matrix and the migration dictionary to obtain a transformer fault voiceprint enhancement sample.

10. A computer-readable storage medium storing a computer program, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images