Substation primary equipment partial discharge multi-source sensing fusion positioning method and system

Abstract

This application relates to a multi-source sensor fusion localization method and system for partial discharge in primary equipment of a substation. The method includes: constructing a digital twin model based on the equipment's three-dimensional geometric parameters, electromagnetic properties of the internal medium, and acoustic properties; acquiring ultra-high frequency electromagnetic wave signals and acoustic emission signals and performing multipath decomposition to obtain electromagnetic wave multipath delay cluster feature vectors and acoustic wave multipath delay cluster feature vectors; calculating path mismatch degrees for all hypothetical source point locations based on the feature vectors to obtain a path dependency matrix; calculating a wave velocity collaborative correction factor for each hypothetical source point location based on this matrix, and performing joint likelihood estimation by combining electromagnetic wave and acoustic wave path mismatch degrees to obtain a preliminary spatial likelihood distribution; and performing variational Bayesian inference on this distribution to obtain the localization result and confidence interval evaluation index. This method can effectively eliminate false localization points and significantly improve the accuracy and reliability of partial discharge localization in complex structural equipment.

Description

Technical Field

[0001] This invention belongs to the field of substation discharge location technology, and in particular relates to a multi-source sensor fusion location method and system for partial discharge of primary equipment in substations. Background Technology

[0002] With the continuous improvement of power system voltage levels and the ongoing expansion of power grid scale, the safe and stable operation of primary equipment in substations is of paramount importance to ensuring the reliability of the entire power system. Partial discharge, as a major symptom and manifestation of equipment insulation degradation, has always been a research hotspot in the field of condition-based maintenance. Utilizing the ultra-high frequency electromagnetic waves and acoustic waves simultaneously excited during partial discharge for joint location is a mainstream technology widely used in online monitoring of primary equipment such as transformers and GIS systems. Electromagnetic waves propagate at approximately the speed of light inside the equipment, while acoustic waves propagate at only a few kilometers per second in insulating media. This significant speed difference provides a natural physical basis for location methods based on time difference of arrival.

[0003] Traditional partial discharge localization methods typically employ a geometric localization approach based on time delay estimation. This method first uses multiple sensors deployed on the device's casing to simultaneously acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals. Then, it constructs a hyperbolic equation system using the time difference of arrival of the electromagnetic waves, or a spherical equation system using the time difference of arrival of the electromagnetic and acoustic waves at the same sensor. Finally, it uses numerical optimization algorithms such as the least squares method or Newton's iteration method to determine the spatial coordinates of the discharge source. Its core assumption is that electromagnetic waves propagate in a straight line with a constant speed inside the device, and that acoustic waves propagate at a uniform speed in a single medium, thus directly converting the time difference into a geometric distance difference for localization calculation.

[0004] However, in the complex structures of actual substations, the traditional positioning methods based on the assumptions of straight-line and uniform propagation face serious applicability challenges. The complex core and winding structures inside transformers, as well as components such as metal elbows and basin-type insulators within the GIS cavity, inevitably cause reflection, diffraction, and scattering of ultra-high frequency electromagnetic waves during propagation, meaning the sensors often do not receive the direct wave. Simultaneously, sound waves undergo refraction and mode conversion at interfaces with different acoustic impedance media, such as transformer oil, insulating paper, and SF6 gas, resulting in curved propagation paths and non-constant speeds. Because the propagation model used in traditional methods has a structural mismatch with the actual physical process of wave propagation, there is a complex nonlinear deviation between the geometric distance calculated based on the incorrect model and the physical distance derived from the actual time delay. Furthermore, the path distortion patterns of electromagnetic waves and sound waves are drastically different, making iterative algorithms prone to converging to erroneous pseudo-location points, severely impacting the confidence level of the positioning results and their engineering application value. Summary of the Invention

[0005] Therefore, it is necessary to provide a multi-source sensor fusion positioning method and system for partial discharge of primary equipment in substations, which can correct and eliminate false positioning points based on the propagation path and improve positioning accuracy, in order to address the above-mentioned technical problems.

[0006] In a first aspect, this application provides a multi-source sensor fusion localization method for partial discharge of primary equipment in a substation, including:

[0007] S1. Based on the three-dimensional geometric parameters, electromagnetic properties, and acoustic properties of the internal medium of the primary equipment in the substation, a digital twin model is constructed. The digital twin model is used to simultaneously simulate the propagation behavior of ultra-high frequency electromagnetic waves and sound waves inside the equipment, following Huygens' principle and Snell's law.

[0008] S2. Acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals, and perform multipath decomposition on the ultra-high frequency electromagnetic wave signals and acoustic wave signals respectively to obtain the electromagnetic wave multipath delay cluster feature vector and the acoustic wave multipath delay cluster feature vector; the ultra-high frequency electromagnetic wave signals and acoustic emission signals are synchronously acquired by a multi-source sensor array deployed on the equipment.

[0009] S3. Based on the eigenvectors of electromagnetic wave multipath delay clusters and acoustic wave multipath delay clusters, the path mismatch degree of all assumed source point positions in the computational domain of the digital twin model is calculated to obtain the path dependency matrix.

[0010] S4. Based on the path dependence matrix, the wave velocity cooperative correction factor is calculated for each hypothetical source location. Then, joint likelihood estimation is performed based on the wave velocity cooperative correction factor corresponding to each hypothetical source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix to obtain the preliminary spatial likelihood distribution results of the discharge source.

[0011] S5. Perform variational Bayesian inference on the preliminary spatial likelihood distribution results to obtain the local discharge source location results and the confidence interval evaluation index of the location results.

[0012] In one embodiment, S1 includes:

[0013] S11. Based on the three-dimensional geometric structure parameters, construct a three-dimensional geometric model of the primary equipment of the substation;

[0014] S12. Based on electromagnetic and acoustic characteristic parameters, different medium regions in the three-dimensional geometric model are assigned corresponding electromagnetic and acoustic property materials to obtain a geometric model with material assignments.

[0015] S13. Perform finite element mesh generation on the geometric model of the material assignment to generate a computational mesh model that simultaneously includes electromagnetic field solution nodes and acoustic field solution nodes, thus obtaining a digital twin model.

[0016] In one embodiment, S2 includes:

[0017] S21. Multipath component decomposition of ultra-high frequency electromagnetic wave signals is performed using the matched pursuit algorithm to extract the arrival time of each reflected wave and generate electromagnetic wave multipath delay cluster feature vectors.

[0018] S22. Perform mode separation of longitudinal and transverse waves on the acoustic emission signal using wavelet packet transform to obtain longitudinal wave waveform data and transverse wave waveform data;

[0019] S23. Perform multipath component decomposition on the longitudinal wave waveform data and the transverse wave waveform data respectively using the matching pursuit algorithm, extract the arrival time of each refracted wave, generate the acoustic longitudinal wave multipath delay cluster feature vector and the acoustic transverse wave multipath delay cluster feature vector, and obtain the acoustic multipath delay cluster feature vector.

[0020] In one embodiment, S3 includes:

[0021] S31. Discretize the computational domain of the digital twin model in three dimensions according to a preset step size to generate a set of hypothetical source point meshes consisting of all hypothetical source point locations.

[0022] S32. Based on the digital twin model, the electromagnetic wave propagation path is simulated in the forward direction by using the ray tracing algorithm for each hypothetical source point location in the hypothetical source point grid set, so as to obtain the electromagnetic wave simulation multipath delay cluster corresponding to each hypothetical source point location.

[0023] S33. Based on the digital twin model, the acoustic wave propagation path is simulated in the forward direction by solving the equation of the process function for each hypothetical source point location in the hypothetical source point grid set, so as to obtain the acoustic wave longitudinal wave simulation multipath delay cluster and acoustic wave transverse wave simulation multipath delay cluster corresponding to each hypothetical source point location.

[0024] S34. Perform path-by-path comparison calculations between the electromagnetic wave simulation multipath delay cluster corresponding to each assumed source point location and the electromagnetic wave multipath delay cluster feature vector to obtain the electromagnetic wave path mismatch degree for each assumed source point location.

[0025] S35. Perform path-by-path comparison calculations between the acoustic longitudinal wave simulation multipath delay cluster and the acoustic longitudinal wave multipath delay cluster feature vector corresponding to each assumed source point location to obtain the acoustic longitudinal wave path mismatch degree for each assumed source point location.

[0026] S36. Perform path-by-path comparison calculations between the acoustic transverse wave simulation multipath delay cluster and the acoustic transverse wave multipath delay cluster feature vector corresponding to each assumed source point location to obtain the acoustic transverse wave path mismatch degree for each assumed source point location.

[0027] S37. Combine the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch at each hypothetical source location into a multidimensional mismatch vector, and then arrange the multidimensional mismatch vectors of all hypothetical source locations into a matrix within the computational domain to obtain the path dependence matrix.

[0028] In one embodiment, S4 calculates the wave velocity cooperative correction factor for each assumed source location, including:

[0029] S41. Based on the path dependence matrix, extract the electromagnetic wave path mismatch degree of each hypothetical source point location, and calculate the equivalent path velocity of the electromagnetic wave corresponding to each hypothetical source point location according to the ratio of the straight distance from the hypothetical source point to each sensor to the electromagnetic wave simulation propagation time.

[0030] S42. Based on the path dependence matrix, extract the path mismatch of the longitudinal wave of the acoustic wave at each assumed source point location, and calculate the equivalent path velocity of the longitudinal wave of the acoustic wave corresponding to each assumed source point location based on the ratio of the straight distance from the assumed source point to each sensor to the simulated propagation time of the longitudinal wave.

[0031] S43. Based on the path dependence matrix, extract the acoustic transverse wave path mismatch degree of each hypothetical source point location, and calculate the equivalent path velocity of the acoustic transverse wave corresponding to each hypothetical source point location according to the ratio of the straight distance from the hypothetical source point to each sensor to the simulated propagation time of the acoustic transverse wave.

[0032] S44. Based on the preset nominal velocities of electromagnetic waves, longitudinal waves, and transverse waves, calculate the degree of variation of the equivalent path velocities of electromagnetic waves, longitudinal waves, and transverse waves relative to their corresponding nominal velocities, and obtain three coefficients of variation.

[0033] S45. Based on the three coefficients of variation, a weighted summation is performed to obtain the wave velocity co-correction factor corresponding to each assumed source point location.

[0034] In one embodiment, S4 performs joint likelihood estimation based on the wave velocity cooperative correction factor corresponding to each assumed source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix, including:

[0035] S46. Based on the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch corresponding to each assumed source point location, the multiphysics path mismatch likelihood value of the corresponding assumed source point location is calculated using the Gaussian likelihood function.

[0036] S47. Based on the wave velocity collaborative correction factor corresponding to each assumed source point location, the wave velocity collaborative correction penalty value for the corresponding assumed source point location is calculated using the exponential penalty function.

[0037] S48. Multiply the multiphysics path mismatch likelihood value of each hypothetical source point location by the wave velocity co-correction penalty value to obtain the joint likelihood function value of that hypothetical source point location.

[0038] S49. Normalize the joint likelihood function values ​​of all hypothetical source locations within the computational domain to generate the preliminary spatial likelihood distribution results of the discharge source.

[0039] In one embodiment, S5 includes:

[0040] S51. Initialize the Gaussian mixture model parameters based on the preliminary spatial likelihood distribution results to obtain the initialized variational distribution parameters;

[0041] S52. Based on the preliminary spatial likelihood distribution results, the initialized variational distribution parameters are iteratively optimized by minimizing the KL divergence between the variational distribution and the true posterior distribution to obtain the optimized variational distribution parameters.

[0042] S53. Determine whether the optimized variational distribution parameters meet the preset iterative convergence conditions. If not, repeat step S52. If they meet, use the current optimized variational distribution parameters as the converged variational distribution parameters.

[0043] S54. Extract the mean vector from the converged variational distribution parameters as the local discharge source location result;

[0044] S55. Extract the covariance matrix from the converged variational distribution parameters and use it as the confidence interval evaluation index for the positioning results.

[0045] Secondly, this application also provides a multi-source sensor fusion positioning system for partial discharge of primary equipment in a substation, comprising:

[0046] The digital twin construction module is used to build a digital twin model based on the three-dimensional geometric parameters, electromagnetic properties, and acoustic properties of the internal medium of the primary equipment in the substation. The digital twin model is used to simultaneously simulate the propagation behavior of ultra-high frequency electromagnetic waves and sound waves inside the equipment, following Huygens' principle and Snell's law.

[0047] The multipath decomposition module is used to acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals, and to perform multipath decomposition on the ultra-high frequency electromagnetic wave signals and acoustic wave signals respectively to obtain the electromagnetic wave multipath delay cluster feature vector and the acoustic wave multipath delay cluster feature vector; the ultra-high frequency electromagnetic wave signals and acoustic emission signals are synchronously acquired by a multi-source sensor array deployed on the equipment.

[0048] The path mismatch calculation module is used to calculate the path mismatch of all assumed source point positions in the computational domain of the digital twin model based on the eigenvectors of electromagnetic wave multipath delay clusters and acoustic wave multipath delay clusters, and obtain the path dependency matrix.

[0049] The joint likelihood estimation module is used to calculate the wave velocity cooperative correction factor for each hypothetical source location based on the path dependence matrix, and to perform joint likelihood estimation based on the wave velocity cooperative correction factor corresponding to each hypothetical source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix, so as to obtain the preliminary spatial likelihood distribution result of the discharge source.

[0050] The variational Bayesian inference module is used to perform variational Bayesian inference on the preliminary spatial likelihood distribution results to obtain the local discharge source location results and the confidence interval evaluation index of the location results.

[0051] Thirdly, this application also provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the multi-source sensor fusion positioning method for partial discharge of primary equipment in a substation as described in the first aspect.

[0052] Fourthly, this application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the multi-source sensor fusion positioning method for partial discharge of primary equipment in a substation as described in the first aspect.

[0053] The aforementioned multi-source sensor fusion localization method and system for partial discharge of primary equipment in substations constructs a digital twin model capable of simultaneously simulating the propagation of ultra-high frequency electromagnetic waves and sound waves within the equipment according to Huygens' principle and Snell's law. It utilizes the multipath delay cluster feature vectors of electromagnetic waves and sound waves extracted from measured signals to calculate the path mismatch degree of all assumed source point locations within the model's computational domain, constructing a path dependence matrix reflecting the propagation distortion characteristics of each location. Based on this matrix, it calculates the wave velocity collaborative correction factor for each assumed source point location, jointly determined by multi-physics path distortion. This correction factor is then combined with the path mismatch degree for joint likelihood estimation to obtain a preliminary spatial likelihood distribution. Finally, variational Bayesian inference outputs the localization result and corresponding confidence interval from this likelihood distribution. This achieves systematic correction of electromagnetic and sound wave propagation path distortion in complex structures, effectively solving the problem of easy convergence to pseudo-localization points due to propagation model mismatch, and significantly improving the accuracy and reliability of the localization results. Attached Figure Description

[0054] To more clearly illustrate the technical solutions in the embodiments or related technologies of this application, the accompanying drawings used in the description of the embodiments or related technologies will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0055] Figure 1 This is a flowchart illustrating a multi-source sensor fusion positioning method for partial discharge of primary equipment in a substation, provided by the present invention.

[0056] Figure 2 This is a schematic diagram of the structure of a multi-source sensor fusion positioning system for partial discharge of primary equipment in a substation, provided by the present invention. Detailed Implementation

[0057] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0058] In one embodiment, such as Figure 1 As shown, a multi-source sensor fusion localization method for partial discharge of primary equipment in a substation is provided. This embodiment illustrates the application of this method to a terminal. It is understood that this method can also be applied to a server, and further to a system including both a terminal and a server, and is implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0059] S1. Based on the three-dimensional geometric parameters, electromagnetic properties, and acoustic properties of the internal medium of the primary equipment in the substation, a digital twin model is constructed. The digital twin model is used to simultaneously simulate the propagation behavior of ultra-high frequency electromagnetic waves and sound waves inside the equipment, following Huygens' principle and Snell's law.

[0060] Specifically, the three-dimensional geometric parameters of the substation's primary equipment are read and imported. These parameters include the spatial coordinates and topological connections of the equipment's outer shell, internal conductive components, insulating support components, and dielectric interfaces. Simultaneously, the electromagnetic and acoustic properties of the internal dielectric are imported. The electromagnetic properties include dielectric constant, conductivity, and permeability, while the acoustic properties include sound velocity, acoustic impedance, and attenuation coefficient. Preferably, spatial meshing is performed according to the actual physical structure and dielectric distribution of the equipment, dividing the internal space of the equipment into continuous and mutually compatible discrete computational units. This constructs a digital twin model that is completely consistent with the actual equipment in terms of geometry and dielectric distribution. Furthermore, the ultra-high frequency electromagnetic wave propagation control equation and the acoustic wave propagation control equation are embedded in the digital twin model, enabling the model to simultaneously follow Huygens' principle and Snell's law to simulate the reflection, diffraction, scattering, and refraction behavior of ultra-high frequency electromagnetic waves within the dielectric and at interfaces, as well as the refraction, mode conversion, and attenuation behavior of acoustic waves propagating in different dielectrics.

[0061] S2. Acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals, and perform multipath decomposition on the ultra-high frequency electromagnetic wave signals and acoustic wave signals respectively to obtain the multipath delay cluster feature vectors of electromagnetic wave and acoustic wave. The ultra-high frequency electromagnetic wave signals and acoustic emission signals are synchronously acquired by a multi-source sensor array deployed on the equipment.

[0062] Specifically, the system receives ultra-high frequency electromagnetic wave signals and acoustic emission signals synchronously acquired by a multi-source sensor array deployed at preset locations on the device casing. The multi-source sensor array includes ultra-high frequency sensing units and acoustic emission sensing units. Each sensing unit uses a unified clock reference to achieve synchronous signal acquisition, ensuring the consistency of the ultra-high frequency electromagnetic wave signals and acoustic emission signals in the time dimension. For example, the ultra-high frequency electromagnetic wave signals undergo joint time-domain and frequency-domain analysis. Through peak detection and time delay extraction, signal components corresponding to different propagation paths are separated, forming an electromagnetic wave multipath delay cluster feature vector. Furthermore, the same time delay extraction and component separation method is applied to the acoustic emission signals to eliminate the influence of environmental noise and interference signals, retaining the effective signal components excited by partial discharge and propagating to the sensing units through different paths, obtaining an acoustic wave multipath delay cluster feature vector. This feature vector can completely characterize the multipath propagation time delay distribution characteristics of both the ultra-high frequency electromagnetic wave and the acoustic wave.

[0063] S3. Based on the eigenvectors of electromagnetic wave multipath delay clusters and acoustic wave multipath delay clusters, the path mismatch degree of all assumed source point positions in the computational domain of the digital twin model is calculated to obtain the path dependency matrix.

[0064] Specifically, within the computational domain of the digital twin model, all hypothetical source locations are traversed according to a preset spatial step size, with each hypothetical source location simulating a potential spatial location of a partial discharge source. Further, for each hypothetical source location, the theoretical propagation paths and theoretical propagation delays of UHF electromagnetic waves and sound waves from that hypothetical source location to each sensing unit are simulated and calculated. The theoretical propagation delays are compared component-by-component with the actual measured delays in the feature vectors of the electromagnetic wave multipath delay clusters and the sound wave multipath delay clusters to calculate the delay deviations of the corresponding paths. Preferably, the electromagnetic wave path delay deviations and sound wave path delay deviations corresponding to all hypothetical source locations are uniformly regularized and stored in a matrix to form a path dependency matrix. This matrix completely records the degree of matching between the propagation paths of UHF electromagnetic waves and sound waves and the actual physical propagation process under different hypothetical source locations.

[0065] S4. Based on the path dependence matrix, the wave velocity cooperative correction factor is calculated for each hypothetical source location. Then, a joint likelihood estimation is performed based on the wave velocity cooperative correction factor corresponding to each hypothetical source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix to obtain the preliminary spatial likelihood distribution result of the discharge source.

[0066] Specifically, based on the electromagnetic wave path mismatch and acoustic wave path mismatch corresponding to each hypothetical source location stored in the path dependency matrix, electromagnetic wave propagation velocity correction terms and acoustic wave propagation velocity correction terms are constructed respectively. A wave velocity collaborative correction factor corresponding to each hypothetical source location is obtained through weighted fusion calculation of the two terms. This correction factor can compensate for wave velocity deviations caused by medium distribution and path curvature. Further, the wave velocity collaborative correction factor, along with the electromagnetic wave path mismatch and acoustic wave path mismatch in the path dependency matrix, are substituted into a preset joint likelihood function to complete the numerical calculation of the likelihood probability, obtaining the probability value that each hypothetical source location is the actual discharge source location. For example, the probability values ​​of all hypothetical source locations are spatially mapped and interpolated within the computational domain to form a continuous preliminary spatial likelihood distribution result. This result reflects the probability distribution of the discharge source at different spatial locations within the device.

[0067] S5. Perform variational Bayesian inference on the preliminary spatial likelihood distribution results to obtain the local discharge source location results and the confidence interval evaluation index of the location results.

[0068] Specifically, the preliminary spatial likelihood distribution results are used as the prior probability distribution input into the variational Bayesian inference process. The approximate posterior probability distribution between the latent variables and the observed variables is solved through iterative optimization, reducing the interference of spurious peaks and outlier likelihood values ​​on positioning accuracy. Further, the spatial coordinates corresponding to the probability peaks are extracted from the optimized posterior probability distribution; these spatial coordinates represent the positioning result of the local discharge source. Preferably, based on the dispersion and peak stability of the posterior probability distribution, a confidence interval evaluation index for the positioning result is calculated. This index can quantitatively characterize the reliability of the positioning result, providing a quantitative reference for the engineering application of the positioning result.

[0069] In the aforementioned multi-source sensor fusion localization method for partial discharge of primary equipment in substations, a digital twin model that fits the actual physical characteristics of the equipment is constructed to complete the multipath decomposition of ultra-high frequency electromagnetic wave signals and acoustic emission signals. The path mismatch degree of the assumed source point position in the computational domain is calculated to obtain the path dependence matrix. Combined with the wave velocity cooperative correction factor, joint likelihood estimation is completed. Then, variational Bayesian inference is used to obtain accurate localization results and confidence interval evaluation indicators. This method effectively corrects the mismatch between the wave propagation model and the actual physical process in traditional localization methods, significantly reduces the localization deviation caused by path distortion, and improves the accuracy and confidence of partial discharge localization of primary equipment in substations.

[0070] In one embodiment, S1 includes:

[0071] S11. Based on the three-dimensional geometric structure parameters, construct a three-dimensional geometric model of the primary equipment of the substation.

[0072] Specifically, the three-dimensional geometric parameters of the substation's primary equipment are read. These parameters comprehensively include the spatial contour data of all components, such as the equipment casing, internal core, windings, conductive busbars, insulating supports, and basin insulators, as well as the relative positional relationships and topological connections between these components, ensuring that the parameters can completely replicate the actual physical structure of the equipment. Preferably, a three-dimensional modeling algorithm is used to analyze and reconstruct the read three-dimensional geometric parameters. Each component is then modeled in three dimensions according to its actual geometric form. All individual components are then spatially assembled according to their topological connections. During assembly, the connection interfaces and relative positions of each component are strictly matched to avoid geometric interference and positional deviations. Furthermore, the assembled three-dimensional model undergoes geometric verification to eliminate redundant geometric surfaces, overhanging edges, and other defects generated during the modeling process, ensuring that the three-dimensional geometric model accurately reproduces the actual geometric structure of the substation's primary equipment.

[0073] S12. Based on electromagnetic and acoustic characteristic parameters, different medium regions in the three-dimensional geometric model are assigned corresponding electromagnetic and acoustic property materials to obtain a geometric model with material assignment.

[0074] Specifically, the electromagnetic and acoustic properties of the internal medium of the substation's primary equipment are read. The electromagnetic properties include the dielectric constant, conductivity, and permeability of each medium, while the acoustic properties include the sound velocity, acoustic impedance, and attenuation coefficient. All parameters correspond one-to-one with the actual medium type used in the equipment. For example, the three-dimensional geometric model is divided into regions based on the actual medium distribution. Each region corresponds to a specific filling medium, and the spatial extent and boundary coordinates of each region are clearly defined. Further, the electromagnetic and acoustic properties are matched to the corresponding medium regions according to the medium type. Each medium region is assigned corresponding electromagnetic and acoustic properties materials. During the assignment process, it is ensured that the parameters accurately correspond to the medium region; that is, the electromagnetic and acoustic properties materials for the same medium region match the characteristics of the actual filling medium in that region. After the assignment is completed, a geometric model with assigned materials is obtained, giving the model the physical basis for simulating the propagation of electromagnetic and sound waves.

[0075] S13. Perform finite element mesh generation on the geometric model of the material assignment to generate a computational mesh model that simultaneously includes electromagnetic field solution nodes and acoustic field solution nodes, thus obtaining a digital twin model.

[0076] Specifically, a finite element method (FEM) mesh generation algorithm is used to mesh the geometric model assigned to the material. During the meshing process, differentiated meshing strategies are adopted for different regions, taking into account the geometric complexity and solution accuracy requirements of the model. For regions with complex geometry and drastic changes in medium properties, a denser meshing method is used, while for regions with regular geometry and uniform medium properties, a sparse meshing method is used, balancing solution accuracy and computational efficiency. Furthermore, during the meshing process, electromagnetic field and acoustic field solution nodes are generated simultaneously. Each mesh cell corresponds to a unique electromagnetic field and acoustic field solution node, and the nodes store the geometric coordinates, electromagnetic property parameters, and acoustic property parameters of the corresponding mesh cell, ensuring that electromagnetic wave and acoustic wave propagation simulations can be carried out simultaneously. After meshing, the generated computational mesh model is checked for mesh quality, and unqualified meshes such as distorted meshes and mesh warping are removed to ensure the regularity and continuity of the mesh cells. After successful verification, a computational grid model containing both electromagnetic field and acoustic field solution nodes is obtained. This model is a digital twin model capable of simulating the propagation behavior of ultra-high frequency electromagnetic waves and acoustic waves.

[0077] In the above embodiment, a precise three-dimensional geometric model is first constructed using the three-dimensional geometric structural parameters of the substation's primary equipment. Then, corresponding material properties are assigned to each medium region of the model by combining electromagnetic and acoustic characteristic parameters. Finally, a computational mesh model containing dual solver nodes is generated through finite element mesh generation, gradually completing the construction of the digital twin model. This embodiment closely matches the actual physical structure and medium characteristics of the equipment, solving the problem of mismatch between traditional models and actual equipment. The constructed digital twin model can accurately simulate the actual propagation behavior of electromagnetic waves and sound waves, providing a realistic and reliable simulation foundation for multi-source sensor fusion positioning of partial discharge, effectively improving the accuracy and confidence of subsequent positioning results.

[0078] In one embodiment, S2 includes:

[0079] S21. Multipath component decomposition of ultra-high frequency electromagnetic wave signals is performed using the matched pursuit algorithm to extract the arrival time of each reflected wave and generate electromagnetic wave multipath delay cluster feature vectors.

[0080] Specifically, the ultra-high frequency electromagnetic wave signal acquired by the multi-source sensor array is read and converted into a time-domain data format suitable for the Matching Pursuit Algorithm (MPA). The MPA iteratively selects the atom that best matches the signal residual from an overcomplete atom library, gradually approximating the original signal through linear combination. Preferably, a Gaussian wavelet atom library suitable for ultra-high frequency electromagnetic wave signals is constructed. The time-frequency characteristics of the atoms in this atom library match the waveform characteristics of the ultra-high frequency electromagnetic wave signal excited by partial discharge, enabling accurate characterization of signal components along different propagation paths. Further, using the ultra-high frequency electromagnetic wave signal as the initial residual, the correlation between the initial residual and all atoms in the atom library is calculated through inner product operations. The atom with the highest correlation is selected as the best matching atom, and the time-domain position corresponding to this atom is determined as the arrival time of a first reflected wave. The updated residual is then obtained by subtracting the linear projection of the best matching atom from the initial residual. The iterative process of correlation calculation, best matching atom selection, and residual update is repeated until the energy of the residual is lower than a preset convergence threshold. At this point, the iteration stops, and all extracted arrival times are summarized. For example, all arrival times are sorted in chronological order and vectorized by combining them with the corresponding sensor unit identifiers to form an electromagnetic wave multipath delay cluster feature vector that can fully characterize the multipath propagation characteristics of ultra-high frequency electromagnetic waves.

[0081] S22. Perform mode separation of longitudinal and transverse waves on the acoustic emission signal using wavelet packet transform to obtain longitudinal wave waveform data and transverse wave waveform data.

[0082] Specifically, the acoustic emission signal acquired by the multi-source sensor array is read and converted into a continuous time-domain sequence that can be processed by wavelet packet transform. Wavelet packet transform decomposes the signal into wavelet packet subspaces of different scales and frequencies, enabling fine frequency domain division of the signal. Preferably, a wavelet basis function suitable for mode separation of acoustic emission signals is selected. This wavelet basis function has good time-frequency localization characteristics and can effectively distinguish the frequency differences and propagation characteristic differences between longitudinal and transverse waves. Further, wavelet packet decomposition is performed on the acoustic emission signal, and the decomposed wavelet packet coefficients are classified according to frequency intervals. The frequency interval coefficients corresponding to the propagation characteristics of longitudinal waves are classified as longitudinal wave wavelet packet coefficients, and the frequency interval coefficients corresponding to the propagation characteristics of transverse waves are classified as transverse wave wavelet packet coefficients. For example, inverse wavelet packet transform is performed on the longitudinal wave wavelet packet coefficients and the transverse wave wavelet packet coefficients respectively, restoring the frequency domain coefficient data to time domain waveform data, and finally obtaining independent longitudinal wave waveform data and transverse wave waveform data, realizing mode separation of longitudinal and transverse waves in the acoustic emission signal.

[0083] S23. Perform multipath component decomposition on the longitudinal wave waveform data and the transverse wave waveform data respectively using the matching pursuit algorithm, extract the arrival time of each refracted wave, generate the acoustic longitudinal wave multipath delay cluster feature vector and the acoustic transverse wave multipath delay cluster feature vector, and obtain the acoustic multipath delay cluster feature vector.

[0084] Specifically, P-wave and S-wave waveform data are read separately. Using the same Gaussian wavelet atom library and convergence logic as in S21, multipath component decomposition using the matching pursuit algorithm is performed on the P-wave waveform data. Further, during the iteration process, the atom that best matches the residual of the P-wave waveform data is selected, and the arrival time corresponding to each refracted wave is extracted. These arrival times are then encoded in chronological order to generate a P-wave multipath delay cluster feature vector. For example, the same matching pursuit algorithm multipath component decomposition process is performed on the S-wave waveform data to extract the arrival time of each refracted wave and encode it to generate a S-wave multipath delay cluster feature vector. Preferably, the P-wave and S-wave multipath delay cluster feature vectors are dimensionally aligned and jointly concatenated. During the concatenation process, their respective modal identifiers and sensor unit identifiers are preserved, ultimately forming a multipath delay cluster feature vector containing information on the propagation of both P-wave and S-wave multipaths.

[0085] In the above embodiment, the multipath components of the UHF electromagnetic wave signal are first decomposed using a matching pursuit algorithm to generate corresponding time delay cluster feature vectors. Then, wavelet packet transform is used to separate the P-wave and S-wave modes in the acoustic emission signal. Finally, the separated P-wave and S-wave waveform data are decomposed using the matching pursuit algorithm and jointly generated to generate acoustic multipath time delay cluster feature vectors, thus completely extracting the multipath propagation time delay features of UHF electromagnetic waves and acoustic waves. This embodiment accurately separates signal components with different propagation paths and different waveforms, solving the problem that traditional methods cannot effectively distinguish multipath signals and different acoustic modes, and effectively improving the matching degree between signal characteristics and actual propagation behavior during partial discharge localization.

[0086] In one embodiment, S3 includes:

[0087] S31. Discretize the computational domain of the digital twin model in three dimensions according to a preset step size to generate a set of hypothetical source point grids consisting of all hypothetical source point locations.

[0088] Specifically, the spatial computational domain corresponding to the digital twin model is read, and this domain completely covers all areas within the primary equipment of the substation that may have localized discharge sources. Preferably, the spatial computational domain is discretized sequentially in three dimensions at equal intervals according to a preset step size, maintaining a consistent step size throughout the entire computational domain to ensure uniformity during traversal and calculation. Further, the spatial coordinate points formed by each discretization are marked as independent hypothetical source point locations. All hypothetical source point locations are uniformly stored and indexed according to their spatial arrangement order, forming a hypothetical source point grid set composed of all hypothetical source point locations. This set can completely cover the potential discharge source areas within the equipment.

[0089] S32. Based on the digital twin model, the electromagnetic wave propagation path is simulated in the forward direction by using the ray tracing algorithm for each hypothetical source point location in the hypothetical source point grid set, so as to obtain the electromagnetic wave simulation multipath delay cluster corresponding to each hypothetical source point location.

[0090] Specifically, each hypothetical source point location in the hypothetical source point grid set is sequentially read and input into the digital twin model as the equivalent emission point of the UHF electromagnetic wave. Further, a ray tracing algorithm is used to simulate the propagation process of the UHF electromagnetic wave from the hypothetical source point location. The ray tracing algorithm follows Huygens' principle and Snell's law, and can simulate the reflection, diffraction, and refraction propagation paths of electromagnetic waves at interfaces between different media. For example, during the simulation, the propagation path and corresponding propagation delay of the electromagnetic wave from the hypothetical source point location to each sensing location are recorded segment by segment. The propagation delays corresponding to all reachable paths are extracted and arranged in chronological order to form a multipath delay cluster for each hypothetical source point location. This delay cluster fully reflects the theoretical multipath propagation delay distribution of the UHF electromagnetic wave within the digital twin model.

[0091] S33. Based on the digital twin model, the acoustic wave propagation path is simulated in the forward direction for each hypothetical source point location in the hypothetical source point grid set by solving the equation of the process function, so as to obtain the acoustic wave longitudinal wave simulation multipath delay cluster and acoustic wave transverse wave simulation multipath delay cluster corresponding to each hypothetical source point location.

[0092] Specifically, keeping the assumed source location unchanged, and based on the medium distribution information with material assignments already completed in the digital twin model, a functional equation solving algorithm is used to perform a forward simulation of the sound wave propagation process. The functional equation describes the propagation law of the wavefront in a non-uniform medium, accurately reflecting the refraction and mode conversion characteristics of the sound wave at the interface of media with different acoustic impedances. Furthermore, during the simulation, the independent propagation paths of the longitudinal and transverse waves are traced separately, and the multipath propagation delays of the longitudinal and transverse waves from the same assumed source location to each sensing location are recorded. These are then compiled into simulated multipath delay clusters for the longitudinal and transverse waves corresponding to each assumed source location.

[0093] S34. Perform path-by-path comparison calculations between the electromagnetic wave simulation multipath delay cluster corresponding to each assumed source point location and the characteristic vector of the electromagnetic wave multipath delay cluster to obtain the electromagnetic wave path mismatch degree for each assumed source point location.

[0094] Specifically, the simulated multipath time delay clusters of electromagnetic waves corresponding to the same hypothetical source location are compared path by path with the previously extracted feature vectors of the electromagnetic wave multipath time delay clusters, according to the same path number. Preferably, the difference between the simulated time delay and the measured time delay corresponding to each path is calculated to obtain the single-path time delay deviation. Then, the time delay deviations of all paths are normalized and energy-weighted summed to obtain a quantitative value that can comprehensively characterize the degree of deviation between the theoretical propagation path and the measured propagation path. Further, this quantitative value is defined as the electromagnetic wave path mismatch degree at the current hypothetical source location. The smaller the mismatch degree value, the higher the degree of agreement between the simulated electromagnetic wave propagation process and the actual physical propagation process; conversely, the larger the mismatch degree value, the greater the deviation between the two.

[0095] S35. Perform path-by-path comparison calculations between the acoustic longitudinal wave simulation multipath delay cluster and the feature vector of the acoustic longitudinal wave multipath delay cluster corresponding to each assumed source point location to obtain the acoustic longitudinal wave path mismatch degree for each assumed source point location.

[0096] Specifically, the acoustic P-wave simulated multipath delay clusters and their eigenvectors corresponding to the same hypothetical source location are compared path-by-path according to their path correspondence. The comparison calculation method is consistent with that for electromagnetic wave path mismatch: first, the single-path delay deviation is calculated, and then a weighted sum is used to obtain the overall deviation quantization value. Further, this overall deviation quantization value is determined as the acoustic P-wave path mismatch at the current hypothetical source location. This mismatch independently reflects the degree of path conformity between the theoretical simulation environment and the actual propagation environment, providing an independent mismatch basis for P-wave direction in multimodal fusion judgment.

[0097] S36. Perform path-by-path comparison calculations between the acoustic transverse wave simulation multipath delay cluster and the eigenvector of the acoustic transverse wave multipath delay cluster corresponding to each assumed source point location to obtain the acoustic transverse wave path mismatch degree for each assumed source point location.

[0098] Specifically, using the same calculation logic as for the longitudinal wave path mismatch, the characteristic vectors of the simulated transverse wave multipath delay clusters and the transverse wave multipath delay clusters corresponding to the same assumed source location are compared path-by-path delay and weighted by deviation. Furthermore, the calculated overall deviation quantification value is defined as the transverse wave path mismatch at the current assumed source location. This mismatch independently characterizes the degree of deviation between the theoretical and measured propagation paths of the transverse wave. By calculating the longitudinal and transverse wave path mismatches separately, the differences in path distortion generated by different acoustic modes during propagation can be fully distinguished, avoiding evaluation bias caused by a single mode mismatch.

[0099] S37. Combine the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch at each hypothetical source location into a multidimensional mismatch vector, and then arrange the multidimensional mismatch vectors of all hypothetical source locations into a matrix within the computational domain to obtain the path dependence matrix.

[0100] Specifically, for each hypothetical source location, the calculated electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch are combined in a fixed dimensional order to form a multidimensional mismatch vector that simultaneously reflects the path matching degree of the three propagation modes. Further, according to the spatial arrangement order of the hypothetical source grid set within the computational domain of the digital twin model, the multidimensional mismatch vectors corresponding to all hypothetical source locations are sequentially regularized and stored in a matrix, so that each row in the matrix corresponds to a hypothetical source location, and each column corresponds to the path mismatch degree of a propagation mode. The resulting path dependency matrix completely records the matching relationship between all potential discharge source locations and the measured multipath delay characteristics within the computational domain.

[0101] In the above embodiments, a set of hypothetical source point meshes is generated by three-dimensional discretization of the computational domain of the digital twin model. Then, the forward simulation of electromagnetic waves and sound waves is completed using ray tracing algorithm and equation solving algorithm respectively to obtain theoretical multipath time delay clusters. By comparing each path, the path mismatch degrees of electromagnetic waves, longitudinal sound waves, and transverse sound waves are obtained and combined to form a path dependence matrix, thus establishing a complete quantitative correspondence between the hypothetical discharge source location and the measured multipath propagation characteristics. This embodiment fully considers the differences in path distortion of different propagation modes, upgrading the single distance comparison in traditional geometric positioning to a quantitative comparison of mismatch degree of multipath and multimodal propagation, effectively reducing the impact of propagation path distortion on positioning accuracy.

[0102] In one embodiment, S4 calculates the wave velocity cooperative correction factor for each assumed source location, including:

[0103] S41. Based on the path dependence matrix, extract the electromagnetic wave path mismatch degree of each hypothetical source point location, and calculate the equivalent path velocity of the electromagnetic wave corresponding to each hypothetical source point location according to the ratio of the straight distance from the hypothetical source point to each sensor to the electromagnetic wave simulation propagation time.

[0104] Specifically, the electromagnetic wave path mismatch degree corresponding to each hypothetical source point location is read and extracted from the path dependency matrix. This path mismatch degree characterizes the deviation between the simulated propagation path and the measured propagation path of the electromagnetic wave at the current hypothetical source point location. Further, in the spatial coordinate system corresponding to the digital twin model, the straight-line distance from the current hypothetical source point location to each sensor location is calculated, and the simulated propagation time of the electromagnetic wave from the hypothetical source point location to the corresponding sensor location, obtained from the forward simulation of the ray tracing algorithm, is read. Specifically, the ratio of each set of straight-line distances to the corresponding simulated propagation time of the electromagnetic wave is calculated to obtain the electromagnetic wave propagation speed value under a single path. Then, a weighted average is performed on the speed values ​​corresponding to all sensor locations to finally obtain the equivalent path speed of the electromagnetic wave corresponding to the current hypothetical source point location. This speed reflects the equivalent propagation rate of the electromagnetic wave after distortion in the actual propagation path.

[0105] S42. Based on the path dependence matrix, extract the path mismatch of the longitudinal wave of the acoustic wave at each hypothetical source point location, and calculate the equivalent path velocity of the longitudinal wave of the acoustic wave corresponding to each hypothetical source point location according to the ratio of the straight distance from the hypothetical source point to each sensor to the simulated propagation time of the longitudinal wave of the acoustic wave.

[0106] Specifically, the path mismatch degree of the acoustic longitudinal wave corresponding to each hypothetical source location is extracted from the path dependency matrix. This path mismatch degree characterizes the deviation between the simulated and measured propagation paths of the acoustic longitudinal wave at the current hypothetical source location. Further, the straight-line distances from the same hypothetical source location to each sensor location in the digital twin model are used, while the simulated propagation time of the acoustic longitudinal wave from the hypothetical source location to the corresponding sensor location, obtained from the forward simulation of the equation solving algorithm, is read. Specifically, a ratio calculation is performed on each set of straight-line distances and the simulated propagation time of the acoustic longitudinal wave to obtain the acoustic longitudinal wave propagation velocity value under a single path. Then, a weighted average is performed on the velocity values ​​corresponding to all sensor locations to obtain the equivalent path velocity of the acoustic longitudinal wave corresponding to the current hypothetical source location. This velocity reflects the equivalent propagation rate of the acoustic longitudinal wave after refraction and mode conversion at the medium interface.

[0107] S43. Based on the path dependence matrix, extract the acoustic transverse wave path mismatch degree for each hypothetical source point location, and calculate the equivalent path velocity of the acoustic transverse wave corresponding to each hypothetical source point location according to the ratio of the straight distance from the hypothetical source point to each sensor to the simulated propagation time of the acoustic transverse wave.

[0108] Specifically, the path mismatch of the acoustic transverse wave corresponding to each hypothetical source location is extracted from the path dependency matrix. This path mismatch characterizes the deviation between the simulated and measured propagation paths of the acoustic transverse wave at the current hypothetical source location. Further, while keeping the straight-line distance from the same hypothetical source location to each sensor location constant, the simulated propagation time of the acoustic transverse wave from the hypothetical source location to the corresponding sensor location, obtained from the forward simulation of the algorithm solving the equation, is read. Specifically, the ratio of each set of straight-line distances to the simulated propagation time of the acoustic transverse wave is calculated to obtain the acoustic transverse wave propagation velocity value under a single path. Then, a weighted average is performed on the velocity values ​​corresponding to all sensor locations to obtain the equivalent path velocity of the acoustic transverse wave corresponding to the current hypothetical source location. This velocity reflects the equivalent propagation rate of the acoustic transverse wave after the propagation path is bent in a non-uniform medium.

[0109] S44. Based on the preset nominal velocities of electromagnetic waves, longitudinal waves, and transverse waves, calculate the degree of variation of the equivalent path velocities of electromagnetic waves, longitudinal waves, and transverse waves relative to their corresponding nominal velocities, and obtain three coefficients of variation.

[0110] Specifically, the pre-set nominal velocities of electromagnetic waves, longitudinal sound waves, and transverse sound waves are read. These nominal velocities correspond to the theoretical propagation velocities of electromagnetic waves, longitudinal sound waves, and transverse sound waves in a uniform ideal medium, respectively. Preferably, the equivalent path velocity of the electromagnetic wave is calculated by subtracting the nominal velocity of the electromagnetic wave, and the result is normalized to obtain the coefficient of variation of the equivalent path velocity of the electromagnetic wave relative to the nominal velocity. Further, using the same calculation method, the equivalent path velocity of the longitudinal sound wave is processed with the nominal velocity to obtain the coefficient of variation of the longitudinal sound wave, and the equivalent path velocity of the transverse sound wave is processed with the nominal velocity to obtain the coefficient of variation of the transverse sound wave. These three coefficients of variation quantify the degree to which the equivalent path velocity deviates from the ideal propagation velocity under the three propagation modes.

[0111] S45. Based on the three coefficients of variation, a weighted summation is performed to obtain the wave velocity co-correction factor corresponding to each assumed source point location.

[0112] Specifically, based on the influence of the three propagation modes in the positioning process, corresponding weighting coefficients are set, and the values ​​of the weighting coefficients are matched with the sensitivity of each mode's path distortion to the positioning results. Further, the electromagnetic wave variation coefficient, acoustic longitudinal wave variation coefficient, and acoustic transverse wave variation coefficient are multiplied by their respective weighting coefficients, and then the three products are summed to obtain the weighted sum value corresponding to the current assumed source point location. For example, the weighted sum value is subjected to nonlinear mapping processing to map it to a preset numerical range, forming a wave velocity collaborative correction factor that can simultaneously correct the velocity deviations of the three propagation modes. This correction factor can uniformly compensate for wave velocity errors caused by different path distortions.

[0113] In the above embodiments, the equivalent path velocity of electromagnetic waves, the equivalent path velocity of acoustic longitudinal waves, and the equivalent path velocity of acoustic transverse waves corresponding to each assumed source location are calculated sequentially. Then, the coefficient of variation of each equivalent path velocity relative to the nominal velocity is obtained. A wave velocity collaborative correction factor is generated through weighted summation, achieving quantification and collaborative compensation for multimodal propagation velocity deviations. This embodiment addresses the mismatch between the assumption of constant wave velocity and actual wave velocity changes in traditional positioning. By using equivalent velocities and coefficients of variation, it achieves adaptive correction of path distortion, effectively improving the rationality and reliability of joint likelihood estimation and providing a more accurate correction basis for source location.

[0114] In one embodiment, S4 performs joint likelihood estimation based on the wave velocity cooperative correction factor corresponding to each assumed source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix, including:

[0115] S46. Based on the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch corresponding to each assumed source point location, the multiphysics path mismatch likelihood value of the corresponding assumed source point location is calculated using the Gaussian likelihood function.

[0116] Specifically, the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch corresponding to each hypothetical source location are read, and these three types of path mismatch are input into the Gaussian likelihood function as multiphysics path mismatch features. The Gaussian likelihood function follows a normal probability density distribution, which can transform the path mismatch into a probability measure that the corresponding location is a real discharge source. Furthermore, the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch are subjected to exponential operations and normalized weighting, respectively, so that the mismatches of different propagation modes are fused under a unified dimension. For example, the weighted fused result is substituted into the Gaussian likelihood function to complete the numerical solution, obtaining the multiphysics path mismatch likelihood value corresponding to the current hypothetical source location. The larger the likelihood value, the higher the degree of agreement between the various propagation paths at the current hypothetical source location and the measured paths.

[0117] S47. Based on the wave velocity co-correction factor corresponding to each hypothetical source point location, the wave velocity co-correction penalty value for the corresponding hypothetical source point location is calculated using an exponential penalty function.

[0118] Specifically, the wave velocity co-correction factor corresponding to each hypothetical source location is input into a preset exponential penalty function. This exponential penalty function, with its monotonically decreasing characteristic, applies a greater penalty to hypothetical source locations with significant wave velocity variations. Furthermore, the wave velocity co-correction factor is used as the independent variable to perform exponential calculations. During the calculation, preset function parameters control the penalty intensity, ensuring that the calculation result is negatively correlated with the degree of wave velocity deviation. For example, the calculation result is subjected to interval mapping, ensuring that the final output value falls within a preset probability range, forming the wave velocity co-correction penalty value corresponding to the current hypothetical source location. This penalty value effectively suppresses spurious likelihood peaks caused by excessive wave velocity distortion, improving the physical plausibility of the likelihood calculation.

[0119] S48. Multiply the multiphysics path mismatch likelihood value of each hypothetical source location by the wave velocity co-correction penalty value to obtain the joint likelihood function value of that hypothetical source location.

[0120] Specifically, the multiphysics path mismatch likelihood value and wave velocity co-correction penalty value corresponding to the same hypothetical source location are read. These two types of values ​​are then multiplied according to a point-to-point correspondence, achieving a joint fusion of path matching and wave velocity correction characteristics. Furthermore, the multiplication process retains the ability of the multiphysics path mismatch likelihood value to characterize the path fit, while the wave velocity co-correction penalty value eliminates interference from hypothetical source locations with abnormal wave velocity distortion. For example, the result of the multiplication is used as the joint likelihood function value of the current hypothetical source location. This function value integrates both path matching degree and wave velocity rationality constraints, more realistically reflecting the overall confidence level that the hypothetical source location is a real discharge source.

[0121] S49. Normalize the joint likelihood function values ​​of all hypothetical source locations within the computational domain to generate the preliminary spatial likelihood distribution results of the discharge source.

[0122] Specifically, the joint likelihood function values ​​corresponding to all hypothetical source point locations within the computational domain are collected and uniformly organized according to the spatial arrangement order of the hypothetical source point grid set. Preferably, all joint likelihood function values ​​are normalized to map all values ​​to a preset unified probability interval, eliminating the numerical magnitude differences between different hypothetical source point locations. Further, the normalized joint likelihood function values ​​are bound to the spatial coordinates of the corresponding hypothetical source point locations, and spatial mapping and interpolation fitting are performed within the computational domain of the digital twin model to form a continuously distributed spatial probability field. For example, this spatial probability field is the preliminary spatial likelihood distribution result of the discharge source; regions with higher values ​​in the result correspond to a greater probability of the existence of a real discharge source.

[0123] In the above embodiment, the multiphysics path mismatch likelihood value is calculated using the Gaussian likelihood function, and the wave velocity co-correction penalty value is obtained using the exponential penalty function. The two are multiplied to obtain the joint likelihood function value, which is then normalized to generate a preliminary spatial likelihood distribution result, achieving a fusion of the dual constraints of path matching characteristics and wave velocity rationality. This embodiment introduces a wave velocity co-correction mechanism on the basis of traditional path mismatch evaluation, effectively suppressing the positioning deviation caused by propagation path distortion and wave velocity non-constancy, and significantly improving the accuracy and robustness of the confidence evaluation of the assumed source point location.

[0124] In one embodiment, S5 includes:

[0125] S51. Initialize the Gaussian mixture model parameters based on the preliminary spatial likelihood distribution results to obtain the initialized variational distribution parameters.

[0126] Specifically, the preliminary spatial likelihood distribution results are used as input data to initialize the parameters of the Gaussian mixture model. The Gaussian mixture model is a probabilistic model formed by a linear combination of multiple Gaussian probability density functions, capable of approximating complex spatial likelihood distributions. Preferably, based on the number of peaks and spatial distribution characteristics in the preliminary spatial likelihood distribution results, the number of Gaussian components in the Gaussian mixture model is determined. Then, the spatial coordinates corresponding to the high-likelihood regions in the preliminary spatial likelihood distribution results are selected as the initial values ​​for the mean vector in the Gaussian mixture model. Further, the initial values ​​of the covariance matrix are set according to the spatial scale range of the computational domain, and the weight coefficients of the Gaussian mixture model are uniformly assigned, completing the initialization process for all parameters and obtaining the initialized variational distribution parameters.

[0127] S52. Based on the preliminary spatial likelihood distribution results, the initial variational distribution parameters are iteratively optimized by minimizing the KL divergence between the variational distribution and the true posterior distribution to obtain the optimized variational distribution parameters.

[0128] Specifically, the initial spatial likelihood distribution is considered as the target posterior distribution to be fitted, and the Gaussian mixture model corresponding to the initialized variational distribution parameters is used as the approximate variational distribution. Parameter optimization is achieved by minimizing the KL divergence between the variational distribution and the true posterior distribution. The KL divergence quantifies the degree of difference between the two probability distributions; a smaller value indicates a higher degree of approximation. Preferably, an iterative update rule based on variational Bayesian inference is used to iteratively update the mean vector, covariance matrix, and weight coefficients of the Gaussian mixture model. Each iteration adaptively adjusts the parameters based on the fitting deviation between the current variational distribution parameters and the initial spatial likelihood distribution. Furthermore, the difference between the variational distribution and the target posterior distribution is continuously reduced during the iteration process. After one round of iteration, the updated and optimized variational distribution parameters are obtained, allowing the variational distribution to gradually approximate the true posterior distribution.

[0129] S53. Determine whether the optimized variational distribution parameters meet the preset iterative convergence conditions. If not, repeat step S52. If they meet, use the current optimized variational distribution parameters as the converged variational distribution parameters.

[0130] Specifically, the optimized variational distribution parameters obtained in the current iteration are read and compared component-by-component with the variational distribution parameters obtained in the previous iteration to calculate the parameter change magnitude between adjacent iterations. Preferably, the parameter change magnitude is compared with a preset iteration convergence condition, which includes a parameter change threshold and a maximum number of iterations as constraints. Specifically, when the parameter change magnitude is less than the parameter change threshold, or the number of iterations reaches the preset maximum number of iterations, the iteration convergence condition is satisfied, and the currently optimized variational distribution parameters are used as the converged variational distribution parameters. If the parameter change magnitude is greater than the parameter change threshold but the maximum number of iterations has not been reached, the convergence condition is not satisfied, and the iteration optimization step is returned to continue the parameter update operation until the convergence condition is satisfied.

[0131] S54. Extract the mean vector from the converged variational distribution parameters as the local discharge source location result.

[0132] Specifically, the mean vector of the corresponding Gaussian mixture model is extracted from the converged variational distribution parameters. This mean vector contains three-dimensional spatial coordinate components, corresponding to the horizontal, vertical, and longitudinal coordinates within the computational domain of the primary equipment of the strain gauge power station. Preferably, the spatial coordinates represented by the mean vector are aligned and mapped with the spatial coordinate system of the digital twin model, so that the spatial position corresponding to the mean vector can accurately match the actual physical position inside the primary equipment. Further, this spatial coordinate is determined as the positioning result of the local discharge source. This positioning result is the expected spatial position obtained when the variational distribution converges to the optimal state, which can effectively eliminate spurious peak interference in the initial spatial likelihood distribution and has higher positioning stability and accuracy.

[0133] S55. Extract the covariance matrix from the converged variational distribution parameters and use it as the confidence interval evaluation index for the positioning results.

[0134] Specifically, the covariance matrix of the corresponding Gaussian mixture model is extracted from the converged variational distribution parameters. This covariance matrix can quantitatively describe the dispersion and distribution direction of the variational distribution in three-dimensional space. Preferably, eigenvalue decomposition and eigenvector extraction are performed on the covariance matrix. The magnitude of the eigenvalues ​​is used to determine the degree of uncertainty of the positioning result in different spatial directions. The smaller the eigenvalue, the smaller the dispersion of the positioning in that direction, and the more reliable the positioning result. Furthermore, the covariance matrix and the spatial distribution characteristics derived from it are used together as the confidence interval evaluation index of the positioning result. This index can intuitively reflect the credibility range and error level of the positioning result, providing a quantitative reliability evaluation basis for the engineering application of the positioning result.

[0135] In the above embodiment, the parameters of a Gaussian mixture model are initialized based on the preliminary spatial likelihood distribution results. Then, the variational distribution parameters are iteratively optimized until convergence with the goal of minimizing the KL divergence. The mean vector is extracted to obtain the local discharge source location result, and the covariance matrix is ​​extracted to obtain the confidence interval evaluation index, thus completing high-precision positioning and reliability quantitative evaluation. This embodiment uses variational Bayesian inference to achieve the optimal approximation of the complex likelihood distribution, effectively suppressing false positioning points and noise interference. While improving positioning accuracy, it also provides quantitative confidence, solving the problem of insufficient confidence in traditional positioning methods and significantly improving the engineering practicality of partial discharge positioning of primary equipment in substations.

[0136] In the aforementioned multi-source sensor fusion localization method and system for partial discharge of primary equipment in substations, a digital twin model that synchronously follows Huygens' principle and Snell's law is constructed based on the three-dimensional geometric structure parameters, electromagnetic characteristic parameters, and acoustic characteristic parameters of the primary equipment in the substation. Then, multipath decomposition and time delay cluster feature extraction are performed on the ultra-high frequency electromagnetic wave signals and acoustic emission signals collected by the multi-source sensor array. Subsequently, multi-modal propagation path forward simulation and path mismatch degree calculation are carried out on all assumed source point locations within the digital twin model to form a path dependence matrix. Then, based on the path dependence matrix, the wave velocity collaborative correction factor is calculated and multi-physics joint likelihood estimation is completed to obtain the preliminary spatial likelihood distribution results. Finally, the probability distribution is iteratively optimized through variational Bayesian inference to output the localization results of the partial discharge source and the confidence interval evaluation index. This technical solution effectively overcomes the shortcomings of traditional positioning methods, such as the mismatch between the assumptions of electromagnetic wave and sound wave propagation models and the actual complex propagation behavior, and the tendency to converge to false positioning points, by constructing a digital twin model that fits the actual physical propagation process and introducing multipath decomposition, path mismatch calculation, wave velocity cooperative correction and variational Bayesian inference. It significantly improves the accuracy, confidence and engineering applicability of partial discharge positioning of primary equipment in substations.

[0137] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0138] Based on the same inventive concept, this application also provides a system for implementing the above-mentioned method for multi-source sensor fusion localization of partial discharge in substation primary equipment. The solution provided by this system is similar to the solution described in the above method. Therefore, the specific limitations of one or more embodiments of the multi-source sensor fusion localization system for partial discharge in substation primary equipment provided below can be found in the limitations of the multi-source sensor fusion localization method for partial discharge in substation primary equipment described above, and will not be repeated here.

[0139] In one exemplary embodiment, such as Figure 2 As shown, a multi-source sensor fusion positioning system 10 for partial discharge of primary equipment in a substation is provided, comprising:

[0140] The digital twin construction module 11 is used to construct a digital twin model based on the three-dimensional geometric structure parameters, electromagnetic characteristic parameters and acoustic characteristic parameters of the internal medium of the substation primary equipment; the digital twin model is used to simultaneously simulate the propagation behavior of ultra-high frequency electromagnetic waves and sound waves inside the equipment following Huygens' principle and Snell's law;

[0141] The multipath decomposition module 12 is used to acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals, and to perform multipath decomposition on the ultra-high frequency electromagnetic wave signals and acoustic wave signals respectively to obtain the electromagnetic wave multipath delay cluster feature vector and the acoustic wave multipath delay cluster feature vector; the ultra-high frequency electromagnetic wave signals and acoustic emission signals are synchronously acquired by a multi-source sensor array deployed on the device.

[0142] The path mismatch calculation module 13 is used to calculate the path mismatch of all assumed source point positions in the computational domain of the digital twin model based on the electromagnetic wave multipath delay cluster feature vector and the acoustic wave multipath delay cluster feature vector, and obtain the path dependency matrix.

[0143] The joint likelihood estimation module 14 is used to calculate the wave velocity cooperative correction factor for each hypothetical source point location based on the path dependence matrix, and to perform joint likelihood estimation based on the wave velocity cooperative correction factor corresponding to each hypothetical source point location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix to obtain the preliminary spatial likelihood distribution result of the discharge source.

[0144] The variational Bayesian inference module 15 is used to perform variational Bayesian inference on the preliminary spatial likelihood distribution results to obtain the local discharge source location results and the confidence interval evaluation index of the location results.

[0145] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps of the multi-source sensor fusion positioning method for partial discharge of primary equipment in a substation as described above.

[0146] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps of the multi-source sensor fusion localization method for partial discharge of primary equipment in a substation as described above.

[0147] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can be referred to in the description of the method embodiments. The device embodiments described above are merely illustrative. The components described as separate parts may or may not be physically separate, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this disclosure according to actual needs. Those skilled in the art can understand and implement this without creative effort.

[0148] The above-described embodiments are merely illustrative of several implementation methods of the embodiments of this application, and their descriptions are relatively specific and detailed. However, they should not be construed as limiting the scope of the patent application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the embodiments of this application, and these modifications and improvements all fall within the protection scope of the embodiments of this application.

Claims

1. A multi-source sensor fusion localization method for partial discharge of primary equipment in a substation, characterized in that, The method includes: S1. Based on the three-dimensional geometric structural parameters, electromagnetic characteristic parameters, and acoustic characteristic parameters of the internal medium of the substation primary equipment, a digital twin model is constructed; the digital twin model is used to simultaneously simulate the propagation behavior of ultra-high frequency electromagnetic waves and sound waves inside the equipment following Huygens' principle and Snell's law. S2. Acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals, and perform multipath decomposition on the ultra-high frequency electromagnetic wave signals and acoustic wave signals respectively to obtain electromagnetic wave multipath delay cluster feature vectors and acoustic wave multipath delay cluster feature vectors; the ultra-high frequency electromagnetic wave signals and acoustic emission signals are synchronously acquired by a multi-source sensor array deployed on the device. S3. Based on the electromagnetic wave multipath delay cluster feature vector and the acoustic wave multipath delay cluster feature vector, the path mismatch degree of all assumed source point positions in the computational domain of the digital twin model is calculated to obtain the path dependency matrix. S4. Based on the path dependence matrix, the wave velocity cooperative correction factor is calculated for each hypothetical source location, and joint likelihood estimation is performed based on the wave velocity cooperative correction factor corresponding to each hypothetical source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix to obtain the preliminary spatial likelihood distribution result of the discharge source. S5. Perform variational Bayes inference on the preliminary spatial likelihood distribution results to obtain the localization results of the local discharge source and the confidence interval evaluation index of the localization results.

2. The method according to claim 1, characterized in that, S1 includes: S11. Based on the three-dimensional geometric structure parameters, construct a three-dimensional geometric model of the primary equipment of the substation; S12. Based on the electromagnetic characteristic parameters and the acoustic characteristic parameters, different medium regions in the three-dimensional geometric model are respectively assigned corresponding electromagnetic property materials and acoustic property materials to obtain a geometric model with material assignment. S13. Perform finite element meshing on the geometric model of the material assignment to generate a computational mesh model that simultaneously includes electromagnetic field solution nodes and acoustic field solution nodes, thereby obtaining the digital twin model.

3. The method according to claim 1, characterized in that, S2 include: S21. The ultra-high frequency electromagnetic wave signal is decomposed into multipath components using a matching pursuit algorithm to extract the arrival time of each reflected wave and generate the electromagnetic wave multipath delay cluster feature vector. S22. The acoustic emission signal is subjected to mode separation of longitudinal and transverse waves by wavelet packet transform to obtain longitudinal wave waveform data and transverse wave waveform data; S23. The longitudinal wave waveform data and the transverse wave waveform data are decomposed into multipath components using a matching pursuit algorithm, the arrival time of each refracted wave is extracted, and the acoustic longitudinal wave multipath delay cluster feature vector and the acoustic transverse wave multipath delay cluster feature vector are generated to obtain the acoustic multipath delay cluster feature vector.

4. The method according to claim 3, characterized in that, S3 include: S31. Discretize the computational domain of the digital twin model in three dimensions according to a preset step size to generate a set of hypothetical source point grids consisting of all hypothetical source point positions. S32. Based on the digital twin model, the electromagnetic wave propagation path of each hypothetical source point location in the hypothetical source point grid set is simulated forward using the ray tracing algorithm to obtain the electromagnetic wave simulation multipath delay cluster corresponding to each hypothetical source point location. S33. Based on the digital twin model, the acoustic wave propagation path is simulated in the forward direction for each hypothetical source point location in the hypothetical source point grid set by solving the equation of the process function, so as to obtain the acoustic wave longitudinal wave simulation multipath delay cluster and acoustic wave transverse wave simulation multipath delay cluster corresponding to each hypothetical source point location. S34. Perform path-by-path comparison calculations between the electromagnetic wave simulation multipath delay cluster corresponding to each assumed source point location and the feature vector of the electromagnetic wave multipath delay cluster to obtain the electromagnetic wave path mismatch degree for each assumed source point location. S35. Perform path-by-path comparison calculation between the acoustic longitudinal wave simulation multipath delay cluster corresponding to each assumed source point location and the feature vector of the acoustic longitudinal wave multipath delay cluster to obtain the acoustic longitudinal wave path mismatch degree for each assumed source point location. S36. Perform path-by-path comparison calculations between the acoustic transverse wave simulation multipath delay cluster corresponding to each assumed source point location and the feature vector of the acoustic transverse wave multipath delay cluster to obtain the acoustic transverse wave path mismatch degree for each assumed source point location. S37. Combine the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch at each hypothetical source location into a multidimensional mismatch vector, and arrange all the multidimensional mismatch vectors at the hypothetical source locations in a matrix within the computational domain to obtain the path dependence matrix.

5. The method according to claim 4, characterized in that, The wave velocity cooperative correction factor calculated for each assumed source location as described in S4 includes: S41. Based on the path dependency matrix, extract the electromagnetic wave path mismatch degree of each assumed source point location, and calculate the equivalent path velocity of the electromagnetic wave corresponding to each assumed source point location according to the ratio of the straight distance from the assumed source point to each sensor to the electromagnetic wave simulation propagation time. S42. Based on the path dependency matrix, extract the longitudinal wave path mismatch of each hypothetical source point location, and calculate the equivalent path velocity of the longitudinal wave corresponding to each hypothetical source point location according to the ratio of the straight distance from the hypothetical source point to each sensor to the simulated propagation time of the longitudinal wave. S43. Based on the path dependency matrix, extract the acoustic transverse wave path mismatch degree of each assumed source point location, and calculate the equivalent path velocity of the acoustic transverse wave corresponding to each assumed source point location according to the ratio of the straight distance from the assumed source point to each sensor to the simulated propagation time of the acoustic transverse wave. S44. Based on the preset nominal velocity of electromagnetic waves, nominal velocity of longitudinal sound waves, and nominal velocity of transverse sound waves, calculate the degree of variation of the equivalent path velocity of electromagnetic waves, the equivalent path velocity of longitudinal sound waves, and the equivalent path velocity of transverse sound waves relative to the corresponding nominal velocities, and obtain three coefficients of variation. S45. Based on the three coefficients of variation, a weighted summation is performed to obtain the wave velocity co-correction factor corresponding to each assumed source point location.

6. The method according to claim 5, characterized in that, The joint likelihood estimation described in S4, based on the wave velocity cooperative correction factor corresponding to each hypothetical source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix, includes: S46. Based on the electromagnetic wave path mismatch, acoustic longitudinal wave path mismatch, and acoustic transverse wave path mismatch corresponding to each assumed source point location, the multiphysics path mismatch likelihood value of the corresponding assumed source point location is calculated using the Gaussian likelihood function. S47. Based on the wave velocity collaborative correction factor corresponding to each assumed source point location, the wave velocity collaborative correction penalty value for the corresponding assumed source point location is calculated using the exponential penalty function. S48. Multiply the multiphysics path mismatch likelihood value of each hypothetical source point location by the wave velocity co-correction penalty value to obtain the joint likelihood function value of that hypothetical source point location. S49. Normalize the joint likelihood function values ​​of all assumed source locations within the computational domain to generate the preliminary spatial likelihood distribution of the discharge source.

7. The method according to claim 6, characterized in that, S5 include: S51. Initialize the Gaussian mixture model parameters of the preliminary spatial likelihood distribution results to obtain the initialized variational distribution parameters; S52. Based on the preliminary spatial likelihood distribution results, the initialized variational distribution parameters are iteratively optimized by minimizing the KL divergence between the variational distribution and the true posterior distribution to obtain the optimized variational distribution parameters. S53. Determine whether the optimized variational distribution parameters meet the preset iterative convergence conditions. If not, repeat step S52. If they meet, use the current optimized variational distribution parameters as the converged variational distribution parameters. S54. Extract the mean vector from the converged variational distribution parameters as the local discharge source location result; S55. Extract the covariance matrix from the converged variational distribution parameters and use it as the confidence interval evaluation index for the positioning result.

8. A multi-source sensor fusion positioning system for partial discharge of primary equipment in a substation, characterized in that, The system includes: The digital twin construction module is used to construct a digital twin model based on the three-dimensional geometric parameters, electromagnetic properties parameters, and acoustic properties parameters of the internal medium of the substation primary equipment; the digital twin model is used to simultaneously simulate the propagation behavior of ultra-high frequency electromagnetic waves and sound waves inside the equipment according to Huygens' principle and Snell's law; The multipath decomposition module is used to acquire ultra-high frequency electromagnetic wave signals and acoustic emission signals, and to perform multipath decomposition on the ultra-high frequency electromagnetic wave signals and acoustic wave signals respectively to obtain electromagnetic wave multipath delay cluster feature vectors and acoustic wave multipath delay cluster feature vectors; the ultra-high frequency electromagnetic wave signals and acoustic emission signals are synchronously acquired by a multi-source sensor array deployed on the device. The path mismatch calculation module is used to calculate the path mismatch of all assumed source point positions in the computational domain of the digital twin model based on the electromagnetic wave multipath delay cluster feature vector and the acoustic wave multipath delay cluster feature vector, and obtain the path dependency matrix. The joint likelihood estimation module is used to calculate the wave velocity cooperative correction factor for each hypothetical source location based on the path dependence matrix, and to perform joint likelihood estimation based on the wave velocity cooperative correction factor corresponding to each hypothetical source location and the electromagnetic wave path mismatch and acoustic wave path mismatch contained in the path dependence matrix to obtain the preliminary spatial likelihood distribution result of the discharge source. The variational Bayesian inference module is used to perform variational Bayesian inference on the preliminary spatial likelihood distribution results to obtain the localization results of the local discharge source and the confidence interval evaluation index of the localization results.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images