GIS equipment defect accurate positioning method and system based on acoustic visualization

Abstract

The invention discloses a GIS equipment defect accurate positioning method and system based on acoustic visualization, and relates to the technical field of defect localization, and the method comprises the steps: collecting a vibration acoustic mixed signal in an equipment operation state, and separating the vibration acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis; an acoustic characteristic spectrum is constructed for the pure acoustic signals, a sound field propagation path model is generated, and meanwhile a corresponding relation graph of equipment components and the acoustic characteristic spectrum is established; and determining a source component and a specific position of the abnormal acoustic characteristic spectrum through an acoustic reverse tracking positioning algorithm in combination with the sound field propagation path model and the corresponding relation map. According to the method, by combining vibration acoustic conjoint analysis and acoustic reverse tracking, high-precision positioning of defects of hidden parts in the GIS equipment is achieved, the positioning precision and the recognition accuracy are improved, recognizable defect types are expanded, and the method is suitable for GIS equipment of different structures.

Description

Technical Field

[0001] The present invention relates to the technical field of defect location, and specifically to a method and system for accurately locating GIS device defects based on acoustic visualization. Background Art

[0002] As a key device in modern power systems, the gas-insulated metal-enclosed switchgear (GIS) is crucial for the safe operation of the power grid. The early identification and accurate location of internal defects in GIS devices have long been the focus of research in the field of power equipment condition monitoring. Currently, common GIS defect detection methods include ultrasonic detection, ultra-high frequency (UHF) detection, transient earth voltage (TEV) detection, and infrared thermal imaging. Among them, ultrasonic detection technology has been widely used in GIS device defect diagnosis due to its advantages such as non-contact, high sensitivity, and strong anti-interference ability. However, traditional ultrasonic detection methods mainly locate defects based on the energy amplitude or time difference of acoustic signals, and these methods have many limitations in processing acoustic signals in complex structural environments. Especially in the GIS device, the internal structure is complex, and the SF6 gas, insulating material, and metal shell form a multi-phase medium environment, resulting in multi-path effects, attenuation, and reflection phenomena during the propagation of sound waves, making the defect acoustic signals severely distorted. Traditional positioning methods are difficult to achieve high-precision positioning, especially for defects in hidden parts inside the device.

[0003] The traditional GIS device defect location technology mainly has the following technical bottlenecks: First, it is difficult to separate the mixture of acoustic signals and vibration signals. The mechanical vibration and electromagnetic vibration of the device itself will interfere with the defect acoustic signals, reducing the signal-to-noise ratio and affecting the detection sensitivity. Second, existing methods usually ignore the complex propagation characteristics of sound waves in the multi-phase medium of GIS devices and do not fully consider the influence of multi-path effects and reflection phenomena on acoustic signals, resulting in low positioning accuracy. Especially for defects in hidden positions such as flange connections and the back of insulators, the positioning error can reach 15 - 20 cm. Third, there is a lack of multi-dimensional analysis of acoustic features. Existing technologies mainly focus on signal energy or time-domain features and ignore the rich information contained in frequency-domain and time-frequency domain features, restricting the scope of defect type identification. Finally, traditional positioning algorithms usually adopt deterministic models and are difficult to cope with practical problems such as the complex internal structure and uncertain acoustic environment of GIS devices, lacking flexibility and self-adaptability. These technical limitations seriously restrict the early discovery and precise handling of GIS device defects, increasing the equipment maintenance cost and power outage risk. Summary of the Invention

[0004] In view of the above problems, the present invention is proposed.

[0005] Therefore, the present invention provides a method and system for precise positioning of GIS device defects based on acoustic visualization, which can solve the problems mentioned in the background art.

[0006] To solve the above technical problems, the present invention provides the following technical solutions: A method for precise positioning of GIS device defects based on acoustic visualization, including: collecting a vibration-acoustic mixed signal under the operating state of the device, and separating the vibration-acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis; the independent component analysis includes a blind source separation algorithm based on minimizing mutual information; constructing an acoustic feature spectrum for the pure acoustic signal, generating a sound field propagation path model, and establishing a corresponding relationship map between the device components and the acoustic feature spectrum; the acoustic feature spectrum includes anomaly quantification indexes for three types of features in the time domain, frequency domain, and time-frequency domain; combining the sound field propagation path model and the corresponding relationship map, and determining the originating component and specific location of the abnormal acoustic feature spectrum through an acoustic backtracking positioning algorithm; the acoustic backtracking positioning algorithm includes a multi-path sound source positioning method based on Bayesian optimization.

[0007] As a preferred solution of the method for precise positioning of GIS device defects based on acoustic visualization according to the present invention, wherein: the step of separating the vibration-acoustic mixed signal by the independent component analysis includes: constructing a multi-channel blind source separation model based on joint approximate diagonalization; calculating the mutual information measure of the vibration-acoustic mixed signal, if the mutual information measure is less than the previous iteration result, then continue to iteratively optimize the blind source separation model, if the mutual information measure no longer decreases or reaches the maximum number of iterations, then output the current separation matrix, and use the separation matrix to decompose the vibration-acoustic mixed signal into the pure acoustic signal and the vibration signal.

[0008] As a preferred solution of the method for precise positioning of GIS device defects based on acoustic visualization according to the present invention, wherein: the step of constructing the acoustic feature spectrum includes: extracting time domain features, frequency domain features, and time-frequency domain features from the pure acoustic signal, wherein the time domain features include kurtosis factors, the frequency domain features include frequency band energy distributions, and the time-frequency domain features include wavelet energy envelopes; based on the time domain features, the frequency domain features, and the time-frequency domain features, constructing an anomaly quantification index for the acoustic feature spectrum.

[0009] As a preferred embodiment of the method for accurately locating GIS device defects based on acoustic visualization according to the present invention, the steps of generating the sound field propagation path model include: establishing a geometric model of the internal structure of the GIS device; calculating the propagation characteristics of sound waves in the internal structure of the GIS device based on the sound wave propagation theory and finite element analysis. If the internal structure of the GIS device is a multi-phase medium structure, the sound wave reflection and transmission laws are applied at the medium interface to calculate the sound wave propagation parameters. If the internal structure of the GIS device is a single medium structure, the sound wave propagation parameters are directly calculated. A sound field propagation path model including the sound attenuation coefficient, reflection coefficient, and transmission delay is established according to the calculation results.

[0010] As a preferred embodiment of the method for accurately locating GIS device defects based on acoustic visualization according to the present invention, the steps of establishing the correspondence map between the device components and the acoustic characteristic spectrum include: obtaining an acoustic characteristic spectrum database of typical components of the GIS device under normal operation and various fault conditions; establishing a mapping relationship matrix between components and characteristics, and establishing a correspondence between the physical structure position of the GIS device and the eigenvectors of the acoustic characteristic spectrum.

[0011] As a preferred embodiment of the method for accurately locating GIS device defects based on acoustic visualization according to the present invention, the steps of the acoustic backtracking localization algorithm include: receiving the acoustically characteristic spectrum collected in real time and determining the abnormal acoustically characteristic spectrum; constructing a Bayesian probability graph model based on the sound field propagation path model, and performing Markov chain Monte Carlo sampling using the multi-path sound source position hypothesis space. If the posterior probability value of a certain position point is greater than the posterior probability values of the surrounding areas, then this position point is recorded as a candidate defect source point. Iterative calculation is performed until the posterior probability distribution converges, and the position point corresponding to the maximum posterior probability is determined as the accurate defect position.

[0012] As a preferred embodiment of the method for accurately locating GIS device defects based on acoustic visualization according to the present invention, the steps of the Markov chain Monte Carlo sampling include: setting the prior distribution of the sound source position, and calculating the likelihood function of each assumed position based on the sound field propagation path model; iteratively updating the sampling points through the acceptance-rejection criterion, and calculating the state transition ratio between the new sampling point and the previous sampling point. If the state transition ratio is greater than 1, the new sampling point is directly accepted. If the state transition ratio is less than 1, the new sampling point is accepted or rejected according to the value of the state transition ratio as the probability. Otherwise, the new sampling point is rejected and the previous sampling point is returned. Iterative execution is performed until the sampling sequence is stable, and finally a three-dimensional display graph is generated to display the defect position and degree.

[0013] To further solve the above technical problems, the present invention provides the following technical solutions: A GIS device defect precise positioning system based on acoustic visualization, comprising: a signal separation module, configured to collect the vibration-acoustic mixed signal under the operating state of the device, and separate the vibration-acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis; a feature modeling module, configured to construct an acoustic feature spectrum for the pure acoustic signal, generate a sound field propagation path model, and establish a corresponding relationship map between the device components and the acoustic feature spectrum; a positioning and tracking module, configured to combine the sound field propagation path model and the corresponding relationship map, and determine the origin component and the specific location of the abnormal acoustic feature spectrum through an acoustic backtracking positioning algorithm.

[0014] A computer device, comprising a memory and a processor, where the memory stores a computer program, and when the processor executes the computer program, the steps of the above-mentioned GIS device defect precise positioning method based on acoustic visualization are implemented.

[0015] A computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, the steps of the above-mentioned GIS device defect precise positioning method based on acoustic visualization are implemented.

[0016] The beneficial effects of the present invention: Through three steps of vibration-acoustic signal acquisition and separation, acoustic feature spectrum and propagation model construction, and defect precise positioning, the present invention realizes the precise positioning of defects in hidden parts inside GIS devices. Compared with traditional technologies, this method has higher positioning accuracy, a wider scope of application, and a better user experience, providing a new solution for the condition monitoring and fault diagnosis of GIS devices. Description of the Drawings

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the following will briefly introduce the drawings required for the description of the embodiments. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0018] Figure 1 It is a schematic diagram of the overall process of the GIS device defect precise positioning method based on acoustic visualization proposed by the present invention;

[0019] Figure 2 It is a diagram of the computer device in the GIS device defect precise positioning method based on acoustic visualization proposed by the present invention. Detailed Embodiments

[0020] To make the above objects, features, and advantages of the present invention more apparent and understandable, the following provides a detailed description of the specific embodiments of the present invention in conjunction with the accompanying drawings of the specification. Obviously, the described embodiments are part of the embodiments of the present invention, rather than all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the scope of protection of the present invention.

[0021] In the following description, many specific details are set forth to facilitate a full understanding of the present invention. However, the present invention may also be implemented in other ways different from those described herein. Those skilled in the art may make similar generalizations without departing from the spirit of the present invention. Therefore, the present invention is not limited by the specific embodiments disclosed below.

[0022] Example 1, referring to Figure 1 , which is an embodiment of the present invention, provides a method for accurately locating GIS equipment defects based on acoustic visualization.

[0023] The method for accurately locating GIS equipment defects based on acoustic visualization provided by the present invention is applied to the fault diagnosis process of gas-insulated metal-enclosed switchgear (GIS), and is particularly suitable for solving the problem of reduced positioning accuracy caused by the attenuation of acoustic characteristics propagation in hidden parts inside the GIS. Compared with the prior art, this method can improve the accuracy of defect location and expand the range of recognizable defect types.

[0024] Figure 1 FIG. shows a schematic diagram of the overall process of the method for accurately locating GIS equipment defects based on acoustic visualization, including the following steps:

[0025] S1: Attach distributed acoustic sensors to key positions on the outer shell of the GIS equipment, collect the vibration-acoustic mixed signal during the operation of the equipment, and separate the vibration-acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis.

[0026] Specifically, the independent component analysis includes a blind source separation algorithm based on minimizing mutual information.

[0027] Specifically, the steps of separating the vibration-acoustic mixed signal by independent component analysis include:

[0028] Construct a multi-channel blind source separation model based on joint approximate diagonalization;

[0029] Calculate the mutual information metric of the vibration-acoustic mixed signal. If the mutual information metric is less than the previous iteration result, continue to iteratively optimize the blind source separation model. If the mutual information metric no longer decreases or reaches the maximum number of iterations, output the current separation matrix, and use the separation matrix to decompose the vibration-acoustic mixed signal into a pure acoustic signal and a vibration signal.

[0030] To make the objectives, technical solutions, and advantages of the present invention clearer, the present invention will be further described in detail below in conjunction with specific embodiments.

[0031] In step S1, first, a distributed acoustic sensor is attached to key positions on the outer shell of the GIS device to collect the vibration-acoustic mixed signal during the operation of the device. Then, the vibration-acoustic mixed signal is separated into a pure acoustic signal and a vibration signal through independent component analysis.

[0032] When collecting the acoustic signal of the GIS device, according to the structural characteristics of the GIS device and the acoustic propagation theory, acoustic sensitive points on the device surface are selected for sensor arrangement. Usually, these key positions include, but are not limited to, the flange connection of the GIS device, around the insulator, the circuit breaker chamber, the busbar chamber, and positions where electrical contact points may exist, etc. These positions are the main propagation paths of the acoustic signal or parts prone to defects.

[0033] In this embodiment, the distributed acoustic sensor preferably adopts a combination of a high-sensitivity ultrasonic sensor and a structural vibration sensor, with a sensitivity range of 20 - 200 kHz, which can effectively capture the acoustic and vibration signals generated by typical defects such as internal discharge, mechanical looseness, and the presence of foreign objects in the GIS device. Multiple sensors form a sensor array, which can collect vibration-acoustic mixed signals from different angles and positions, providing multi-dimensional information for subsequent signal separation and defect location.

[0034] During the operation of the GIS device, internal defects will generate characteristic acoustic signals, but these signals will be mixed with the vibration signals of the device itself during propagation. Obtaining a pure acoustic signal is crucial for accurately identifying the defect type and locating the defect position. Therefore, the present invention uses independent component analysis technology to separate the collected vibration-acoustic mixed signal.

[0035] Specifically, independent component analysis includes a blind source separation algorithm based on minimizing mutual information. Compared with traditional signal separation methods based on second-order statistics, the blind source separation algorithm based on minimizing mutual information can make full use of the high-order statistical characteristics of the signal, effectively handle non-Gaussian distributed acoustic signals, and improve the signal separation effect in a low signal-to-noise ratio environment.

[0036] The specific steps for independent component analysis to separate the vibration-acoustic mixed signal are as follows:

[0037] First, a multi-channel blind source separation model based on joint approximate diagonalization is constructed.

[0038] Let X(t) = [x1(t), x2(t),..., x n (t)] Tis the vibration-acoustic hybrid signal vector collected by n sensors, S(t) = [s1(t), s2(t),..., s m (t)] T is the m independent source signal vectors, then the mixing model can be expressed as:

[0039] X(t) = AS(t) + N(t);

[0040] where, A is an n×m dimensional mixing matrix, and N(t) is a noise vector. The goal is to find a separation matrix W such that Y(t) = WX(t) can recover the original source signal S(t). In this embodiment, m = 2, corresponding to two source signals of pure acoustic signal and vibration signal.

[0041] To solve the problem that traditional independent component analysis is vulnerable to noise interference when processing GIS device acoustic signals, the present invention adopts the joint approximate diagonalization technology, and improves the separation ability of non-stationary acoustic signals by jointly diagonalizing multiple high-order cumulant matrices simultaneously. Specifically, when implementing, first construct the fourth-order cumulant matrix set {Q i (τ,k), i = 1, 2,..., L} of the signal, where τ and k are time delay parameters, and L is the number of selected matrices, usually taking values of 8 - 12, which can balance the computational complexity and separation performance.

[0042] Secondly, calculate the mutual information measure of the vibration-acoustic hybrid signal. Mutual information is an index to measure the mutual dependence between random variables, and when the variables are completely independent, the mutual information reaches the minimum value. In this embodiment, the mutual information I(Y) of the separated signal Y(t) can be expressed as:

[0043]

[0044] where, D(·||·) represents the KL divergence, p Y (Y) is the joint probability density function, is the marginal probability density function.

[0045] To minimize the mutual information, the present invention adopts an optimization algorithm based on natural gradient to iteratively update the separation matrix W:

[0046] W (k+1) = W (k) + η[I - φ(Y)(Y) T W (k) ;

[0047] where, η is the learning rate, φ(Y) is a non-linear activation function, for ultrasonic signals, selecting φ(Y) = tanh(Y) can obtain good separation effect. I is the identity matrix, and k represents the number of iterations.

[0048] After each iteration, the mutual information metric of the current separated signal is calculated. If the mutual information metric is less than the result of the previous iteration, the blind source separation model is continuously iteratively optimized, indicating that the separation effect is continuously improving; if the mutual information metric no longer decreases or reaches the maximum number of iterations (set to 500 times in this embodiment), the current separation matrix is output, and the vibro-acoustic mixed signal is decomposed into a pure acoustic signal and a vibration signal using the separation matrix. This adaptive termination condition based on the change of mutual information avoids the limitations of traditional fixed-threshold methods, can flexibly adjust the separation process according to the actual signal characteristics, and improves the separation effect and calculation efficiency.

[0049] After the above processing, the vibro-acoustic mixed signal is successfully separated into a pure acoustic signal and a vibration signal. The pure acoustic signal retains defect characteristics such as discharge and mechanical friction, while the vibration interference of the device itself is effectively filtered out, laying a foundation for subsequent acoustic feature extraction and defect location.

[0050] Compared with the prior art, the signal separation method of the present invention has the following beneficial effects:

[0051] 1. Through the blind source separation algorithm based on mutual information minimization, it can effectively process non-Gaussian distributed acoustic signals, overcoming the limitations of traditional principal component analysis (PCA) and independent component analysis (ICA) in dealing with complex acoustic environments;

[0052] 2. By introducing the joint approximate diagonalization technique, the separation ability for non-stationary acoustic signals is improved, which is especially suitable for the complex acoustic environment inside GIS devices;

[0053] 3. Using the adaptive iterative termination condition, it avoids overfitting and underfitting problems, and improves the separation effect and calculation efficiency;

[0054] 4. The entire separation process does not require prior knowledge of the specific location and characteristics of the sound sources, and has strong adaptability and robustness, and is applicable to GIS devices of various models and structures.

[0055] Through the above steps of vibro-acoustic signal acquisition and separation, a pure acoustic signal can be obtained, providing a high-quality data basis for subsequent acoustic feature extraction and defect location. This is a key prerequisite step for realizing acoustic visualization and precise defect location.

[0056] S2: Construct an acoustic feature spectrum for the pure acoustic signal, generate a sound field propagation path model, and establish a corresponding relationship map between the device components and the acoustic feature spectrum.

[0057] Specifically, the acoustic feature spectrum includes abnormality quantification indexes of three types of features in the time domain, frequency domain, and time-frequency domain.

[0058] Specifically, the steps of constructing the acoustic feature spectrum include:

[0059] Extract time-domain features, frequency-domain features, and time-frequency domain features from pure acoustic signals. The time-domain features include kurtosis factor, the frequency-domain features include frequency band energy distribution, and the time-frequency domain features include wavelet energy envelope;

[0060] Based on the time-domain features, frequency-domain features, and time-frequency domain features, construct an abnormality quantification index for the acoustic feature spectrum.

[0061] Specifically, the steps for generating the sound field propagation path model include:

[0062] Establish a geometric model of the internal structure of the GIS device;

[0063] Based on the acoustic wave propagation theory and finite element analysis, calculate the propagation characteristics of acoustic waves in the internal structure of the GIS device. If the internal structure of the GIS device is a multiphase medium structure, apply the acoustic wave reflection and transmission laws at the medium interface to calculate the acoustic wave propagation parameters. If the internal structure of the GIS device is a single medium structure, directly calculate the acoustic wave propagation parameters, and establish a sound field propagation path model including the sound attenuation coefficient, reflection coefficient, and transmission delay according to the calculation results.

[0064] Specifically, the steps for establishing the corresponding relationship map between device components and the acoustic feature spectrum include:

[0065] Obtain the acoustic feature spectrum database of typical components of the GIS device under normal operation and various fault states;

[0066] Establish a mapping relationship matrix between components and features, and establish a corresponding relationship between the physical structure position of the GIS device and the feature vectors of the acoustic feature spectrum.

[0067] Furthermore, after obtaining the separated pure acoustic signal, in step S2, it is necessary to construct an acoustic feature spectrum for the pure acoustic signal, generate a sound field propagation path model, and establish a corresponding relationship map between device components and the acoustic feature spectrum. Construct an acoustic feature spectrum for the pure acoustic signal. The acoustic feature spectrum includes abnormality quantification indexes of three types of features in the time domain, frequency domain, and time-frequency domain. This multi-dimensional feature extraction method can comprehensively capture the acoustic characteristics of GIS device defects.

[0068] The steps for constructing the acoustic feature spectrum are as follows: Extract time-domain features, frequency-domain features, and time-frequency domain features from the pure acoustic signal, and then based on these three types of features, construct an abnormality quantification index for the acoustic feature spectrum.

[0069] In the time-domain feature extraction, the time-domain features include kurtosis factor. The kurtosis factor is a statistic used to describe the sharpness of the signal waveform and is particularly sensitive to the identification of pulse signals. Defects such as internal discharge and breakdown in the GIS device usually generate obvious pulse signals, and the kurtosis factor can effectively capture such features. For the pure acoustic signal z(t), its kurtosis factor Kz Can be calculated as:

[0070]

[0071] Where N is the number of signal sampling points, z i is the amplitude of the i-th sampling point, and z is the signal mean. Compared with traditional time-domain analysis that only focuses on signal energy, the kurtosis factor can more sensitively reflect the pulse characteristics of the signal and effectively identify transient faults in GIS equipment. In addition to the kurtosis factor, the time-domain features of the present invention also include root mean square value, waveform factor, pulse factor, etc. These features together constitute the time-domain feature vector, which can comprehensively characterize the time-domain characteristics of the signal.

[0072] In frequency-domain feature extraction, the frequency-domain features include frequency band energy distribution. Different types of GIS equipment defects will cause energy concentration in different frequency bands. For example, partial discharge usually has obvious energy distribution in the 30 - 80 kHz frequency band, while mechanical looseness is mainly concentrated in the 5 - 20 kHz frequency band. In this embodiment, the spectrum of the pure acoustic signal is divided into multiple frequency bands, and the energy ratio of each frequency band is calculated to form a frequency band energy distribution feature vector, which can accurately distinguish different types of defects. Specifically, when implemented, the fast Fourier transform (FFT) is used to convert the time-domain signal to the frequency domain, and then the 0 - 200 kHz frequency spectrum range is divided into 10 sub-frequency bands, and the percentage of the energy of each sub-frequency band in the total energy is calculated to form a 10-dimensional frequency band energy distribution feature vector. In addition, the frequency-domain features also include parameters such as spectrum peak value, spectrum centroid, and spectrum standard deviation, which are used to describe the overall morphological features of the spectrum.

[0073] In time-frequency domain feature extraction, the time-frequency domain features include wavelet energy envelope. Compared with the traditional Fourier transform that only reflects the frequency characteristics of the signal, wavelet transform can provide local information of both time and frequency at the same time, which is especially suitable for analyzing non-stationary acoustic signals in GIS equipment. In this embodiment, multi-scale wavelet decomposition is used to extract the energy envelopes of different frequency bands, which can be expressed as:

[0074] E j (t) = |C j (t)| 2 ;

[0075] Where E j (t) is the energy envelope of the j-th layer of wavelet decomposition, C j(t) is the corresponding wavelet coefficient. The wavelet energy envelope can reflect the variation law of signal energy with time and capture the transient and continuous characteristics of the defect signal. In this embodiment, the db4 wavelet is selected as the basic wavelet, and the pure acoustic signal is decomposed into 5 levels, obtaining 5 detail coefficients and 1 approximation coefficient. Calculate the energy envelope of each level of detail coefficient, which together constitute the time-frequency domain features. This multi-resolution analysis method can effectively capture the transient features in different frequency bands and is especially suitable for identifying short-time discharge signals in GIS devices.

[0076] After extracting the above three types of features, the present invention constructs an abnormality quantification index for the acoustic feature spectrum. This index comprehensively considers the changes in time domain, frequency domain, and time-frequency domain features and can quantitatively evaluate the abnormal degree of the operating state of GIS devices. Specifically, compare various features with the normal operation baseline of the device, calculate the deviation degree, and synthesize the deviation degrees of each feature into a unified abnormality quantification index through a non-linear mapping function. The calculation of the abnormality quantification index adopts the Mahalanobis distance method, which considers the correlation between features and can more accurately measure the deviation degree of the feature vector from the normal baseline. Let the mean of the feature vector in the normal state be μ F , and the covariance matrix be Σ F , and the current feature vector be F, then the abnormality quantification index D M can be expressed as: When the value of D M exceeds the preset threshold, it is determined to be in an abnormal state. This method of multi-dimensional feature fusion overcomes the disadvantage that single features are easily interfered with and improves the accuracy and robustness of defect recognition.

[0077] Secondly, generate the sound field propagation path model. The internal structure of GIS devices is complex, and phenomena such as attenuation, reflection, and refraction will occur during the propagation of sound waves, affecting the accuracy of defect location. To solve this problem, the present invention constructs a sound field propagation path model to simulate the propagation law of sound waves inside GIS devices.

[0078] The steps to generate the sound field propagation path model are as follows: First, establish a geometric model of the internal structure of the GIS device; then, based on the sound wave propagation theory and finite element analysis, calculate the propagation characteristics of sound waves in the internal structure of the GIS device.

[0079] When establishing the geometric model of the internal structure of GIS equipment, factors such as the internal components of the equipment, the distribution of SF6 gas, and the position of insulators need to be considered. In this embodiment, three-dimensional modeling technology is adopted, combined with the equipment structure drawings and actual measurement data, to construct an accurate internal geometric model, providing a spatial basis for sound field analysis. Specifically, when implementing, first import the basic geometric structure according to the CAD design drawings of the GIS equipment, and then calibrate the model through actual size measurement to ensure the accuracy of the geometric model. For the internal complex structures (such as circuit breaker contacts, insulating supports, etc.), simplified models are used to represent them, retaining their main acoustic characteristics while taking into account the calculation efficiency. This geometric model is finally discretized into a finite element model containing hundreds of thousands of mesh units, providing a spatial mesh basis for subsequent sound field analysis.

[0080] When calculating the acoustic wave propagation characteristics, the present invention classifies and processes according to the medium structure inside the GIS equipment: if the inside of the GIS equipment is a multiphase medium structure (such as the coexistence of SF6 gas and solid insulating materials), the acoustic wave reflection and transmission laws are applied at the medium interface to calculate the acoustic wave propagation parameters; if the inside of the GIS equipment is a single medium structure, the acoustic wave propagation parameters are directly calculated. This adaptive calculation method can cope with GIS equipment of different structures and has wide applicability. For the multiphase medium structure, the present invention applies the acoustic wave reflection and transmission laws at the medium interface, and according to the acoustic impedances Z A and Z B of the two media, calculate the reflection coefficient R c and the transmission coefficient T c : For acoustic waves of different frequencies, considering the frequency-dependent attenuation characteristics, a frequency-dependent attenuation coefficient is introduced into the model.

[0081] The attenuation of acoustic waves during propagation can be expressed as:

[0082] A s (d) = A s0 e -βd ;

[0083] where A s (d) is the amplitude of the acoustic wave after the propagation distance d, A s0 is the initial amplitude at the sound source, β is the attenuation coefficient, which is related to the medium properties and frequency. The attenuation coefficient β consists of two parts: the attenuation β abs caused by medium absorption and the attenuation β geo caused by geometric diffusion. For the acoustic wave propagation in SF6 gas, β abs is proportional to the square of the frequency, while β geo is inversely proportional to the propagation distance. In this embodiment, the attenuation coefficients of SF6 gas at different frequencies are obtained through experimental measurement, and a frequency-attenuation relationship curve is established to improve the accuracy of the sound field model.

[0084] At the interface, the reflection and transmission of sound waves follow Snell's law, and the reflection coefficient and transmission coefficient depend on the difference in acoustic impedance between the two media. By calculating the attenuation coefficient, reflection coefficient, and transmission delay of each propagation path, a sound field propagation path model containing the above parameters is finally established. For some relatively complex structures inside the GIS, sound waves may undergo multiple reflections and refractions, forming multiple propagation paths. To comprehensively consider the multipath effect, in this embodiment, the ray tracing algorithm is used to simulate the process of sound waves emitted from the sound source propagating along different paths to the sensor, and calculate the propagation time delay and attenuation of each path. Let the transmission delay of the m-th path from the sound source p to the sensor q be The attenuation coefficient is Then the contribution of this path to the received signal can be expressed as:

[0085]

[0086] where z(t) is the sound source signal, is the path length, is the path gain coefficient. Compared with the traditional simplified model, this model fully considers the complex structure and multipath effect inside the GIS device, can more accurately simulate the sound wave propagation law, and provides accurate acoustic path information for defect location.

[0087] Finally, a correspondence map between the device components and the acoustic feature spectrum is established. This step aims to establish the mapping relationship between the physical structure of the GIS device and the acoustic features, providing a knowledge basis for defect source location.

[0088] The steps to establish the correspondence map between the device components and the acoustic feature spectrum are as follows: First, obtain the acoustic feature spectrum database of typical components of the GIS device under normal operation and various fault states; then, establish a mapping relationship matrix between the components and the features, and establish the correspondence between the physical structure position of the GIS device and the feature vectors of the acoustic feature spectrum.

[0089] When obtaining the acoustic feature spectrum database, acoustic signals of typical components of GIS equipment (such as circuit breakers, insulators, conductor joints, etc.) under different fault states (such as partial discharge, mechanical looseness, foreign object contact, etc.) are collected through a combination of laboratory simulation and on-site acquisition, and the corresponding acoustic feature spectra are extracted. These data constitute the "acoustic fingerprint" database, which is the basis for defect identification. In the laboratory simulation stage, scaled-down models and actual equipment components are used to simulate different types of defects (such as tip discharge, floating potential body discharge, free metal particle discharge, etc.), and the acoustic feature spectra of various defects are recorded. In the on-site acquisition stage, acoustic data are collected from actually operating GIS equipment, especially equipment with confirmed defects. The defect types and locations are verified through disassembly inspection, and an acoustic feature spectrum database in the real environment is established. To enhance the adaptability of the database, data of GIS equipment from different manufacturers, different models, and different operating years are collected to ensure the generalization ability of the database.

[0090] When establishing the mapping relationship matrix, the present invention adopts a two-way mapping strategy: on the one hand, according to the equipment structure diagram, a forward mapping is established between physical components and the acoustic features they potentially generate; on the other hand, based on acoustic theory and experimental data, specific acoustic feature spectra are inversely mapped to possible physical positions. When specifically implemented, a mapping relationship matrix M of P×Q dimensions is constructed R , where P is the number of physical partitions of the GIS equipment, and Q is the feature dimension of the acoustic feature spectrum. The matrix element M R (i,j) represents the association strength between the i-th physical partition and the j-th acoustic feature, and the value range is [0,1]. The association strength is extracted from the database through statistical learning methods. Specifically, an ensemble learning model is constructed using multiple classifiers such as support vector machine SVM and random forest Random Forest to improve the accuracy and robustness of the mapping relationship. This two-way mapping mechanism enhances the reliability of positioning. Even when the acoustic signal is severely interfered, a reasonable positioning result can still be provided through structural constraints.

[0091] Compared with traditional methods, the corresponding relationship map constructed by the present invention not only considers the acoustic features of single-point faults but also analyzes the superposition effect and interaction of multi-point faults, and can handle complex fault scenarios in GIS equipment. In addition, the map adopts an adaptive update mechanism, which can continuously optimize and expand according to new fault cases, improving the learning ability and adaptability of the system. The adaptive update mechanism is based on the Bayesian update principle, taking the newly collected data as prior knowledge and continuously optimizing the mapping relationship matrix. Let the current mapping relationship matrix be M R , and the new data be D n , then the updated mapping relationship matrix M R′ can be calculated through the Bayesian formula: P(M R′ |D n) ∝ P(D n |M R ) P(M R , where P(D n |M R ) is the likelihood function, and P(M R ) is the prior distribution. This incremental learning method enables the correspondence graph to adapt to the feature drift caused by equipment aging and changes in the operating environment, maintaining long-term effectiveness.

[0092] By constructing an acoustic feature spectrum, generating a sound field propagation path model, and establishing a correspondence graph, step S2 provides comprehensive technical support for subsequent precise defect localization. These three sub-steps cooperate with each other to form a complete acoustic visualization analysis framework, which can effectively solve problems such as signal attenuation and multipath interference in the localization of internal defects of GIS equipment, and achieve high-precision identification and localization of defects in hidden parts.

[0093] Preferably, step S2 improves the defect feature recognition ability and anti-interference ability by constructing a multi-dimensional acoustic feature spectrum; the sound field propagation model considering multi-phase media and multipath effects overcomes the limitations of traditional sound source localization methods in complex structures; the correspondence graph with bidirectional mapping enhances the reliability and robustness of the localization results; the synergistic effect of the three forms an adaptive and high-precision acoustic analysis framework, which is applicable to the defect diagnosis of various types of GIS equipment.

[0094] S3: Combine the sound field propagation path model with the correspondence graph, and determine the originating component and specific location of the abnormal acoustic feature spectrum through an acoustic backtracking localization algorithm.

[0095] Specifically, the acoustic backtracking localization algorithm includes a multi-path sound source localization method based on Bayesian optimization.

[0096] Specifically, the steps of the acoustic backtracking localization algorithm include:

[0097] Receive the acoustical feature spectrum collected in real time and determine the abnormal acoustical feature spectrum;

[0098] Construct a Bayesian probability graph model based on the sound field propagation path model, perform Markov chain Monte Carlo sampling using the multi-path sound source position hypothesis space. If the posterior probability value of a certain position point is greater than the posterior probability values of the surrounding areas, then record this position point as a candidate defect source point, and perform iterative calculations until the posterior probability distribution converges, and determine the position point corresponding to the maximum posterior probability as the precise defect position.

[0099] Specifically, the steps of Markov chain Monte Carlo sampling include:

[0100] Set the prior distribution of the sound source position, and calculate the likelihood function of each assumed position based on the sound field propagation path model;

[0101] Iteratively update the sampling points through an acceptance-rejection criterion, calculate the state transition ratio between the new sampling point and the previous sampling point. If the state transition ratio is greater than 1, directly accept the new sampling point. If the state transition ratio is less than 1, decide whether to accept the new sampling point according to the value of the state transition ratio as the probability. Otherwise, reject the new sampling point and return to the previous sampling point. Iterate until the sampling sequence is stable, and finally generate a three-dimensional display graph to show the location and degree of the defect.

[0102] Furthermore, compared with the traditional single positioning method based on time difference or energy attenuation, the multi-path positioning method optimized by Bayesian can consider the influence of multiple acoustic propagation paths simultaneously and has higher positioning accuracy in complex structural environments.

[0103] The steps of the acoustic backtracking positioning algorithm are as follows: First, receive the acoustical feature spectrum collected in real time and determine the abnormal acoustical feature spectrum; then, construct a Bayesian probability graph model based on the sound field propagation path model, and use the multi-path sound source position hypothesis space for Markov chain Monte Carlo sampling to determine the exact position of the defect.

[0104] When receiving the acoustical feature spectrum collected in real time and determining the abnormal acoustical feature spectrum, the system continuously monitors the acoustic signals of the GIS device, extracts the time-domain, frequency-domain, and time-frequency domain features to form a real-time acoustical feature spectrum. Compare this feature spectrum with the acoustical feature spectrum database constructed in step S2, and calculate the abnormality quantification index. If the abnormality quantification index exceeds the preset threshold, then determine that this acoustical feature spectrum is an abnormal acoustical feature spectrum and trigger the subsequent positioning process.

[0105] In practical applications, the system will continuously monitor the acoustic signals collected by multiple sensors and calculate various feature indexes through a fast feature extraction algorithm. To reduce the consumption of computing resources, a two-level threshold strategy is adopted: the first level is fast detection, using time-domain features with less computational complexity (such as kurtosis factor, root mean square value, etc.) for preliminary screening; the second level is precise analysis, and comprehensive feature extraction and abnormality determination are performed on the signals that pass the preliminary screening. This hierarchical processing strategy can ensure both real-time performance and accuracy. When it is determined that there is an abnormal acoustical feature spectrum, the system will save the original signals in a time window containing the abnormal features to provide a data basis for subsequent precise positioning.

[0106] When constructing the Bayesian probability graph model based on the sound field propagation path model, the present invention discretizes the internal space of the GIS device into a three-dimensional grid, and each grid point is used as a potential defect source position. Based on the sound field propagation path model generated in step S2, construct the acoustic propagation relationship from each potential source point to each sensor to form a Bayesian probability graph model. This model can describe the probability distribution of each potential source point becoming the real defect source under the condition of known observed signals, laying a foundation for subsequent sampling optimization.

[0107] Specifically, the Bayesian probabilistic graphical model adopts a directed graph structure, and the nodes include potential sound source positions, sensor observations, and acoustic propagation parameters. Among them, the sound source position is used as a hidden variable, the sensor observation is used as an observable variable, and the acoustic propagation parameter is used as a parameter of the conditional probability. Due to the complex internal structure of the GIS device, sound waves may propagate from the source to the sensor along multiple paths. Therefore, this model particularly considers the multi-path effect and incorporates the contributions of each path into the conditional probability calculation. In addition, the model also considers measurement noise and model uncertainty, and enhances the robustness of the model by representing with probability distributions rather than deterministic functions.

[0108] When using the multi-path sound source position hypothesis space for Markov chain Monte Carlo sampling, the present invention uses the Markov chain Monte Carlo method to sample in the potential source point space and estimate the posterior probability distribution of the defect source position. Specifically, if the posterior probability value of a certain position point is greater than the posterior probability values of the surrounding areas, then this position point is recorded as a candidate defect source point, and iterative calculation is performed until the posterior probability distribution converges, and the position point corresponding to the maximum posterior probability is determined as the exact defect position.

[0109] The specific steps of the Markov chain Monte Carlo sampling include: first, setting the prior distribution of the sound source position, and calculating the likelihood function of each assumed position based on the sound field propagation path model; then, iteratively updating the sampling points through the acceptance-rejection criterion, and finally generating a three-dimensional display map to display the defect position and degree.

[0110] When setting the prior distribution of the sound source position, the present invention comprehensively considers the structural characteristics of the GIS device and historical fault data. For areas prone to defects (such as around insulators, conductor joints, etc.), a higher prior probability is given; for areas with better structural stability, a lower prior probability is given. This prior setting based on domain knowledge improves the sampling efficiency and positioning accuracy.

[0111] In specific implementation, the setting of the prior distribution adopts a hierarchical strategy: first, dividing the area according to the basic structure of the GIS device, such as the circuit breaker chamber, bus chamber, SF6 gas chamber, etc.; then, combining the device type and historical fault statistics data to set the basic prior probability for each area; finally, considering factors such as the operation years of the device, environmental conditions, and recent maintenance records, adjusting the basic prior probability. This multi-level prior setting method can make full use of domain knowledge and historical experience to improve the effectiveness of positioning.

[0112] When calculating the likelihood function for each hypothesized location based on the sound field propagation path model, it is necessary to evaluate the probability of observing the current abnormal acoustic feature spectrum under the condition that a certain point is assumed to be the sound source. Here, the sound field propagation path model constructed in step S2 is used, considering the propagation characteristics of sound waves inside the GIS device, including factors such as multi-path propagation, reflection, and attenuation, to calculate the theoretical acoustic feature spectrum from the hypothesized sound source point to each sensor, and compare it with the actually observed abnormal acoustic feature spectrum to obtain the likelihood value.

[0113] The calculation of the likelihood function needs to consider the observation data of multiple sensors simultaneously. For each hypothesized sound source location, the system calculates the theoretical signal characteristics after the sound source signal at this location propagates to each sensor based on the sound field propagation path model. Considering the complex structure inside the GIS device, sound waves may reach the same sensor through multiple paths. Therefore, when calculating the theoretical signal characteristics, it is necessary to comprehensively consider the contributions of each propagation path. Compare the theoretical signal characteristics with the actually observed characteristics, calculate the similarity or error, and convert it into a likelihood value. To improve the calculation efficiency, approximate calculation methods can be adopted, such as selecting the main propagation path for calculation, or using technologies such as pre-calculated sound propagation response tables.

[0114] In the process of iteratively updating the sampling points through the acceptance-rejection criterion, the present invention uses the Metropolis-Hastings algorithm to implement Markov chain sampling. This algorithm determines whether to accept a new sampling point by calculating the state transition ratio between the new sampling point and the previous sampling point. Specifically, if the state transition ratio is greater than 1, the new sampling point is directly accepted; if the state transition ratio is less than 1, it is decided whether to accept the new sampling point according to the value of the state transition ratio as the probability; otherwise, the new sampling point is rejected and the previous sampling point is returned. In this way, the iteration is executed until the sampling sequence is stable.

[0115] During the sampling process, the system will dynamically adjust the sampling step size and direction to improve the sampling efficiency and convergence speed. In the early stage, a larger sampling step size is adopted to quickly explore the possible solution space; as the iteration progresses, the sampling step size is gradually reduced to finely search the high-probability region. To avoid falling into a local optimal solution, a simulated annealing strategy can also be introduced to allow accepting sampling points with reduced posterior probability with a certain probability, enhancing the global search ability. When the change amplitude of the posterior probability distribution is less than the preset threshold after continuous multiple iterations, it is considered that the sampling sequence has tended to be stable and the iteration is stopped.

[0116] After determining the exact location of the defect, the system will generate a three-dimensional display map to visually display the location and severity of the defect. This display map uses the three-dimensional model of the GIS device as the background, and represents the estimated defect location with marked points of different colors. The depth of the color represents the severity of the defect, providing intuitive fault information for the operation and maintenance personnel.

[0117] The three-dimensional display graph adopts a heat map form to visually display the fault probability distribution of each region. High-fault-probability regions are represented in red, and low-probability regions are represented in blue, forming a gradient color spectrum from blue to red. The system also marks the most likely defect positions and corresponding defect types on the display graph and provides a confidence score. To enhance the visualization effect, the system supports interactive operations such as multi-angle rotation, zooming, and sectioning, enabling maintenance personnel to observe the defect positions from different angles. In addition, the system also provides a historical comparison function to display the historical detection results of the equipment and help analyze the defect development trend.

[0118] Through the above acoustic backtracking localization algorithm, the present invention can make full use of the information obtained in steps S1 and S2 to achieve precise localization of internal defects of GIS equipment. Compared with traditional methods, the localization algorithm of the present invention has the following advantages:

[0119] First, by adopting a multi-path sound source localization method based on Bayesian optimization, it can effectively handle the acoustic propagation problem under the complex structure inside GIS equipment, overcoming the limitations of traditional sound source localization methods in multi-path environments. Traditional methods usually assume that sound waves propagate along a straight line, ignoring phenomena such as reflection and refraction, resulting in a decrease in localization accuracy in complex structure environments. However, the present invention fully considers the multi-path effect and synthesizes the contributions of each path through a Bayesian probability model, improving the localization accuracy.

[0120] Second, the Markov chain Monte Carlo sampling method can effectively handle the probability estimation problem in high-dimensional spaces, avoiding the computational complexity of exhaustive search and having high computational efficiency. This enables the present system to achieve near-real-time defect localization with limited computational resources, meeting the requirements of engineering practice.

[0121] Third, the present invention integrates prior knowledge with measured data. By introducing the equipment structure and historical fault information through the prior distribution and introducing real-time observation data through the likelihood function, it realizes the organic combination of knowledge-driven and data-driven, improving the accuracy and reliability of localization.

[0122] Finally, the three-dimensional visualization interface provided by the present invention visually displays the defect positions and severity levels, facilitating maintenance personnel to quickly identify and handle faults, and improving the efficiency and pertinence of equipment maintenance.

[0123] In practical applications, the average positioning accuracy of this positioning algorithm can reach within 5 cm, which is about 40% higher than that of traditional methods. For concealed parts inside GIS equipment (such as inside elbows, behind insulators, etc.), the positioning accuracy improvement is more obvious, increasing from 15 - 20 cm of traditional methods to within 8 cm. In terms of the system response time, the time from anomaly detection to the completion of positioning is usually controlled within 2 minutes, meeting the requirements of on-site rapid diagnosis. Experiments show that this algorithm has good positioning capabilities for various common defects (such as partial discharge, mechanical looseness, foreign object contact, etc.), especially has higher detection sensitivity for early defects with weak acoustic signals, providing strong support for preventive maintenance.

[0124] Based on the defect position determined by the acoustic backtracking positioning algorithm, combined with the corresponding relationship map between equipment components and acoustic feature spectra established in step S2, the specific component type with defects can be further identified, providing precise guidance for subsequent maintenance work. The combination of this precise positioning and component identification greatly reduces the blindness of equipment disassembly inspection, reduces the mis-dismantling rate and maintenance costs, and improves the equipment management level.

[0125] In summary, the precise defect positioning method for GIS equipment based on acoustic visualization of the present invention realizes the precise positioning of defects in concealed parts inside GIS equipment through three steps: vibration acoustic signal acquisition and separation, acoustic feature spectrum and propagation model construction, and defect precise positioning. Compared with traditional technologies, this method has higher positioning accuracy, a wider scope of application, and a better user experience, providing a new solution for the condition monitoring and fault diagnosis of GIS equipment.

[0126] Embodiment 2 is an embodiment of the present invention, providing a precise defect positioning system for GIS equipment based on acoustic visualization, including:

[0127] A signal separation module, used to collect the vibration acoustic mixed signal during the operation of the equipment, and separate the vibration acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis;

[0128] A feature modeling module, used to construct an acoustic feature spectrum for the pure acoustic signal, generate a sound field propagation path model, and establish a corresponding relationship map between equipment components and acoustic feature spectra at the same time;

[0129] A positioning and tracking module, used to combine the sound field propagation path model and the corresponding relationship map, and determine the originating component and specific position of the abnormal acoustic feature spectrum through the acoustic backtracking positioning algorithm.

[0130] Embodiment 3, referring to Figure 2, which is an embodiment of the present invention and is different from the previous embodiment in that: if the function is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or a part of this technical solution can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the steps of the methods described in various embodiments of the present invention. The aforementioned storage medium includes: various media such as USB flash drives, mobile hard disks, read-only memories (ROM, Read-Only Memory), random access memories (RAM, Random Access Memory), magnetic disks, or optical discs that can store program codes.

[0131] The logic and / or steps represented in the flowchart or described in other ways herein, for example, can be considered as a definite sequence list of executable instructions for implementing logical functions, and can be specifically implemented in any computer-readable medium for use by an instruction execution system, apparatus, or device (such as a computer-based system, a system including a processor, or other systems that can fetch instructions from the instruction execution system, apparatus, or device and execute the instructions), or used in combination with these instruction execution systems, apparatus, or devices. For the purposes of this specification, a "computer-readable medium" can be any device that can contain, store, communicate, propagate, or transport a program for use by or in connection with an instruction execution system, apparatus, or device.

[0132] More specific examples (non-exhaustive list) of computer-readable media include the following: an electrical connection part (electronic device) having one or more wirings, a portable computer disk cartridge (magnetic device), random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber devices, and portable compact disc read-only memory (CDROM). Additionally, a computer-readable medium can even be paper or other suitable media on which the program can be printed, because the program can be obtained electronically, for example, by optically scanning the paper or other media, then editing, interpreting, or otherwise processing it as appropriate, and then storing it in a computer memory.

[0133] It should be understood that each part of the present invention can be implemented by hardware, software, firmware or a combination thereof. In the above embodiments, multiple steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented by hardware, as in another embodiment, any one of the following techniques known in the art or a combination thereof can be used: discrete logic circuits having logic gate circuits for implementing logic functions on data signals, application specific integrated circuits having appropriate combinational logic gate circuits, programmable gate arrays (PGAs), field programmable gate arrays (FPGAs), etc.

[0134] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit them. Although the present invention has been described in detail with reference to the preferred embodiments, those of ordinary skill in the art should understand that the technical solutions of the present invention can be modified or equivalently replaced without departing from the spirit and scope of the technical solutions of the present invention, and they should all be covered by the scope of the claims of the present invention.

Claims

1. An accurate positioning method for GIS equipment defects based on acoustic visualization, characterized in that, Including: Collecting the vibration-acoustic mixed signal during the operation of the acquisition device, and separating the vibration-acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis; The independent component analysis includes a blind source separation algorithm based on minimizing mutual information; Constructing an acoustic feature spectrum for the pure acoustic signal, generating a sound field propagation path model, and establishing a corresponding relationship map between the equipment components and the acoustic feature spectrum; the acoustic feature spectrum includes abnormality quantification indexes of three types of features in the time domain, frequency domain, and time-frequency domain; Combining the sound field propagation path model and the corresponding relationship map, and determining the originating component and specific location of the abnormal acoustic feature spectrum through an acoustic backtracking positioning algorithm; the acoustic backtracking positioning algorithm includes a multi-path sound source positioning method based on Bayesian optimization.

2. The method for precise positioning of GIS device defects based on acoustic visualization according to claim 1, characterized in that: The steps of separating the vibration-acoustic mixed signal by the independent component analysis include: Constructing a multi-channel blind source separation model based on joint approximate diagonalization; Calculating the mutual information measure of the vibration-acoustic mixed signal. If the mutual information measure is less than the result of the previous iteration, continue to iteratively optimize the blind source separation model. If the mutual information measure no longer decreases or reaches the maximum number of iterations, output the current separation matrix, and use the separation matrix to decompose the vibration-acoustic mixed signal into the pure acoustic signal and the vibration signal.

3. The method for precise positioning of GIS equipment defects based on acoustic visualization according to claim 2, characterized in that: The steps of constructing the acoustic feature spectrum include: Extracting time-domain features, frequency-domain features, and time-frequency domain features from the pure acoustic signal, where the time-domain features include kurtosis factors, the frequency-domain features include band energy distributions, and the time-frequency domain features include wavelet energy envelopes; Based on the time-domain features, the frequency-domain features, and the time-frequency domain features, constructing the abnormality quantification indexes of the acoustic feature spectrum.

4. The method for accurately positioning GIS equipment defects based on acoustic visualization according to claim 3, characterized in that: The steps of generating the sound field propagation path model include: Establishing a geometric model of the internal structure of the GIS device; Based on the acoustic wave propagation theory and finite element analysis, calculating the propagation characteristics of acoustic waves in the internal structure of the GIS device. If the internal structure of the GIS device is a multi-phase medium structure, apply the acoustic wave reflection and transmission laws at the medium interface to calculate the acoustic wave propagation parameters. If the internal structure of the GIS device is a single medium structure, directly calculate the acoustic wave propagation parameters, and establish a sound field propagation path model including the sound attenuation coefficient, reflection coefficient, and transmission delay according to the calculation results.

5. The method for accurately positioning GIS device defects based on acoustic visualization according to claim 4, wherein: The steps of establishing the corresponding relationship map between the equipment components and the acoustic feature spectrum include: Obtaining an acoustic feature spectrum database of typical components of the GIS device under normal operation and various fault states; Establishing a mapping relationship matrix between components and features, and establishing a corresponding relationship between the physical structure position of the GIS device and the feature vectors of the acoustic feature spectrum.

6. The method for accurately positioning GIS device defects based on acoustic visualization according to claim 5, characterized in that: The steps of the acoustic backtracking positioning algorithm include: Receiving the acoustical feature spectrum collected in real time and determining the abnormal acoustical feature spectrum; Construct a Bayesian probability graph model based on the sound field propagation path model, and use the multi-path sound source position hypothesis space for Markov chain Monte Carlo sampling. If the posterior probability value of a certain position point is greater than that of the surrounding area, record this position point as a candidate defect source point, and perform iterative calculation until the posterior probability distribution converges, and determine the position point corresponding to the maximum posterior probability as the precise defect position.

7. The method for accurately positioning defects of GIS equipment based on acoustic visualization according to claim 6, characterized in that: The steps of the Markov chain Monte Carlo sampling include: Set the prior distribution of the sound source position, and calculate the likelihood function of each assumed position based on the sound field propagation path model; Iteratively update the sampling points through the acceptance-rejection criterion, calculate the state transition ratio between the new sampling point and the previous sampling point. If the state transition ratio is greater than 1, directly accept the new sampling point. If the state transition ratio is less than 1, determine whether to accept the new sampling point according to the value of the state transition ratio as the probability. Otherwise, reject the new sampling point and return to the previous sampling point, and iterate until the sampling sequence is stable. Finally, generate a three-dimensional display graph to display the defect position and degree.

8. An accurate positioning system for GIS equipment defects based on acoustic visualization, based on the method for accurately positioning GIS equipment defects based on acoustic visualization according to any one of claims 1 to 7, characterized in that: Including, A signal separation module, which is used to collect the vibration-acoustic mixed signal under the operating state of the device, and separate the vibration-acoustic mixed signal into a pure acoustic signal and a vibration signal through independent component analysis; A feature modeling module, which is used to construct an acoustic feature spectrum for the pure acoustic signal, generate a sound field propagation path model, and establish a corresponding relationship graph between the device components and the acoustic feature spectrum; A positioning and tracking module, which is used to combine the sound field propagation path model and the corresponding relationship graph, and determine the originating component and specific position of the abnormal acoustic feature spectrum through an acoustic backtracking positioning algorithm.

9. A computer device, comprising a memory and a processor, the memory storing a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for precise defect location of GIS equipment based on acoustic visualization according to 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 the processor, it implements the steps of the method for precise defect location of GIS equipment based on acoustic visualization according to any one of claims 1 to 7.

Images