Substation partial discharge positioning and imaging method based on acoustic phased array

Abstract

The invention relates to the field of power equipment insulation monitoring and fault positioning, in particular to a transformer substation partial discharge positioning and imaging method based on an acoustic phased array. The method comprises the following steps: deploying an acoustic phased array sensor array to collect a partial discharge multichannel original acoustic signal, and carrying out noise suppression and signal enhancement preprocessing to obtain array frequency domain signal data; a spatial spectrum is calculated through an adaptive beam forming algorithm, and multiple discharge source candidate spatial positions are separated through peak detection and clustering analysis; a three-dimensional space coordinate is obtained through accurate positioning by using a time difference positioning principle, and energy intensity information is fused to generate a partial discharge three-dimensional space distribution image. The method effectively solves the core problems that in the prior art, the partial discharge positioning dimension is limited, a visual imaging function is lacked, the multi-discharge source separation effect is poor, and the anti-interference capability is insufficient.

Description

Technical Field

[0001] This invention relates to the field of insulation monitoring and fault location of power equipment, and in particular to a method for partial discharge location and imaging in substations based on an acoustic phased array. Background Technology

[0002] The safe and stable operation of power systems is the core guarantee of reliable energy supply. Substations, as key hubs for power transmission and conversion, directly impact the overall operational safety of the system due to the insulation performance of their internal electrical equipment. Partial discharge is a significant indicator of insulation degradation in power equipment. Timely and accurate location of partial discharge sources and understanding of their distribution characteristics provide precise information for equipment maintenance, preventing insulation faults from escalating and causing major power outages. Currently, various partial discharge location technologies have been developed, including ultra-high frequency (UHF) methods, ultrasonic methods, time-of-flight (TOF) methods, and angle measurement methods. Among these, the UHF method is widely used due to its strong anti-interference capabilities and fast propagation speed. For example, existing technology CN108490325B discloses a two-stage substation partial discharge signal location method. This method uses UHF sensors to collect signals, combining preliminary clustering location with compressed sensing for precise location, thus reducing hardware requirements while ensuring location accuracy. However, existing technologies still have many limitations: First, the positioning dimensions are limited, mostly only achieving two-dimensional planar positioning, unable to obtain the three-dimensional spatial coordinates of the discharge source, making it difficult to meet the monitoring needs of multiple devices and multiple levels in complex substation environments; second, they lack visualization imaging capabilities, only outputting single or multiple discrete positioning points, unable to intuitively present the spatial distribution and energy intensity differences of partial discharge, which is not conducive to maintenance personnel quickly judging the scope of fault impact; third, the ability to separate multiple discharge sources is insufficient, when multiple simultaneous partial discharge sources exist in the substation, existing methods are prone to signal confusion, leading to increased positioning deviations; fourth, the anti-interference performance needs to be improved, the complex electromagnetic environment and background noise in the substation can easily interfere with the acquired signals, affecting positioning accuracy; fifth, the multi-dimensional information of the discharge signal is not fully integrated, making it difficult to achieve comprehensive analysis of the characteristics and location of the discharge source. These problems make it difficult for existing technologies to fully meet the actual needs of accurate monitoring, intuitive presentation, and efficient operation and maintenance of partial discharge in substations, urgently requiring a partial discharge monitoring technology that can achieve three-dimensional positioning, visualization imaging, multi-source separation, and strong anti-interference capabilities. Summary of the Invention

[0003] This invention provides a method for partial discharge localization and imaging in substations based on an acoustic phased array, aiming to solve the problems of limited localization dimensions, lack of visualization imaging function, poor separation effect of multiple discharge sources, and insufficient anti-interference capability in the existing technology.

[0004] To achieve the above objectives, the following technical solution is adopted.

[0005] A method for localizing and imaging partial discharge in substations based on an acoustic phased array includes the following steps: deploying an acoustic phased array sensor array in the substation monitoring area to collect multi-channel raw acoustic signals generated by partial discharge, obtaining array time-domain signal data; preprocessing the array time-domain signal data, including noise suppression and signal enhancement, to obtain preprocessed array frequency-domain signal data; calculating the spatial spectrum based on the preprocessed array frequency-domain signal data using an adaptive beamforming algorithm to obtain an initial spatial spectrum distribution containing energy peaks; performing peak detection and cluster analysis on the energy peaks in the initial spatial spectrum distribution to separate multiple candidate spatial locations corresponding to different discharge sources; using the time difference positioning principle and the array time-domain signal data, performing precise positioning calculations for each candidate spatial location to obtain the three-dimensional spatial coordinates of each discharge source; fusing the three-dimensional spatial coordinates of the discharge source and its energy intensity information in the initial spatial spectrum distribution to generate a three-dimensional spatial distribution image of partial discharge in the substation monitoring area.

[0006] Optionally, the step of preprocessing the array time-domain signal data to obtain preprocessed array frequency-domain signal data specifically includes: Wavelet threshold denoising is performed on the array time-domain signal data of each channel to obtain denoised array time-domain signal data. Adaptive blind source separation is then performed on the denoised array time-domain signal data to separate the background noise component and the potential discharge acoustic signal component, resulting in a pre-separated acoustic signal. Bandpass filtering is then applied to the pre-separated acoustic signal to retain the frequency band matching the acoustic emission characteristics of partial discharge, resulting in a filtered acoustic signal. Short-time Fourier transform is then performed on the filtered acoustic signal to convert the signal from the time domain to the time-frequency domain, resulting in a time-frequency domain signal representation. In the time-frequency domain, the Wiener filtering algorithm is applied to further enhance the time-frequency domain signal representation, suppressing residual non-stationary noise, resulting in an enhanced time-frequency domain signal. Frequency domain integration is then performed on the enhanced time-frequency domain signal to extract the energy distribution within a specific frequency band, finally obtaining the pre-processed array frequency-domain signal data.

[0007] Optionally, the step of calculating the spatial spectrum based on the preprocessed array frequency domain signal data to obtain the initial spatial spectrum distribution specifically includes: A precise geometric position model of the acoustic phased array sensor array is established. Based on the geometric position model, the theoretical acoustic wave propagation delay from each scan point to each array element within the monitoring area is calculated to form an array manifold matrix. A minimum variance distortionless response beamforming algorithm is used to invert the covariance matrix of the preprocessed array frequency domain signal data and, combined with the array manifold matrix, calculate the spatial power spectral density of each scan point. A diagonal loading technique is introduced to regularize the covariance matrix to improve the numerical stability of the algorithm under coherent signal environments. The calculated original spatial power spectral density is logarithmically compressed and normalized to form a visualized initial spatial spectrum distribution.

[0008] Optionally, the step of performing peak detection and cluster analysis on the energy peaks in the initial spatial spectrum distribution to separate multiple candidate spatial locations corresponding to different discharge sources specifically includes: A dynamic energy threshold is set in the initial spatial spectrum distribution, and all local maxima points above the dynamic energy threshold are extracted to obtain an initial set of peak points. The three-dimensional spatial coordinates and energy values ​​of each point in the initial set of peak points are calculated to form a three-dimensional spatial energy point cloud. A density-based spatial clustering algorithm is applied to analyze the three-dimensional spatial energy point cloud, and peak points with similar spatial locations and energy values ​​are grouped into the same cluster to obtain multiple candidate peak clusters. For each candidate peak cluster, its energy-weighted centroid is calculated, and the three-dimensional spatial coordinates of the centroid are used as the candidate spatial location of a candidate discharge source represented by the cluster. Invalid candidate peak clusters with a total energy below a set threshold are eliminated, and the final separated candidate spatial locations and their corresponding energy information are output.

[0009] Optionally, the step of using the time difference positioning principle and array time-domain signal data to perform precise positioning calculations for each candidate spatial location to obtain the three-dimensional spatial coordinates of each power source specifically includes: For each candidate spatial location, the time-domain signals of several sensor channels with the strongest energy are selected as reference signals from the denoised array time-domain signal data. The cross-correlation function between the time-domain signals of the remaining sensor channels and the reference signals is calculated. By finding the global maximum point of the cross-correlation function, the time difference between the sound wave propagation to different array elements is accurately estimated, and a high-precision time difference estimate is obtained. Based on the precise geometric position model of the acoustic phased array sensor array and the propagation speed model of sound waves in the complex environment of the substation, a set of time difference positioning equations with the candidate spatial location as the initial solution is established. The iterative least squares algorithm is used to solve the set of time difference positioning equations, and the three-dimensional coordinates of the candidate spatial location are continuously optimized until the residual between the time difference estimate and the theoretical time difference calculated based on the current coordinates is minimized. Finally, the optimized three-dimensional coordinates are determined as the precise positioning result of the power source.

[0010] Optionally, the step of generating a three-dimensional spatial distribution image of partial discharge in the substation monitoring area by fusing the three-dimensional spatial coordinates of the discharge source and its energy intensity information in the initial spatial spectrum distribution specifically includes: A three-dimensional voxel space grid covering the substation monitoring area is constructed. The three-dimensional spatial coordinates of each precisely located discharge source are mapped to the three-dimensional voxel space grid to find its corresponding voxel cell. Based on the energy intensity information corresponding to the discharge source in the initial spatial spectrum distribution, the voxel cell is assigned a corresponding grayscale value or color value. For voxel cells not directly occupied by the discharge source, an inverse distance weighted interpolation algorithm based on distance is used to interpolate and fill the cells according to the values ​​of neighboring discharge source voxels, forming a continuous energy intensity spatial distribution field. Three-dimensional image rendering technology is applied to volumetrically render the energy intensity data in the three-dimensional voxel space grid to generate a three-dimensional spatial distribution image of substation partial discharge containing spatial location and discharge intensity information.

[0011] Optionally, a discharge type-assisted identification step may also be included: Extract the array time-domain signal data segment corresponding to the precisely located discharge source to obtain the target discharge signal segment; perform feature extraction on the target discharge signal segment, calculate its time-domain waveform features, frequency-domain spectral features, and time-frequency joint features to form a discharge acoustic signature feature vector; input the discharge acoustic signature feature vector into a pre-trained discharge mode classifier, which is based on a support vector machine and deep neural network fusion architecture, and outputs the probability distribution of the discharge source belonging to different typical discharge types; associate the discharge type label with the highest probability with the three-dimensional spatial coordinates and energy intensity information of the discharge source, and mark it on the three-dimensional spatial distribution image of the partial discharge to form an enhanced three-dimensional positioning imaging result with discharge type information.

[0012] Optionally, the method also includes a step of dynamically updating and analyzing monitoring data: repeating all the steps in claim 1 at a preset cycle to obtain multiple frames of three-dimensional spatial distribution images of partial discharge at different timestamps; establishing a three-dimensional discharge activity database based on time series, storing the coordinates of the discharge source, energy intensity, and occurrence time of each frame image in chronological order; performing spatiotemporal clustering analysis on the three-dimensional discharge activity database to identify stable discharge hotspots that exist continuously in space or have regular energy growth; calculating the statistical characteristics of each stable discharge hotspot, including the average energy growth rate, spatial location drift rate, and discharge frequency change trend; and predicting the development trend of discharge hotspots based on the statistical characteristics, and providing graded early warning for hotspots that may pose an insulation fault risk.

[0013] Optionally, the steps for establishing the propagation speed model of sound waves in the complex environment of a substation specifically include: Multiple acoustic reference sources with known spatial coordinates are arranged within the substation monitoring area. The acoustic reference sources are controlled to emit specific pulse signals, which are received by an acoustic phased array sensor array to obtain calibration signal data. Based on the calibration signal data and the known coordinates of the acoustic reference sources, the equivalent sound velocity of the sound wave along different propagation paths within the monitoring area is calculated, resulting in a set of discrete path sound velocity values. Using the Kriging spatial interpolation algorithm, the set of discrete path sound velocity values ​​is spatially interpolated to fit a continuously varying spatial distribution field of sound velocity throughout the entire monitoring area. This spatial distribution field of sound velocity is then correlated with environmental parameters collected by temperature and humidity sensors to establish a propagation velocity model describing the dynamic changes of sound velocity with spatial location and environmental parameters. This model is then used to solve the time difference positioning equations.

[0014] Optionally, the steps for deploying an acoustic phased array sensor array include: Based on the three-dimensional structural model and electromagnetic interference distribution map of the substation monitoring area, an initial deployment scheme for the acoustic phased array sensor array is planned to ensure coverage of all monitored devices and avoid areas with strong electromagnetic interference. A flexible network of distributed sensor nodes is adopted, with each node containing a microphone array module, a signal preprocessing module, and a wireless communication module. Each sensor node is installed on an insulating support according to the initial deployment scheme, and the final precise three-dimensional coordinates of each array element within each node are measured and recorded to form a precise geometric position model of the array. During system operation, based on the preliminary positioning results and signal quality feedback, the pointing and attitude of some sensor nodes are adaptively fine-tuned via a pan-tilt unit to optimize the reception sensitivity of discharge signals from specific directions.

[0015] Compared with the prior art, the present invention has the following beneficial effects: This application provides a substation partial discharge localization and imaging method based on acoustic phased array, which effectively solves the core problems of limited dimensionality of partial discharge localization, lack of visualization imaging function, poor separation effect of multiple discharge sources, and insufficient anti-interference capability in existing technologies. This method acquires multi-channel raw acoustic signals by deploying an acoustic phased array sensor array, improves signal quality through noise suppression and signal enhancement preprocessing, calculates the spatial spectrum to obtain the initial energy distribution using an adaptive beamforming algorithm, accurately separates candidate locations of multiple discharge sources through peak detection and cluster analysis, achieves precise three-dimensional spatial coordinate positioning by combining time-difference positioning principles, and finally fuses energy intensity information to generate a three-dimensional spatial distribution image. This not only achieves precise three-dimensional localization of discharge sources but also intuitively presents the discharge distribution pattern through visualization imaging, significantly improving the efficiency of maintenance personnel in judging the fault location and impact range.

[0016] The various technical aspects of this method work together to further expand its application value and technical advantages: In signal preprocessing, multiple techniques such as wavelet threshold denoising, adaptive blind source separation, and Wiener filtering are employed to effectively suppress noise interference in the complex environment of substations, ensuring the purity of the discharge signal; the adaptive beamforming algorithm combined with diagonal loading technology improves numerical stability in coherent signal environments, resulting in more accurate spatial spectrum distribution; the density-based spatial clustering algorithm efficiently separates spatially adjacent and energy-similar discharge sources, solving the signal confusion problem when multiple discharge sources coexist; and the introduction of a sound wave propagation velocity model in time-difference positioning, combined with actual substation environmental parameters to calibrate the sound velocity, further... The method improves the accuracy of 3D coordinate positioning; 3D imaging, through voxel mesh construction and interpolation filling, forms a continuous energy intensity distribution field, making the discharge distribution characteristics readily apparent; the discharge type auxiliary identification function, by extracting discharge acoustic features and combining them with a fusion classifier, achieves the correlation labeling of discharge type with location and intensity, providing a more comprehensive basis for fault cause analysis; the dynamic updating and trend analysis steps of monitoring data can track the spatiotemporal changes of discharge hotspots, enabling fault risk classification and early warning, transforming passive maintenance into proactive prevention and control; the flexibly deployed distributed acoustic phased array sensor array can adaptively adjust its attitude, optimize signal receiving sensitivity, and adapt to the monitoring needs of substations with different structures. These technical features work synergistically to significantly improve the method in terms of positioning accuracy, anti-interference capability, visualization effect, and operational practicality, providing a more comprehensive and efficient solution for substation partial discharge monitoring. Detailed Implementation

[0017] The present invention will now be described in detail with reference to embodiments. It should be noted that, unless otherwise specified, the embodiments and features described in the embodiments of this application can be combined with each other.

[0018] The following detailed description is exemplary and intended to provide further detailed explanation of the invention. Unless otherwise specified, all technical terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. The terminology used in this invention is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention.

[0019] Example 1 This embodiment provides a method for partial discharge localization and imaging in substations based on acoustic phased arrays. Through a complete technical solution encompassing acoustic signal acquisition, multi-dimensional signal processing, precise localization calculation, and three-dimensional visualization imaging, it achieves three-dimensional spatial localization of partial discharges in substations, separation of multiple discharge sources, and intuitive presentation of the discharge status. This method is suitable for monitoring equipment insulation in complex substation environments. The specific implementation process is as follows: The first step is to deploy the acoustic phased array sensor array and acquire signals. Before deployment, a 3D structural model and electromagnetic interference distribution map of the substation monitoring area need to be obtained. The 3D structural model can be obtained through laser scanning technology or Building Information Modeling (BIM) modeling, accurately reproducing the equipment layout, spatial dimensions, and obstacle distribution within the monitoring area. The electromagnetic interference distribution map is obtained by measuring multiple points within the monitoring area using an electromagnetic spectrum analyzer, marking the location and interference frequency band of strong electromagnetic interference sources, providing a basis for sensor placement. Based on the 3D structural model and electromagnetic interference distribution map, an initial placement scheme is planned to ensure that the sensor array can fully cover all monitored equipment such as transformers, switchgear, and cable terminals, while avoiding areas with strong electromagnetic interference such as high-voltage busbars and surge arresters, reducing the impact of electromagnetic interference on acoustic signal acquisition.

[0020] The system employs a flexible, distributed sensor node architecture. Each sensor node integrates a microphone array module, a signal preprocessing module, and a wireless communication module. The microphone array module uses high-sensitivity electret microphones or MEMS microphones, with the number of array elements configured according to positioning accuracy requirements. The array elements are arranged in a uniform circular or linear array to ensure good spatial directivity in both horizontal and vertical directions. The signal preprocessing module incorporates a low-noise amplifier and an anti-aliasing filter, which can initially amplify and filter the acquired raw acoustic signals to suppress circuit noise interference. The wireless communication module supports LoRa, 5G, or WiFi communication protocols, and an adaptation scheme is selected based on the substation's communication environment to ensure the stability and real-time performance of data transmission.

[0021] Each sensor node is mounted on an insulating bracket made of high-strength epoxy resin, which possesses excellent insulation performance and mechanical stability. The mounting height is adjusted according to the height of the monitoring equipment to ensure that the microphone array can effectively receive the acoustic signals generated by the discharge. After installation, a high-precision laser rangefinder is used to measure the final accurate three-dimensional coordinates of each element within each node. The coordinate measurement error is controlled within millimeters. The three-dimensional coordinates of all elements are compiled and archived to form a precise geometric position model of the array. During system operation, the signal quality of each channel is monitored in real time. The reception effect is evaluated by parameters such as signal-to-noise ratio and amplitude. When the signal quality in a certain direction is poor, the pointing and attitude of the corresponding sensor node are adaptively fine-tuned by controlling the pan-tilt unit. The pan-tilt unit's pitch angle adjustment range is -30 degrees to 30 degrees, and the azimuth angle adjustment range is 0 degrees to 360 degrees. This fine-tuning optimizes the sensor array's sensitivity to receiving discharge signals from specific directions.

[0022] After the sensor array is deployed, the signal acquisition process is initiated. The acquisition trigger condition is set to the signal amplitude exceeding a preset threshold. The threshold is determined based on the background noise level of the monitoring area. A reasonable trigger threshold is set through prior environmental noise acquisition and analysis to avoid false triggering. The sampling frequency is selected based on the frequency band characteristics of the partial discharge acoustic signal. The frequency of the acoustic signal generated by partial discharge is typically in the range of tens to hundreds of kHz. To satisfy the Nyquist sampling theorem, the sampling frequency is set to more than twice the highest frequency of the signal to ensure complete preservation of the signal's frequency information. During acquisition, a clock synchronization chip is used to achieve synchronous acquisition of multi-channel signals, with synchronization errors controlled at the microsecond level to avoid the impact of time synchronization deviations on subsequent time difference calculations. The acquired multi-channel raw acoustic signals are stored in time-domain data format. The data for each channel includes a timestamp and corresponding amplitude information, constituting the array's time-domain signal data.

[0023] Next, array time-domain signal data preprocessing is performed to improve signal quality through noise suppression and signal enhancement. First, wavelet thresholding denoising is applied to the array time-domain signal data for each channel. The wavelet basis chosen is either db4 or sym8 wavelets, which possess good temporal localization characteristics and frequency resolution, effectively separating signal from noise. An adaptive thresholding method is used to determine the threshold, dynamically adjusting the threshold based on the signal's local variance. Wavelet coefficients with signal amplitudes greater than the threshold are retained, while coefficients with amplitudes less than the threshold are set to zero or reduced. After processing, the denoised array time-domain signal data is reconstructed through inverse wavelet transform.

[0024] Adaptive blind source separation is performed on the denoised array time-domain signal data. The Fast Independent Component Analysis (FastICA) algorithm or the Joint Approximate Diagonalization (JADE) algorithm is selected. These algorithms do not require prior information about the signal source and transmission channel and can separate independent signal components from the mixed signal. Through blind source separation, the background noise component is separated from the potential discharge acoustic signal component. The background noise component includes equipment operating noise, ambient airflow noise, etc., while the potential discharge acoustic signal component is the target signal, resulting in a preliminarily separated acoustic signal.

[0025] The initially separated acoustic signals are subjected to bandpass filtering, with the filtering band determined based on the acoustic emission characteristics of partial discharge. Different types of partial discharge (such as corona discharge, surface discharge, and internal discharge) produce acoustic signals with different frequency bands. Acoustic signals from different types of discharge are collected through preliminary experiments, their spectral characteristics are analyzed, and the frequency band range covering the main discharge types is determined. The passband of the bandpass filter is set to this range, filtering out low-frequency and high-frequency noise outside the passband to obtain the filtered acoustic signal.

[0026] A short-time Fourier transform (SFT) is performed on the filtered acoustic signal to convert it from the time domain to the time-frequency domain. A Hanning or Hamming window is chosen as the window function, and its length is adjusted according to the time-varying characteristics of the signal to ensure both frequency variation capture and time resolution. The window length is typically set to 256 or 512 points, with an overlap rate of 50% to 75% to reduce spectral leakage caused by window truncation. The SFT then yields the time-frequency domain representation of the signal, i.e., the spectral distribution corresponding to each time segment.

[0027] In the time-frequency domain, the Wiener filtering algorithm is applied to further enhance the time-frequency domain signal representation. The core of Wiener filtering is to adaptively adjust the filter coefficients based on the power spectral density ratio of the signal to the noise. First, the power spectral density of the noise is estimated through the region with low signal energy in the time-frequency domain. Then, the power spectral density ratio of the signal to the noise is calculated to construct the Wiener filter, which filters the time-frequency domain signal, suppresses residual non-stationary noise, and obtains the enhanced time-frequency domain signal.

[0028] The enhanced time-frequency domain signal is integrated in the frequency domain to extract the energy distribution within a specific frequency band. The specific frequency band is the passband range of the bandpass filter. The amplitude of the time-frequency domain signal within this frequency band is integrated and summed to obtain the frequency domain energy value of each channel at different time segments. The frequency domain energy values ​​of all channels are then compiled and summarized to finally obtain the preprocessed array frequency domain signal data.

[0029] Based on the preprocessed array frequency domain signal data, the spatial spectrum is calculated using an adaptive beamforming algorithm to obtain the initial spatial spectrum distribution. First, a precise geometric position model of the acoustic phased array sensor array is established, containing the three-dimensional coordinate information of all array elements, serving as the basis for subsequent calculations. According to this geometric position model, the area to be monitored is divided into a three-dimensional mesh, generating a large number of points to be scanned, each with known three-dimensional coordinates. For each point to be scanned, the theoretical acoustic wave propagation delay from it to each array element is calculated. The propagation delay is initially estimated based on the straight-line distance between the point to be scanned and the array element and the speed of sound propagation in air. This yields the phase difference at which each array element receives the discharge signal at that point. Based on the phase difference, a steering vector is constructed, and the steering vectors of all points to be scanned form the array manifold matrix.

[0030] The spatial power spectral density is calculated using the Minimum Variance Distortionless Response (MVDR) beamforming algorithm. First, based on preprocessed array frequency domain signal data, the covariance matrix between each channel is calculated. The covariance matrix reflects the correlation between signals in each channel. During the calculation, piecewise averaging of the signal is performed to improve the estimation accuracy of the covariance matrix. Then, the covariance matrix is ​​inverted, and the inverse matrix is ​​combined with the array manifold matrix and its conjugate transpose. This inverse matrix is ​​then substituted into the MVDR algorithm formula to calculate the spatial power spectral density of each scanned point. The magnitude of the power spectral density reflects the probability of a discharge source at that point.

[0031] To improve the numerical stability of the algorithm under coherent signal environments, a diagonal loading technique is introduced to regularize the covariance matrix. The diagonal loading amount is adaptively adjusted based on the signal-to-noise ratio (SNR). When the SNR is high, a smaller loading amount is chosen to avoid affecting the algorithm's resolution; when the SNR is low, a larger loading amount is chosen to improve the singularity of the matrix. The loading amount can be determined experimentally or through an adaptive algorithm to ensure that the covariance matrix is ​​invertible and the algorithm's performance is stable.

[0032] The calculated raw spatial power spectral density is logarithmically compressed and normalized to form a visualized initial spatial spectral distribution. Logarithmic compression enhances the identification of weak signals by taking the natural logarithm or common logarithm of the power spectral density amplitude, making the differences between signals in different amplitude ranges more obvious. Normalization maps the logarithmically compressed power spectral density to the range of 0 to 1, facilitating subsequent peak detection and visualization. Finally, an initial spatial spectral distribution containing energy peaks is obtained, and the location corresponding to the energy peaks is the approximate region of the potential discharge source.

[0033] Peak detection and cluster analysis are performed on the energy peaks in the initial spatial spectrum distribution to separate multiple candidate spatial locations corresponding to different discharge sources. First, a dynamic energy threshold is set, which is calculated based on the statistical characteristics of the initial spatial spectrum distribution, such as the signal mean and standard deviation. The threshold is set to the mean plus k times the standard deviation, and the value of k is adjusted according to the noise level of the monitoring environment, usually between 2 and 5. All local maxima points above this dynamic energy threshold are extracted. These maxima points are the potential discharge signal peak points, forming the initial peak point set.

[0034] Calculate the three-dimensional spatial coordinates and energy value of each point in the initial set of peak points. The three-dimensional spatial coordinates are indexed to the three-dimensional coordinates of the point to be scanned according to the position of the peak point in the spatial spectrum distribution. The energy value is the normalized power spectral density value of the point. The coordinates and energy values ​​are stored together to form a three-dimensional spatial energy point cloud.

[0035] A density-based spatial clustering algorithm (DBSCAN) is applied to analyze 3D spatial energy point clouds. This algorithm does not require pre-specifying the number of clusters and can automatically identify clusters of arbitrary shapes. Algorithm parameters include the neighborhood radius and the minimum number of points. The neighborhood radius is determined based on the density of the point cloud to ensure that spatially adjacent peak points are grouped into the same neighborhood. The minimum number of points is set as the minimum number of peak points required to form an effective cluster, typically determined based on experimental or empirical data. The DBSCAN algorithm groups spatially adjacent peak points with similar energy values ​​into the same cluster, with each cluster corresponding to a potential discharge source, resulting in multiple candidate peak clusters.

[0036] For each candidate peak cluster, its energy-weighted centroid is calculated, and the three-dimensional spatial coordinates of this centroid are used as the candidate spatial location of a candidate discharge source represented by that cluster. The calculation of the energy-weighted centroid uses the energy value of each peak point as a weight, and performs a weighted average of the three-dimensional coordinates of all peak points within the cluster. Points with larger weights have a greater influence on the centroid location, ensuring that the candidate spatial location is closer to the actual discharge source location.

[0037] An energy sum threshold is set to eliminate invalid candidate peak clusters whose energy sum is below the threshold. The energy sum threshold is determined based on historical discharge data or experiments and is used to filter out false peak clusters caused by noise or interference, retaining candidate clusters with actual discharge significance. Finally, the separated candidate spatial locations and their corresponding energy information are output.

[0038] By utilizing the time-difference positioning principle and array time-domain signal data, precise positioning calculations are performed for each candidate spatial location to obtain the three-dimensional spatial coordinates of each power source. For each candidate spatial location, the time-domain signals of several sensor channels with the strongest energy are selected from the denoised array time-domain signal data as reference signals. Typically, the top three channels with the strongest energy are chosen to ensure that the reference signals have a high signal-to-noise ratio and signal quality, providing a foundation for accurate time-difference estimation.

[0039] The cross-correlation function between the time-domain signals of the remaining sensor channels and the reference signal is calculated. The cross-correlation function reflects the similarity between the two signals, and the time corresponding to its peak value is the time difference between the two signals. A Fast Fourier Transform (FFT) is used to quickly calculate the cross-correlation function, improving computational efficiency. The global maximum point of the cross-correlation function is found, and the x-coordinate corresponding to this point represents the time difference between the sound wave propagating to the reference channel and the current channel, thus obtaining a high-precision time difference estimate.

[0040] Based on the precise geometric position model of the acoustic phased array sensor array and the propagation speed model of sound waves in the complex environment of a substation, a set of time-difference positioning equations is established. The propagation speed model is a pre-constructed spatial distribution field of sound speed. Based on the approximate coordinates of the point to be located (candidate spatial location), the sound wave propagation speed corresponding to that location is obtained by querying the spatial distribution field of sound speed. Based on the three-dimensional coordinates of the sensor array elements, the coordinates of the point to be located, and the sound speed, a time difference equation is established between each channel and the reference channel. Multiple channels correspond to multiple time difference equations, forming a set of time-difference positioning equations.

[0041] An iterative least squares algorithm is used to solve the time difference positioning equations. Candidate spatial locations are used as initial solutions, and their values ​​are substituted into the equations to calculate the residual for each equation. The residual is the difference between the estimated measured time difference and the calculated theoretical time difference. Based on the magnitude and direction of the residual, the three-dimensional coordinates of the point to be located are adjusted, and the theoretical time difference and residual are recalculated. This iterative process is repeated until the sum of squares of the residuals is less than a preset threshold, at which point the iteration converges. The optimized three-dimensional coordinates at this point represent the precise positioning result of the discharge source.

[0042] By fusing the three-dimensional spatial coordinates of the discharge source and its energy intensity information in the initial spatial spectrum distribution, a three-dimensional spatial distribution image of partial discharge in the substation monitoring area is generated. First, a three-dimensional voxel spatial grid covering the substation monitoring area is constructed. The size of the voxel grid is determined according to the positioning accuracy requirements and the size of the monitoring area. The higher the positioning accuracy requirements, the smaller the voxel size. Typically, the voxel size is set to 1 cm × 1 cm × 1 cm to 5 cm × 5 cm × 5 cm to ensure that the spatial distribution details of the discharge source can be clearly presented.

[0043] The three-dimensional spatial coordinates of each precisely located discharge source are mapped onto a three-dimensional voxel grid. Based on the starting coordinates and voxel size of the voxel grid, the voxel index corresponding to the discharge source coordinates is calculated to locate the voxel cell containing it. According to the energy intensity information corresponding to the discharge source in the initial spatial spectrum distribution, the corresponding voxel cell is assigned a corresponding grayscale value or color value. Energy intensity is positively correlated with grayscale / color value; the higher the energy intensity, the brighter the grayscale value or the closer the color is to red; the lower the energy intensity, the darker the grayscale value or the closer the color is to blue. A unified mapping rule is established to ensure the consistency of the imaging results.

[0044] For voxel cells not directly occupied by the discharge source, an inverse distance-weighted interpolation algorithm is used to fill the gaps based on the values ​​of neighboring discharge source voxels, forming a continuous spatial distribution field of energy intensity. Inverse distance-weighted interpolation uses the distance between the interpolation point and neighboring discharge source voxels as the weight; the closer the distance, the greater the weight. The calculation formula is that the energy value of the interpolation point equals the sum of the energy values ​​of each neighboring discharge source voxel multiplied by its weight, where the weight is the reciprocal or square of the distance. Through interpolation filling, the energy distribution of the entire monitoring area shows a continuous change, more intuitively reflecting the influence range of the discharge.

[0045] By applying 3D image rendering technology, volume rendering is performed on the energy intensity data in a 3D voxel grid to generate a 3D spatial distribution image of partial discharge in a substation, containing information on spatial location and discharge intensity. The volume rendering technology employs either a ray casting algorithm or a texture slicing algorithm. The ray casting algorithm emits a ray from each pixel of the image plane, passing through the voxel grid, sampling the voxel energy values ​​along the ray path, and accumulating the pixel's color value through color mapping and opacity, thus generating a realistic 3D image. The texture slicing algorithm slices the voxel grid along a certain direction, generating a series of 2D texture images, which are then superimposed to form a 3D effect. The rendered 3D image can be interactively viewed through rotation, scaling, and translation, allowing maintenance personnel to observe the spatial location and energy distribution of the discharge source from different angles.

[0046] This method also includes a discharge type auxiliary identification step, providing a more comprehensive basis for fault analysis. The array time-domain signal data segment corresponding to the precisely located discharge source is extracted. Based on the discharge source's location time and the signal's timestamp, a segment containing the complete discharge signal is extracted to obtain the target discharge signal segment. The segment length is determined according to the duration of the discharge signal, ensuring that the rising edge, peak value, and falling edge of the signal are included.

[0047] Feature extraction is performed on the target discharge signal segment, calculating its time-domain waveform features, frequency-domain spectral features, and time-frequency joint features to construct a discharge acoustic signature feature vector. Time-domain waveform features include the signal's peak value, mean, variance, kurtosis, waveform factor, and peak factor, reflecting the signal's time-domain amplitude distribution and morphological characteristics. Frequency-domain spectral features are obtained by performing a Fourier transform on the target discharge signal segment, including peak frequency, center frequency, frequency variance, spectral kurtosis, and spectral entropy, reflecting the signal's frequency distribution characteristics. Time-frequency joint features are obtained through wavelet packet transform or short-time Fourier transform, including wavelet packet energy entropy, peak energy in the time-frequency plane, and time-frequency centroid, comprehensively reflecting the signal's time-domain and frequency-domain variation characteristics. All extracted features are normalized to eliminate the influence of different feature dimensions, constructing a discharge acoustic signature feature vector with unified dimensions.

[0048] The discharge acoustic signature feature vector is input into a pre-trained discharge pattern classifier. This classifier is based on a fusion architecture of Support Vector Machine (SVM) and Deep Neural Network (DNN), which fully leverages the classification advantages of SVM on small sample data and the deep learning capabilities of DNN. The classifier training process is as follows: Sample signals of typical discharge types such as corona discharge, surface discharge, internal discharge, and floating potential discharge are collected, with a sufficient number of samples collected for each type, and the discharge type of the samples is labeled; features are extracted from each sample signal to construct training and testing sets; the SVM model and DNN model are trained separately. For the SVM model, an appropriate kernel function (such as the radial basis function kernel) is selected, and for the DNN model, a multilayer perceptron structure is used, with a reasonable number of hidden layers and neurons; the outputs of the two models are fused using a weighted fusion strategy to obtain the final classification probability distribution, outputting the probability distribution of the discharge source belonging to different typical discharge types.

[0049] The label of the most probable discharge type is associated with the three-dimensional spatial coordinates and energy intensity information of the discharge source, and then marked on the three-dimensional spatial distribution image of partial discharge. The marking method can be to display the type name near the location of the discharge source or to use a specific symbol to form an enhanced three-dimensional positioning imaging result with discharge type information, which helps maintenance personnel to quickly determine the type and severity of the discharge.

[0050] This method also includes steps for dynamic updating and trend analysis of monitoring data to achieve continuous tracking and risk warning of discharge status. All the above steps are repeated at a preset cycle, which is determined according to the discharge development speed and monitoring requirements, and is usually set to 5 to 30 minutes. This yields multiple frames of three-dimensional spatial distribution images of partial discharge at different time stamps. Each frame contains information on the location of the discharge source, energy intensity, and discharge type at the corresponding time point.

[0051] A time-series-based three-dimensional discharge activity database was established. The database stores the coordinates, energy intensity, discharge type, and occurrence time of each image frame in chronological order, using a structured data format to support fast querying and data retrieval. Spatiotemporal clustering analysis was performed on the three-dimensional discharge activity database using the Spatiotemporal Density Clustering Algorithm (ST-DBSCAN). This algorithm introduces a time dimension to the DBSCAN algorithm, considering both spatial distance and time interval. It clusters spatially adjacent and temporally continuous discharge sources into the same stable discharge hotspot, identifying stable discharge hotspots that are spatially persistent or exhibit regular energy growth, while excluding accidental transient discharges.

[0052] The statistical characteristics of each stable discharge hotspot are calculated, including the average energy growth rate, spatial location drift rate, and discharge frequency trend. The average energy growth rate is obtained by calculating the ratio of the change in hotspot energy in adjacent time frames to the time interval, reflecting the trend of discharge intensity. The spatial location drift rate is obtained by calculating the ratio of the change in the hotspot center coordinates in adjacent time frames to the time interval, reflecting the movement trend of the discharge source. The discharge frequency trend is obtained by statistically analyzing the change in the number of times the hotspot appears per unit time, reflecting the frequency of discharge.

[0053] Based on statistical characteristics, the development trend of discharge hotspots is predicted. Time series prediction algorithms (such as ARIMA models and LSTM neural networks) are used to predict features such as average energy growth rate and spatial location drift rate to determine whether the hotspots will continue to strengthen, expand, or move. Based on the prediction results, hotspots that may pose a risk of insulation failure are given graded early warnings: Level 1, Level 2, and Level 3. Level 1 warnings correspond to hotspots with rapid energy growth and high drift rates, indicating a high risk of failure and requiring immediate action. Level 2 warnings correspond to hotspots with slow energy growth and stable locations, requiring timely action. Level 3 warnings correspond to hotspots with stable energy and no significant changes, requiring continuous monitoring. Early warning information is promptly communicated to maintenance personnel, achieving a shift from passive maintenance to proactive prevention.

[0054] The steps for establishing a sound wave propagation speed model in the complex environment of a substation are as follows: Multiple acoustic reference sources with known spatial coordinates are arranged within the substation monitoring area. The acoustic reference sources are selected from sound source devices capable of emitting specific pulse signals. The placement locations are chosen from key locations within the monitoring area, including different environmental areas such as densely populated equipment areas and open areas, to ensure coverage of the main propagation path of the monitoring area. The coordinates of the reference sources are measured by high-precision positioning equipment with a positioning error of less than millimeters.

[0055] The acoustic reference source is controlled to emit specific pulse signals. Narrow pulse signals are selected, with frequencies covering the frequency band of the partial discharge acoustic signal to ensure consistency with its propagation characteristics. The pulse signals emitted by the reference source are received by an acoustic phased array sensor array to obtain calibration signal data, and the timestamps of each reference source signal being received by each sensor channel are recorded.

[0056] Based on calibration signal data and known acoustic reference source coordinates, the equivalent sound velocity of the sound wave along different propagation paths within the monitoring area is calculated. For each reference source and each sensor channel, the propagation distance is calculated based on the reference source coordinates and sensor channel coordinates. The propagation time is obtained by combining the difference between the receiving timestamp and the transmitting timestamp. The equivalent sound velocity is equal to the ratio of the propagation distance to the propagation time, resulting in a set of discrete path sound velocity values.

[0057] The Kriging spatial interpolation algorithm is used to spatially interpolate discrete path sound velocity values, fitting a continuously varying spatial distribution field of sound velocity across the entire monitoring area. Based on regionalized variable theory, the Kriging interpolation algorithm constructs a semi-variogram to describe the spatial correlation of sound velocity. Based on the spatial location and correlation of known path sound velocity values, it predicts the sound velocity value at unknown locations, generating a continuous spatial distribution field that reflects the variation of sound velocity with spatial location.

[0058] The spatial distribution field of sound velocity is correlated with environmental parameters collected by temperature and humidity sensors to form a model. Multiple temperature and humidity sensors are deployed in the monitoring area to collect temperature and humidity data at different locations in real time. A multivariate regression model or machine learning model (such as random forest or neural network) is established for sound velocity and temperature and humidity. The model input is temperature, humidity and spatial location, and the output is the corresponding sound velocity value. This model can describe the dynamic change of sound velocity with spatial location and environmental parameters. It can be used to solve the time difference positioning equations to improve the accuracy of time difference positioning.

[0059] As is known from common technical knowledge, this invention can be implemented through other embodiments that do not depart from its spirit or essential characteristics. Therefore, the disclosed embodiments described above are merely illustrative in all respects and are not the only ones. All modifications within the scope of this invention or its equivalents are included in this invention.

Claims

1. A method for localizing and imaging partial discharge in substations based on an acoustic phased array, characterized in that, Includes the following steps: An acoustic phased array sensor array is deployed in the substation monitoring area to collect multi-channel raw acoustic signals generated by partial discharge and obtain array time-domain signal data. The array time-domain signal data is preprocessed, including noise suppression and signal enhancement, to obtain preprocessed array frequency-domain signal data. Based on the preprocessed array frequency domain signal data, the spatial spectrum is calculated using an adaptive beamforming algorithm to obtain an initial spatial spectrum distribution containing energy peaks; Peak detection and cluster analysis are performed on the energy peaks in the initial spatial spectrum distribution to separate multiple candidate spatial locations corresponding to different discharge sources; using the time difference positioning principle and the array time domain signal data, precise positioning calculations are performed on each candidate spatial location to obtain the three-dimensional spatial coordinates of each discharge source; the three-dimensional spatial coordinates of the discharge source and its energy intensity information in the initial spatial spectrum distribution are fused to generate a three-dimensional spatial distribution image of partial discharge in the substation monitoring area.

2. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, The steps for preprocessing array time-domain signal data to obtain preprocessed array frequency-domain signal data specifically include: Wavelet threshold denoising is performed on the array time-domain signal data of each channel to obtain denoised array time-domain signal data. Adaptive blind source separation is then performed on the denoised array time-domain signal data to separate the background noise component and the potential discharge acoustic signal component, resulting in a pre-separated acoustic signal. Bandpass filtering is then applied to the pre-separated acoustic signal to retain the frequency band matching the acoustic emission characteristics of partial discharge, resulting in a filtered acoustic signal. Short-time Fourier transform is then performed on the filtered acoustic signal to convert the signal from the time domain to the time-frequency domain, resulting in a time-frequency domain signal representation. In the time-frequency domain, the Wiener filtering algorithm is applied to further enhance the time-frequency domain signal representation, suppressing residual non-stationary noise, resulting in an enhanced time-frequency domain signal. Frequency domain integration is then performed on the enhanced time-frequency domain signal to extract the energy distribution within a specific frequency band, finally obtaining the pre-processed array frequency-domain signal data.

3. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, The steps for calculating the spatial spectrum and obtaining the initial spatial spectrum distribution based on the preprocessed array frequency domain signal data using an adaptive beamforming algorithm include: A precise geometric position model of the acoustic phased array sensor array is established. Based on the geometric position model, the theoretical acoustic wave propagation delay from each scan point to each array element within the monitoring area is calculated to form an array manifold matrix. A minimum variance distortionless response beamforming algorithm is used to invert the covariance matrix of the preprocessed array frequency domain signal data and, combined with the array manifold matrix, calculate the spatial power spectral density of each scan point. A diagonal loading technique is introduced to regularize the covariance matrix to improve the numerical stability of the algorithm under coherent signal environments. The calculated original spatial power spectral density is logarithmically compressed and normalized to form a visualized initial spatial spectrum distribution.

4. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, The step of performing peak detection and cluster analysis on the energy peaks in the initial spatial spectrum distribution to separate multiple candidate spatial locations corresponding to different discharge sources specifically includes: A dynamic energy threshold is set in the initial spatial spectrum distribution, and all local maxima points above the dynamic energy threshold are extracted to obtain an initial set of peak points. The three-dimensional spatial coordinates and energy values ​​of each point in the initial set of peak points are calculated to form a three-dimensional spatial energy point cloud. A density-based spatial clustering algorithm is applied to analyze the three-dimensional spatial energy point cloud, and peak points with similar spatial locations and energy values ​​are grouped into the same cluster to obtain multiple candidate peak clusters. For each candidate peak cluster, its energy-weighted centroid is calculated, and the three-dimensional spatial coordinates of the centroid are used as the candidate spatial location of a candidate discharge source represented by the cluster. Invalid candidate peak clusters with a total energy below a set threshold are eliminated, and the final separated candidate spatial locations and their corresponding energy information are output.

5. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, The step of using the time difference positioning principle and array time-domain signal data to accurately locate each candidate spatial position and obtain the three-dimensional spatial coordinates of each power source specifically includes: For each candidate spatial location, the time-domain signals of several sensor channels with the strongest energy are selected as reference signals from the denoised array time-domain signal data. The cross-correlation function between the time-domain signals of the remaining sensor channels and the reference signals is calculated. By finding the global maximum point of the cross-correlation function, the time difference between the sound wave propagation to different array elements is accurately estimated, and a high-precision time difference estimate is obtained. Based on the precise geometric position model of the acoustic phased array sensor array and the propagation speed model of sound waves in the complex environment of the substation, a set of time difference positioning equations with the candidate spatial location as the initial solution is established. The iterative least squares algorithm is used to solve the set of time difference positioning equations, and the three-dimensional coordinates of the candidate spatial location are continuously optimized until the residual between the time difference estimate and the theoretical time difference calculated based on the current coordinates is minimized. Finally, the optimized three-dimensional coordinates are determined as the precise positioning result of the power source.

6. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, The step of generating a three-dimensional spatial distribution image of partial discharge in the substation monitoring area by fusing the three-dimensional spatial coordinates of the discharge source and its energy intensity information in the initial spatial spectrum distribution specifically includes: A three-dimensional voxel space grid covering the substation monitoring area is constructed. The three-dimensional spatial coordinates of each precisely located discharge source are mapped to the three-dimensional voxel space grid to find its corresponding voxel cell. Based on the energy intensity information corresponding to the discharge source in the initial spatial spectrum distribution, the voxel cell is assigned a corresponding grayscale value or color value. For voxel cells not directly occupied by the discharge source, an inverse distance weighted interpolation algorithm based on distance is used to interpolate and fill the cells according to the values ​​of neighboring discharge source voxels, forming a continuous energy intensity spatial distribution field. Three-dimensional image rendering technology is applied to volumetrically render the energy intensity data in the three-dimensional voxel space grid to generate a three-dimensional spatial distribution image of substation partial discharge containing spatial location and discharge intensity information.

7. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, It also includes a discharge type auxiliary identification step: Extract the array time-domain signal data segment corresponding to the precisely located discharge source to obtain the target discharge signal segment; perform feature extraction on the target discharge signal segment, calculate its time-domain waveform features, frequency-domain spectral features, and time-frequency joint features to form a discharge acoustic signature feature vector; input the discharge acoustic signature feature vector into a pre-trained discharge mode classifier, which is based on a support vector machine and deep neural network fusion architecture, and outputs the probability distribution of the discharge source belonging to different typical discharge types; associate the discharge type label with the highest probability with the three-dimensional spatial coordinates and energy intensity information of the discharge source, and mark it on the three-dimensional spatial distribution image of the partial discharge to form an enhanced three-dimensional positioning imaging result with discharge type information.

8. The method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, It also includes a monitoring data dynamic update and trend analysis step: repeating all the steps in claim 1 at a preset cycle to obtain multi-frame three-dimensional spatial distribution images of partial discharge under different timestamps; establishing a three-dimensional discharge activity database based on time series, storing the discharge source coordinates, energy intensity and occurrence time in each frame image in chronological order; performing spatiotemporal clustering analysis on the three-dimensional discharge activity database to identify stable discharge hotspots that exist continuously in space or have regular energy growth; Calculate the statistical characteristics of each stable discharge hotspot, including the average energy growth rate, spatial location drift rate, and discharge frequency variation trend; based on the statistical characteristics, predict the development trend of discharge hotspots, and provide graded early warning for hotspots that may pose an insulation fault risk.

9. A method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 5, characterized in that, The steps for establishing a sound wave propagation velocity model in the complex environment of a substation include: Multiple acoustic reference sources with known spatial coordinates are arranged within the substation monitoring area. The acoustic reference sources are controlled to emit specific pulse signals, which are received by an acoustic phased array sensor array to obtain calibration signal data. Based on the calibration signal data and the known coordinates of the acoustic reference sources, the equivalent sound velocity of the sound wave along different propagation paths within the monitoring area is calculated, resulting in a set of discrete path sound velocity values. Using the Kriging spatial interpolation algorithm, the set of discrete path sound velocity values ​​is spatially interpolated to fit a continuously varying spatial distribution field of sound velocity throughout the entire monitoring area. This spatial distribution field of sound velocity is then correlated with environmental parameters collected by temperature and humidity sensors to establish a propagation velocity model describing the dynamic changes of sound velocity with spatial location and environmental parameters. This model is then used to solve the time difference positioning equations.

10. A method for localizing and imaging partial discharge in a substation based on an acoustic phased array according to claim 1, characterized in that, The steps for deploying an acoustic phased array sensor array specifically include: Based on the three-dimensional structural model and electromagnetic interference distribution map of the substation monitoring area, an initial deployment scheme for the acoustic phased array sensor array is planned to ensure coverage of all monitored devices and avoid areas with strong electromagnetic interference. A flexible network of distributed sensor nodes is adopted, with each node containing a microphone array module, a signal preprocessing module, and a wireless communication module. Each sensor node is installed on an insulating support according to the initial deployment scheme, and the final precise three-dimensional coordinates of each array element within each node are measured and recorded to form a precise geometric position model of the array. During system operation, based on the preliminary positioning results and signal quality feedback, the pointing and attitude of some sensor nodes are adaptively fine-tuned via a pan-tilt unit to optimize the reception sensitivity of discharge signals from specific directions.