Transformer partial discharge positioning method based on partial discharge wave velocity adaptive compensation

Abstract

The application discloses a transformer partial discharge positioning method based on a partial discharge wave speed adaptive compensation. The method comprises the following steps: converting a three-dimensional semantic grid model of a transformer into a three-dimensional semantic voxel matrix; determining a target sound speed tensor matrix corresponding to the three-dimensional semantic voxel matrix; determining a time-of-arrival strategy template of each voxel in the target sound speed tensor matrix according to a semantic label identified by the three-dimensional semantic voxel matrix; generating a plurality of three-dimensional time field matrices according to the respective corresponding time-of-arrival strategy templates based on the target sound speed tensor matrix; and in the case of detecting a partial discharge signal of the transformer, determining a partial discharge positioning result of the transformer based on the plurality of three-dimensional time field matrices and observed time difference data. The application solves the technical problem of low transformer partial discharge positioning accuracy in related technologies due to the neglect of the anisotropic propagation characteristics of sound waves of the transformer winding structure.

Description

Technical Field

[0001] This invention relates to the field of power equipment fault monitoring and diagnosis, and more specifically, to a transformer partial discharge location method based on adaptive compensation for partial discharge velocity. Background Technology

[0002] In transformer partial discharge localization technology, the Time Difference of Arrival (TDoA) method, based on ultra-high frequency (UHF) or acoustic emission (AE) sensor arrays, is commonly used. Its core relies on accurate assumptions about the propagation speed of sound waves inside the transformer. However, existing technologies typically simplify the internal medium of the transformer to a homogeneous isotropic oil body, or simply consider the windings as an equivalent homogeneous mixed medium, assigning a single fixed sound velocity value (e.g., 1400 m / s). This severely ignores the complex anisotropic structure of the transformer windings, which consists of periodically stacked copper conductors, insulating paper, and oil channels. Experimental measurements show that the axial or tangential propagation speed of sound waves along the copper conductors can reach 3800–4000 m / s, while the radial propagation speed perpendicularly penetrating the insulation layer attenuates to 1400–1600 m / s, exhibiting a tortuous path and significant attenuation. Because the methods in related technologies ignore the anisotropic propagation characteristics of acoustic waves in the transformer winding structure, there is a systematic velocity error of 20%–40% in the calculation of the signal path through the winding area, resulting in a positioning result offset of more than 10cm, which makes it difficult to meet the high-precision diagnostic requirements of power equipment and leads to low accuracy in locating partial discharge in transformers.

[0003] There is currently no effective solution to the above problems. Summary of the Invention

[0004] This invention provides a transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity, which at least solves the technical problem in related technologies that the accuracy of transformer partial discharge location is low due to neglecting the anisotropic propagation characteristics of acoustic waves in the transformer winding structure.

[0005] According to one aspect of the present invention, a transformer partial discharge localization method based on adaptive compensation for partial discharge velocity is provided, comprising: converting a three-dimensional semantic mesh model of the transformer into a three-dimensional semantic voxel matrix, wherein the three-dimensional semantic mesh model is composed of multiple sub-meshes, each sub-mesh corresponding one-to-one with multiple physical components inside the transformer, each sub-mesh carrying a semantic label of the corresponding physical component for identifying the device identifier and device attributes of the corresponding physical component, and the value of each voxel in the semantic voxel matrix indicating the semantic label of the corresponding physical component; determining a target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix, wherein the value of each voxel in the target sound velocity tensor matrix indicates the physical sound velocity characteristics of the corresponding physical component; and determining the arrival values ​​of each voxel included in the target sound velocity tensor matrix according to the semantic labels identified by the three-dimensional semantic voxel matrix. A time strategy template is used to determine the arrival time of the partial discharge signal at the corresponding voxel location. Based on the target sound velocity tensor matrix, multiple three-dimensional time field matrices are generated for multiple sensors on the transformer surface according to their respective arrival time strategy templates. Each sensor corresponds one-to-one with a specific three-dimensional time field matrix. The value of each voxel in the three-dimensional time field matrix indicates the shortest physical time required for the partial discharge signal emitted from the corresponding physical component location to reach the corresponding reference sensor. When a partial discharge signal is detected in the transformer, the partial discharge location result of the transformer is determined based on the multiple three-dimensional time field matrices and the observation arrival time difference data corresponding to each of the multiple sensors. The observation arrival time difference data indicates the time difference between the capture of the partial discharge signal by the corresponding sensor and the reference sensor.

[0006] According to another aspect of the present invention, a non-volatile storage medium is also provided, which stores multiple instructions, any one of which is adapted to be loaded by a processor for a transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity.

[0007] According to another aspect of the present invention, an electronic device is also provided, including one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement any one of the transformer partial discharge location methods based on adaptive compensation of partial discharge wave velocity.

[0008] According to another aspect of the present invention, a computer program product is also provided, including a computer program that, when executed by a processor, implements the steps of any one of the transformer partial discharge location methods based on adaptive compensation for partial discharge wave velocity.

[0009] In this embodiment of the invention, the three-dimensional semantic mesh model of the transformer is transformed into a three-dimensional semantic voxel matrix. The three-dimensional semantic mesh model consists of multiple sub-meshes, each corresponding to a physical component within the transformer. Each sub-mesh carries a semantic tag for the corresponding physical component, used to identify the device identifier and device attributes of the corresponding physical component. The value of each voxel in the semantic voxel matrix indicates the semantic tag of the corresponding physical component. A target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix is ​​determined, where the value of each voxel in the target sound velocity tensor matrix indicates the physical sound velocity characteristics of the corresponding physical component. According to the semantic tags identified by the three-dimensional semantic voxel matrix, an arrival time strategy template is determined for each voxel included in the target sound velocity tensor matrix. The arrival time strategy template indicates the strategy for determining the arrival time of the partial discharge signal at the corresponding voxel location. Based on the target sound velocity tensor matrix and according to their respective arrival time strategy templates, multiple three-dimensional time field matrices corresponding to multiple sensors on the transformer surface are generated. The multiple sensors are associated with multiple three-dimensional time field matrices. The three-dimensional time field matrix is ​​one-to-one; the value of each voxel in the three-dimensional time field matrix is ​​used to indicate the shortest physical time required for the partial discharge signal emitted from the corresponding physical component location to reach the corresponding reference sensor; when a partial discharge signal of the transformer is detected, the partial discharge location result of the transformer is determined based on multiple three-dimensional time field matrices and the observation arrival time difference data corresponding to multiple sensors. The observation arrival time difference data is used to indicate the time difference between the corresponding sensor and the reference sensor capturing the partial discharge signal. This achieves the goal of generating a multi-sensor three-dimensional time field matrix by transforming the three-dimensional semantic mesh model into a voxel matrix carrying semantic labels and dynamically matching the corresponding sound velocity features and arrival time calculation strategies for each voxel, thereby generating a multi-sensor three-dimensional time field matrix and performing high-precision partial discharge location based on the measured time difference and the pre-calculated time field. This achieves the technical effect of improving the accuracy of partial discharge signal location inside the transformer, and solves the technical problem of low accuracy of transformer partial discharge location caused by ignoring the anisotropic propagation characteristics of sound waves in the transformer winding structure in related technologies. Attached Figure Description

[0010] The accompanying drawings, which are included to provide a further understanding of the invention and form part of this application, illustrate exemplary embodiments of the invention and, together with their description, serve to explain the invention and do not constitute an undue limitation thereof. In the drawings:

[0011] Figure 1 This is a flowchart of a transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity according to an embodiment of the present invention;

[0012] Figure 2 This is a flowchart of an optional target sound velocity tensor matrix construction according to an embodiment of the present invention;

[0013] Figure 3 This is a flowchart of an optional transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity according to an embodiment of the present invention;

[0014] Figure 4 This is a schematic diagram of a transformer partial discharge location device based on adaptive compensation of partial discharge wave velocity according to an embodiment of the present invention.

[0015] Figure 5 This is a schematic diagram of an electronic device structure according to an embodiment of the present invention. Detailed Implementation

[0016] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0017] It should be noted that the terms "first," "second," etc., in the specification, claims, and accompanying drawings of this invention are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that such data can be interchanged where appropriate so that the embodiments of the invention described herein can be implemented in orders other than those illustrated or described herein. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover a non-exclusive inclusion; for example, a process, method, system, product, or apparatus that comprises a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to such processes, methods, products, or apparatus.

[0018] According to an embodiment of the present invention, a method for locating partial discharge in a transformer based on adaptive compensation of partial discharge wave velocity is provided. It should be noted that the steps shown in the flowchart in the accompanying drawings can be executed in a computer system such as a set of computer-executable instructions. Furthermore, although a logical order is shown in the flowchart, in some cases, the steps shown or described may be executed in a different order than that shown here.

[0019] Figure 1 This is a flowchart of a transformer partial discharge location method based on adaptive compensation for partial discharge velocity according to an embodiment of the present invention, as shown below. Figure 1 As shown, the method includes the following steps:

[0020] Step S102: The three-dimensional semantic mesh model of the transformer is transformed into a three-dimensional semantic voxel matrix. The three-dimensional semantic mesh model consists of multiple sub-mesh, which correspond one-to-one with multiple physical components inside the transformer. Each sub-mesh carries a semantic label for the corresponding physical component, which is used to identify the device identifier and device attributes of the corresponding physical component. The value of each voxel in the semantic voxel matrix is ​​used to indicate the semantic label of the corresponding physical component.

[0021] In this step, the three-dimensional semantic mesh model of the transformer is first transformed into a three-dimensional semantic voxel matrix. This model consists of multiple sub-mesh, each of which corresponds to a physical component inside the transformer and carries the semantic label of that component to clearly identify its equipment identifier and equipment attributes. During the transformation process, each voxel in the semantic voxel matrix is ​​assigned a value, which is used to reflect the semantic label of the physical component corresponding to its spatial location, thereby realizing the discretized semantic mapping of the internal spatial structure of the transformer.

[0022] Optionally, the 3D semantic mesh model can be obtained, but is not limited to, through a reverse modeling method based on laser SLAM scanning and multimodal point cloud registration. Specifically, this includes: using a high-precision 3D laser scanner to perform non-contact point cloud acquisition of the transformer's external casing and key internal structures (such as the core, winding outline, tap changer, etc.); combining transformer nameplate parameters, prior knowledge of typical structures, and finite element model guidance; and reconstructing a 3D geometric model containing key components such as the core, windings, oil tank, and tap changer through point cloud segmentation, semantic recognition, and multi-view registration techniques; further, based on component topology relationships and engineering drawing mapping, assigning semantic tags such as core, windings, tap changer, and transformer oil to each sub-mesh, forming a 3D semantic mesh model carrying physical attribute identifiers. This modeling method does not require disassembling the equipment or rely on original design drawings, making it suitable for old transformers without or with incomplete drawings on-site. It also effectively integrates external scanning data with internal structural experience models, providing a highly adaptable geometric foundation for subsequent voxelization and sound velocity tensor modeling.

[0023] Optionally, through spatial voxelization and discrete semantic mapping of the 3D semantic mesh model, the generated continuous geometric surface model of the transformer is transformed into a discrete 3D matrix (i.e., a 3D semantic voxel matrix) that can be used by a computer to calculate wave velocity fields, and the identity information of the geometric components is solidified into each spatial mesh point. The specific implementation process is as follows:

[0024] S1.1: Input data parsing and coordinate system alignment.

[0025] Input object: Receives a 3D semantic mesh model that has completed internal and external registration. .

[0026] Data format requirements: This 3D semantic mesh model consists of multiple independent sub-meshes, each carrying a unique semantic label. The sub-mesh... Marked as "Iron Core", with the attribute of Total Reflection Obstacle; Sub-mesh Marked as "winding", with the property of anisotropic medium; submesh Marked as "Tap Switch", with attributes of mixed medium or obstacle (depending on subsequent settings, typically an insulating cylinder structure); subgrid Marked as "oil tank", with the attribute being boundary constraint; the remaining space defaults to "transformer oil".

[0027] Pre-set coordinates to ensure the origin of the model coordinate system. Consistent with the world coordinate system established by external laser SLAM scanning to ensure the coordinates of externally installed partial discharge sensors. It can be directly mapped to this model space.

[0028] S1.2: Solving for the bounding box and setting the resolution. Traversing the fuel tank model. The axially aligned bounding box (AABB) of the computational model, with all vertices defined, determines the physical boundary of the computational domain. ,in, This represents the minimum x-coordinate of the bounding box. This represents the maximum x-coordinate of the bounding box. This represents the minimum ordinate of the bounding box. This represents the maximum ordinate of the bounding box. Set the voxelization resolution parameters. ( ), calculate the dimensions of the discretized three-dimensional matrix ,in, .

[0029] S1.3: Construct the initial semantic state matrix Allocate a dimension in memory for Three-dimensional integer matrix Initialize all elements in the matrix to (This represents transformer oil, for example, let the value be 0), which means assuming that the space is filled with oil.

[0030] S1.4: Semantic Injection Based on "Inclusion Detection". Using ray casting or polygon scanline algorithms, the 3D semantic mesh model is mapped into a matrix, resulting in a 3D semantic voxel matrix. Considering potential component overlap (e.g., insulating paper attached to metal), the write priority is set as follows: core > tap changer > winding > oil. The specific execution logic is as follows:

[0031] S1.4.1: Traverse the center point of each voxel in the 3D semantic mesh model Its physical coordinates are: .

[0032] S1.4.2: Determine whether point P is located at... Inside the closed grid of the (iron core). If so, then let (For example, let the value be 1).

[0033] S1.4.3: If not in the iron core, determine whether point P is located within the iron core. Inside the closed grid of (winding). If so, then let (For example, let the value be 2).

[0034] S1.4.4: If P is in Inside, then order (For example, let the value be 3).

[0035] S1.4.5: Repeat the above process until all components of the transformer (or all sub-mesh in the 3D semantic mesh model) have been traversed.

[0036] The output of this step is a three-dimensional semantic voxel matrix carrying semantic information. In this matrix, if If so, it indicates that the point is oil (isotropic wave velocity, low loss); if This indicates that the point is an iron core (an acoustically hard boundary that cannot be penetrated and requires diffraction); if If this indicates that the point is a winding (anisotropic region, requiring special calculation); if This indicates that the tap changer is given a special sound velocity value (usually slightly higher than oil, approximately 2000-2500 m / s equivalent).

[0037] Step S104: Determine the target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix, wherein the value of each voxel in the target sound velocity tensor matrix is ​​used to indicate the physical sound velocity characteristics of the corresponding physical component.

[0038] In this step, determining the target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix means directly mapping the semantic identifier information carried by each voxel in the three-dimensional semantic voxel matrix to the corresponding physical sound velocity feature expression, forming a three-dimensional sound velocity tensor field that corresponds one-to-one with the spatial location. This process assigns each voxel a tensor value that can characterize the acoustic propagation characteristics of its corresponding physical component, thereby realizing the transformation from discrete semantic classification to continuous physical attributes, so that the sound velocity attribute at each location in three-dimensional space can be uniquely determined by its semantic identifier.

[0039] In one alternative embodiment, Figure 2 This is a flowchart of an optional target sound speed tensor matrix construction according to an embodiment of the present invention, such as... Figure 2 As shown, determining the target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix includes:

[0040] Step S202: Based on the device attributes of multiple physical components, determine the initial sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix;

[0041] Step S204: The transition region in the initial sound velocity tensor matrix is ​​fuzzed to obtain the first sound velocity tensor matrix. The fuzzing is used to smooth the sound velocity tensor of the transition region, which is the boundary between the transformer oil and the winding in the transformer.

[0042] Step S206: Based on the target oil speed temperature change factor and the target winding structure aging factor, the first sound velocity tensor matrix is ​​corrected to obtain the target sound velocity tensor matrix. The target oil speed temperature change factor is used to correct the sound velocity deviation caused by oil temperature change, and the target winding structure aging factor is used to correct the sound velocity deviation caused by insulation aging or fastening force change.

[0043] In this embodiment, the transition zone width can be set to 2 to 4 voxels to allow the sound velocity to smoothly transition from the isotropic oil side to the anisotropic winding side, eliminating numerical oscillations caused by hard boundaries. By transforming the three-dimensional semantic mesh model of the transformer's internal structure into a voxel matrix carrying semantic labels, the equipment attributes of each physical component are accurately mapped, and an initial sound velocity tensor matrix is ​​constructed accordingly. Further fuzzy smoothing is applied to the sound velocity values ​​at the interface between the transformer oil and windings, effectively mitigating sound velocity jump distortion caused by abrupt changes in material interfaces. Subsequently, by combining the real-time monitored target oil velocity temperature variation factor and the target winding structure aging factor, the first sound velocity tensor matrix is ​​dynamically corrected, allowing the sound velocity characteristics to adaptively adjust with oil temperature fluctuations and insulation aging or changes in winding fastening force. This constructs a high-precision target sound velocity tensor model that truly reflects the transformer's operating state. This model guides the arrival time strategy matching and three-dimensional time field generation for different semantic regions, significantly improving the physical accuracy of partial discharge signal propagation path modeling. Ultimately, it achieves strong robustness of the positioning results to changes in equipment operating conditions, effectively avoiding positioning drift and error accumulation problems caused by using a static sound velocity model, and achieving high-precision and high-stability partial discharge positioning effects even under complex operating conditions.

[0044] In an optional embodiment, when multiple physical components include transformer oil, tap changer, iron core, and winding, the initial sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix is ​​determined based on the device attributes of the multiple physical components. This includes: determining the initial sound velocity corresponding to the transformer oil based on the oil temperature parameter of the transformer oil; determining the initial sound velocity corresponding to the tap changer as a preset first value; determining the initial sound velocity corresponding to the iron core as a preset second value; determining the predetermined sound velocity tensor of the winding based on the preset empirical sound velocities of the winding in each of the three principal directions, wherein the three principal directions include the axial, tangential, and radial directions of the winding; determining the initial sound velocity of the winding according to the sound wave propagation direction of the winding and the predetermined sound velocity tensor; and arranging the initial sound velocities corresponding to the transformer oil, tap changer, iron core, and winding according to the voxel positions in the three-dimensional semantic voxel matrix to obtain the initial sound velocity tensor matrix.

[0045] In this embodiment, multiple physical components inside the transformer—including transformer oil, tap changer, core, and windings—are mapped one-to-one with voxels in a three-dimensional semantic voxel matrix. An initial sound velocity tensor matrix is ​​constructed based on the specific equipment attributes and physical characteristics of each component. The sound velocity is dynamically calculated based on the actual oil temperature parameters of the transformer oil, ensuring that the sound velocity of the oil medium responds to changes in actual operating temperature. Preset first and second values ​​are set for the tap changer and core, consistent with their material structure and acoustic characteristics, to ensure that their sound velocity values ​​have engineering measurement basis. In particular, considering the anisotropic characteristics of sound waves in the winding structure, preset empirical sound velocities in the three main directions—axial, tangential, and radial—are introduced to construct the winding structure. By using the sound velocity tensor and combining it with the vector projection relationship between the sound wave propagation direction and this tensor, the initial sound velocity at each voxel position of the winding is accurately calculated. Then, based on the spatial position of the voxels, the initial sound velocity values ​​of each component are arranged in an orderly manner to form a complete, spatially corresponding, and physically interpretable initial sound velocity tensor matrix. This significantly improves the accuracy of the sound velocity model in depicting the acoustic characteristics of the complex internal structure of the transformer. It lays a solid and realistic physical foundation for subsequent adaptive correction using oil temperature variation factors and winding aging factors, effectively avoiding the problem of sound velocity assumption distortion caused by ignoring component characteristic differences and winding anisotropy. Ultimately, it achieves a high-precision positioning effect with a more realistic partial discharge signal propagation path and a significantly reduced positioning error.

[0046] Optionally, the anisotropic sound velocity tensor of the winding region is constructed based on the local principal axis directions, which include radial, axial, and tangential directions. The radial sound velocity is 1400 to 1600 m / s, and the axial and tangential sound velocities are 3800 to 4000 m / s, corresponding to the propagation characteristics of the sound wave along the penetration direction of the insulating paperboard and along the direction of the copper conductor. The construction of the three-dimensional sound velocity tensor includes: for each winding voxel, calculating its radial, axial, and tangential vectors relative to the winding central axis, and projecting the empirical sound velocity values ​​into a 3×3 diagonal matrix based on this local coordinate system to achieve propagation direction-dependent velocity weighting.

[0047] Optionally, a semantically based non-uniform anisotropic sound velocity tensor field is constructed to obtain an initial sound velocity tensor matrix. The three-dimensional semantic voxel matrix is ​​then obtained based on the aforementioned embodiment. This transforms abstract semantic IDs into specific physical sound speed attributes (i.e., physical sound speed characteristics). The core principle is that, for windings, instead of assigning a single numerical value, a "velocity ellipsoid" model is constructed; and for tap changers and the core, specific equivalent medium properties are assigned. The specific implementation process is as follows:

[0048] S2.1: Scalar assignment of isotropic media (fundamental physical fields). Traversing the matrix. For regions exhibiting isotropic behavior, a scalar velocity of sound is directly assigned. as follows:

[0049] S2.1.1: Transformer oil area ( ): Assignment (Based on oil temperature parameter T) Make minor adjustments;

[0050] S2.1.2: Tap changer area ( Tap changers typically consist of copper contacts encased in an insulating cylinder, and are considered "semi-transmitting high-impedance dielectrics," thus assigning a value... ;

[0051] S2.1.3: Core area ( Sound waves have extremely high reflectivity at the oil-steel interface (>90%), and are considered "obstacles" in path planning algorithms. .

[0052] S2.2: Solving the local structure vector field of the winding (core: solving the problem of "which side is horizontal and which side is vertical"). Considering that the winding is cylindrical, in the voxel... At this point, sound waves propagate fastest along the copper conductor (tangentially), followed by the height direction (axially), and slowest through the insulating cardboard (radially). Therefore, it is necessary to calculate the local coordinate system for each voxel point. The specific calculation logic is as follows: extract the coordinates of the winding's central axis from the model parameters. For each marked as voxels Calculate its three principal unit vectors: radial vector (penetration direction). : The direction from the axis to the voxel point, where, Axial vector (height direction) Parallel to the Z-axis, where Tangential vector (copper wire direction) The cross product of radial and axial directions, where .

[0053] S2.3: Construction of the anisotropic velocity tensor for the windings. For each winding voxel, construct a 3×3 metric tensor M or velocity matrix D. Define the preset empirical sound velocities for the three principal directions as follows: (Along the copper wire / axial direction, fast); (Along the copper wire / tangentially, quickly); (Penetrating the insulating paper, slowly); the formula for constructing the sound velocity tensor is as follows: Its physical meaning is that when the direction of sound wave propagation... Its effective velocity when passing through this voxel The weighting will be based on the angle between the propagation direction and the three principal axes: That is, if the wave passes horizontally through the winding, the speed automatically decreases to 1500; if it runs along the winding, the speed automatically increases to 3800.

[0054] In one optional embodiment, the transition region in the initial sound velocity tensor matrix is ​​blurred to obtain a first sound velocity tensor matrix, including: converting the sound velocity tensors of voxels belonging to transformer oil in a predetermined neighborhood of the transition region in the initial sound velocity tensor matrix into a diagonal tensor matrix, wherein the diagonal elements of the diagonal tensor matrix are the scalar sound velocities of transformer oil; obtaining the sound velocity tensors of the voxels corresponding to the windings in the first predetermined neighborhood of the transition region in the initial sound velocity tensor matrix; determining the oil attribute proportion weight of the voxels corresponding to transformer oil in the transition region based on the Euclidean distance from the transition region to the pure oil region and the Euclidean distance from the transition region to the region corresponding to the windings; performing linear interpolation on the diagonal tensor matrix and the sound velocity tensors of the voxels corresponding to the windings in the first predetermined neighborhood according to the oil attribute proportion weight to generate a smoothed sound velocity tensor corresponding to the transition region; and replacing the sound velocity tensor of the transition region in the initial sound velocity tensor matrix with the smoothed sound velocity tensor to obtain the first sound velocity tensor matrix.

[0055] In this embodiment, to address the path calculation oscillations and non-physical refraction artifacts caused by abrupt changes in the sound velocity tensor in the transition zone between transformer oil and windings, the sound velocity tensor of transformer oil voxels within a predetermined neighborhood of the transition zone is uniformly converted into a diagonal tensor matrix with scalar sound velocity as the diagonal element. This accurately characterizes the isotropic acoustic properties of the transformer oil. Simultaneously, the anisotropic sound velocity tensor of winding voxels within the first predetermined neighborhood of the transition zone is obtained. Then, based on the Euclidean distance from the transition zone voxels to the pure oil region and the winding region, the oil property weight is dynamically calculated. This weight is then used to linearly interpolate the diagonal tensor of the transformer oil and the anisotropic tensor of the windings, achieving a continuous, gradient physical transition of the sound velocity tensor from isotropic to anisotropic in the oil-winding interface. This effectively eliminates discontinuous jumps in the sound velocity field, thereby significantly improving the stability and physical consistency of sound wave propagation path calculation in subsequent three-dimensional time field matrix construction and arrival time strategy applications, ultimately improving the partial discharge positioning accuracy and system robustness.

[0056] In one optional embodiment, the sound velocity tensor of the transition region in the initial sound velocity tensor matrix is ​​replaced with a smoothed sound velocity tensor to obtain a first sound velocity tensor matrix. This includes: replacing the sound velocity tensor of the transition region in the initial sound velocity tensor matrix with a smoothed sound velocity tensor, marking a first predetermined region as an isolation region, and introducing a preset uncertainty penalty factor for the voxel corresponding to the second predetermined region to obtain the first sound velocity tensor matrix. The first predetermined region is the boundary region between the transformer oil and the iron core, the second predetermined region is the minimum region (such as the minimum directional bounding box, cylindrical envelope, etc.) covering the tap changer at the boundary position with the transformer oil, and the isolation region represents the region that does not support sound wave propagation.

[0057] In this embodiment, after replacing the transition region sound velocity tensor of the transformer oil and winding interface in the initial sound velocity tensor matrix with a smoothed sound velocity tensor, the interface between the transformer oil and the iron core is further marked as an isolation region. This forces the voxel of this region to be identified as an impenetrable region in the sound wave propagation path planning, effectively preventing the sound wave from erroneously penetrating due to the modeling error of the iron core boundary. At the same time, for the minimum region of the tap changer covering the interface with the transformer oil, a preset uncertainty penalty factor is introduced. By artificially increasing the propagation cost of this region during the construction of the three-dimensional time field matrix, the sound wave path is guided to preferentially choose to bypass the high-confidence medium, thereby suppressing the interference of low-energy, high-noise direct path caused by the complex structure and uneven materials of the tap changer. Combined with the semantic tag-driven arrival time strategy template and the synergistic effect of the sound velocity tensor field, a refined physical constraint on the sound wave propagation behavior is achieved, significantly improving the accuracy and robustness of the calculation of the arrival time difference of the partial discharge signal. This effectively avoids the positioning deviation problem caused by the physical blocking characteristics of the structural boundary and the propagation uncertainty of complex devices, achieving the effect of accurately locating the partial discharge source in a complex electromagnetic-acoustic coupling environment.

[0058] Optionally, for the initial sound velocity tensor matrix mentioned above, boundary fuzzing and tolerance processing based on morphological gradients are used to obtain the first sound velocity tensor matrix. Since the geometric model generated in the preceding steps may have ±10... A 30mm spatial error (originating from SLAM drift or inversion parameter estimation bias) means that directly using sharp "hard boundaries" may cause the calculated propagation path to erroneously refract or reflect at the model edges. This embodiment aims to construct a "physical property transition zone" to prevent algorithm crashes due to minute boundary displacements, thereby improving system robustness. The specific implementation process is as follows:

[0059] S3.1: Automatic identification and classification of 3D semantic boundaries. The operation object is the traversed and generated 3D semantic voxel matrix. For any voxel Check the voxel IDs within its 26-neighborhood (3x3x3 range). If at least one voxel in the neighborhood has an ID different from P, then mark P as a "boundary voxel". Based on the contact medium, boundaries are divided into two categories:

[0060] Type I (transmission interface) is: oil-winding ( Oil-tap switch Such boundaries allow sound waves to pass through, primarily affecting the angle of refraction.

[0061] Type II (reflective interface) is: oil-iron core Oil - fuel tank Such boundaries block sound waves, primarily affecting the diffraction path.

[0062] S3.2: "Safety Margin Expansion" (morphological processing) for Type II hard obstacles. For "absolute obstacles" like iron cores, drawing the model too small will cause sound waves to "clippage" (incorrectly calculating straight paths), while drawing the model too large only slightly lengthens the diffraction path (the error is controllable). For those marked as... Perform a three-dimensional morphological "dilation" operation on the region. Set the dilation radius. (Approximately 2-4 voxel widths). Push the physical boundaries of the iron core outwards. During path search, the sound waves are forced to avoid this expanded "no-fly zone" (i.e., isolation zone) to effectively offset the risk of core position shift caused by modeling errors.

[0063] S3.3: "Sound velocity gradient smoothing" (Gaussian blurring) for Type I transmission interfaces. To address computational oscillations caused by model boundary geometric errors when sound waves enter the "anisotropic winding" from the "isotropic oil," this step forcibly unifies the sound velocity field near the interface into tensor form and performs linear interpolation, constructing a physical property buffer. The specific execution steps are as follows:

[0064] S3.3.1: Tensor Quantization Unification of Physical Fields: Before mixing, the data dimensions must be unified, specifically including: on the oil side, unifying the scalar velocity of sound. Transform into a diagonal tensor matrix : On the winding side (Coil), the anisotropic tensor in the initial sound velocity tensor matrix is ​​directly invoked. .

[0065] S3.3.2: Geometric Delineation of the Transition Zone: Defining the width of the transition zone Using a morphological edge extraction algorithm, the semantic matrix is ​​extracted. and interface voxel set .by Expanding outwards and inwards from the center. Determine the smooth computation domain.

[0066] S3.3.3: Hybrid weight calculation based on distance field: For any voxel P in the smooth computation domain, calculate its Euclidean distance to the pure winding core region. Euclidean distance to the core area of ​​pure oil The normalized weighting factor is calculated as follows: (Represents the proportion of oil attributes, i.e., the weight of oil attributes): .

[0067] S3.3.4: Attribute Interpolation Generation: Calculate the final smooth tensor of the transition voxel P as follows: This resulted in a continuous medium field that smoothly transitioned from "anisotropic" to "isotropic", completely eliminating the numerical reflection artifacts caused by hard boundaries.

[0068] S3.4: Further modeling of the tap changer's "probabilistic envelope and path penalty". Considering the extremely complex geometry of the tap changer (making it impossible to accurately reconstruct every contact) and its high probability of blocking or scattering signals, instead of pursuing geometric precision, an "envelope + penalty" strategy is adopted to force signal paths to "avoid unnecessary crossings" at the algorithmic level. The specific execution steps are as follows:

[0069] S3.4.1: Envelope Geometry Voxelization: Read the center axis and maximum rotation radius of the tap changer. Construct an oriented bounding box (OBB) or cylindrical envelope covering all possible mechanical protrusions of the tap changer. Voxels within this envelope are then processed. Forced to be marked as a special state .

[0070] S3.4.2: Define "virtual slowness" and "penalty cost": Unlike the iron core's "absolute obstruction (speed = 0)," the tap changer allows signals to pass through, but at an extremely high cost. Assign an equivalent composite speed of sound to this region. In the subsequent solution of the equations, an uncertainty penalty factor is introduced for this region. Set the cost of regular area movement as follows: The cost of moving the tap changer area is set as follows: .

[0071] S3.4.3: From an engineering perspective of path decision-making, if the partial discharge source is inside the tap changer, or if passing through the tap changer is the only path, the algorithm can still calculate the result (because it's not a dead end). If the signal can "bypass" the tap changer to reach the sensor, even if the detour is slightly longer, the algorithm will prioritize the detour (because the cost of passing through is reduced). (This is artificially amplified). This conforms to the physical detection logic of "low energy scattered signal and high energy direct wave", effectively eliminating low-quality paths that pass through the tap changer.

[0072] In an optional embodiment, before correcting the first sound velocity tensor matrix based on the target oil speed temperature variation factor and the target winding structure aging factor to obtain the target sound velocity tensor matrix, the method further includes: obtaining the target oil speed temperature variation factor and the target winding structure aging factor by: selecting an acoustic propagation path through which the connection between at least two sensors passes through the core region of the winding as a calibration path; using an external acoustic calibration source to excite a standard pulse signal at the starting point of the calibration path, and recording the measured arrival time of each sensor on the calibration path, wherein the measured arrival time refers to the time when the corresponding sensor actually receives the excited standard pulse signal; determining the theoretical arrival time of each sensor on the calibration path based on the target sound velocity tensor matrix; constructing a residual objective function between the measured arrival time and the theoretical arrival time of each sensor on the calibration path, wherein the residual objective function is the sum of squares of the differences between the measured arrival time and the theoretical arrival time; optimizing the initial oil speed temperature variation factor and the initial winding structure aging factor with the minimum function value of the residual objective function as the optimization objective to obtain the target oil speed temperature variation factor and the target winding structure aging factor.

[0073] In this embodiment, the acoustic propagation path through the core region of the transformer winding is selected as the calibration path by connecting at least two sensors. A standard pulse signal is excited at the starting point of the path using an external acoustic calibration source, and the measured arrival time of the signal received by each sensor is collected. The theoretical arrival time is calculated by combining the target sound velocity tensor matrix constructed by the three-dimensional semantic voxel matrix. Then, the sum of squares of the difference between the measured and theoretical arrival times is constructed as the residual objective function. The optimization objective is to minimize this function. The initial oil velocity temperature change factor and the winding structure aging factor are adaptively and iteratively adjusted until they converge to the optimal value. This achieves accurate correction of the sound velocity parameters in the first sound velocity tensor matrix, enabling the model to dynamically respond to changes in acoustic characteristics caused by oil temperature fluctuations and insulation aging. This effectively eliminates the positioning deviation caused by static parameter assumptions and significantly improves the spatiotemporal positioning accuracy and robustness of partial discharge signals in complex anisotropic media.

[0074] Optionally, for the first sound velocity tensor matrix, in-situ calibration of the sound velocity field based on sensor array mutual measurement constraints is used to obtain the target sound velocity tensor matrix. Using sensors installed on the transformer tank wall (whose precise coordinates have been locked via SLAM scanning) as "known anchor points," several "detection sound paths" traversing the transformer's interior are constructed. By comparing the measured time-of-flight (ToF) of these paths with the model-calculated time of flight, the global sound velocity scaling factor for the entire field is corrected in reverse. Specifically, this includes:

[0075] S4.1: Construction of the calibration path and acquisition of observations. Before the transformer is energized or using an external acoustic calibration source (such as an acoustic percussion device), at a specific location (coordinate) on the tank wall. Given a standard acoustic pulse signal, the system automatically selects K receiving sensors from N sensors. Select the location of the sound source. The connection can pass through the sensor in the core area of ​​the winding (i.e., located on the opposite side of the oil tank), eliminating paths that transmit only through pure oil (because pure oil paths are not sensitive to winding parameters). The recorded signal originates from the source. To each sensor The actual arrival time is denoted as the observation set: .

[0076] S4.2: Define the set of parameters to be optimized (inversion variables): Instead of directly modifying the value of each voxel (which would be computationally too expensive and prone to overfitting), two global correction coefficients are introduced as variables to be solved, namely the oil speed temperature variation factor. Used to correct background sound velocity deviation caused by oil temperature changes; and winding structure aging factor. This is used to correct anisotropic tensor deviations caused by insulation aging or changes in fastening force. Based on the above factors, the corrected oil sound velocity tensor is calculated as follows: The corrected winding sound velocity tensor is calculated as follows: .

[0077] S4.3: Construct the residual objective function. Utilize the current parameters. Call the anisotropic equation solver to calculate from the source To each sensor Theoretical flight time .

[0078] Construct a least-squares objective function of the following form to characterize the deviation between the model and the actual results: .

[0079] S4.4: Iterative parameter update based on gradient descent. Since there are only two variables, the Gauss-Newton method or grid search method is used for fast optimization. The iterative process is as follows: Initialization ; Calculate the current residual ; Calculate the gradient And update the parameters: ; when the residual Less than the preset threshold Stop when the corresponding distance error is less than 2cm or when the maximum number of iterations is reached.

[0080] It will eventually converge. Applied to the above embodiments generated The sound velocity tensor matrix of all voxels in the field is updated to obtain the target sound velocity tensor matrix.

[0081] Step S106: According to the semantic tags identified by the three-dimensional semantic voxel matrix, determine the arrival time strategy templates corresponding to each voxel included in the target sound velocity tensor matrix. The arrival time strategy templates are used to indicate the determination strategy of the arrival time of the partial discharge signal at the corresponding voxel position.

[0082] In this step, based on the semantic labels identified by each voxel in the three-dimensional semantic voxel matrix, a corresponding arrival time strategy template is matched for each voxel location. This template explicitly indicates the calculation method for the arrival time of the partial discharge signal at that voxel. The semantic labels are directly associated with specific media types, and different labels correspond to different time calculation rules, thereby ensuring that the strategy for determining the signal propagation time matches the physical properties represented by the voxel, achieving differentiated responses to propagation behavior in different regions.

[0083] In one optional embodiment, the arrival time strategy templates corresponding to each voxel in the target sound velocity tensor matrix are determined according to the semantic tags identified by the three-dimensional semantic voxel matrix. This includes: determining the arrival time strategy template corresponding to the voxel with the semantic tag "transformer oil" in the target sound velocity tensor matrix as a first strategy template, wherein the first strategy template is used to indicate that, based on the diagonal sound velocity tensor of the corresponding voxel, a univariate quadratic equation is solved in six orthogonal neighborhoods using a specified difference scheme to calculate the minimum arrival time of the corresponding voxel; determining the arrival time strategy template corresponding to the voxel in the target sound velocity tensor matrix, ... The arrival time strategy template corresponding to the voxel with semantic label "winding" is the second strategy template. The second strategy template is used to indicate that the anisotropic sound velocity tensor based on voxels searches for the simplex that best matches the main velocity direction in twenty-six neighborhoods, and obtains the minimum arrival time that satisfies the specified equation through matrix transformation and iterative solution. The arrival time strategy template corresponding to the voxel with semantic label "core" in the target sound velocity tensor matrix is ​​the third strategy template. The third strategy template is used to set the arrival time of the corresponding voxel to a specified value to isolate the sound wave propagation.

[0084] In this embodiment, the first strategy template can be understood as a six-neighborhood isotropic fast-moving update template, and the specified difference scheme can be the Godunov difference scheme; the second strategy template can be understood as a twenty-six-neighborhood anisotropic simplex update template, and the specified equation can be the Eikonal equation; the third strategy template can be understood as a skip update template, and the specified value can be set to infinity to effectively isolate the propagation of sound waves. This embodiment models the internal structure of the transformer as a three-dimensional semantic voxel matrix carrying semantic tags, and combines it with the target sound velocity tensor matrix corresponding to the physical component characteristics. Different arrival time calculation strategies are applied to voxels with different semantic tags: for the transformer oil region, based on its diagonal sound velocity tensor characteristics, a univariate quadratic equation is solved in six orthogonal neighborhoods using a specified difference scheme to achieve efficient and accurate calculation of sound wave propagation time in isotropic media; for the winding region, its anisotropic sound velocity tensor characteristics are fully utilized to search for the simplex that best matches the main velocity direction in the twenty-six neighborhoods, and matrix transformation and iteration are used to calculate the time of sound wave propagation in the isotropic medium. The minimum arrival time that satisfies the equation is solved to accurately characterize the directional influence of the winding fiber structure on the propagation path and velocity of sound waves. For the iron core region, the arrival time of the corresponding voxel is directly set to a preset maximum or invalid value to physically isolate the sound wave propagation path and avoid false propagation path interference caused by the high impedance characteristics of the iron core being mistakenly modeled as a waveguide medium. The three strategies work together to make the propagation model of partial discharge signals in complex multi-medium environments more consistent with actual physical behavior, significantly reduce the positioning deviation caused by ignoring structural anisotropy or false propagation paths, and ultimately improve the accuracy and robustness of spatial positioning of partial discharge sources.

[0085] Step S108: Based on the target sound velocity tensor matrix, generate multiple three-dimensional time field matrices corresponding to multiple sensors on the transformer surface according to their respective arrival time strategy templates. The multiple sensors correspond one-to-one with the multiple three-dimensional time field matrices. The value of each voxel in the three-dimensional time field matrix is ​​used to indicate the shortest physical time required for the partial discharge signal emitted from the corresponding physical component location to reach the corresponding reference sensor.

[0086] In this step, based on the target sound velocity tensor matrix, multiple three-dimensional time field matrices are generated for the multiple sensors on the transformer surface according to their respective arrival time strategy templates. Each sensor corresponds one-to-one with one of the multiple three-dimensional time field matrices. This process uses the preset sound velocity tensor matrix as its physical basis, independently performing time propagation calculations for each sensor. Based on the arrival time strategy template corresponding to that sensor, the physical time required for the signal to propagate from any location in space to that sensor is calculated point-by-point, ultimately forming a three-dimensional time distribution field covering the entire computational domain. Because each sensor has a different location, its corresponding time propagation path, refraction and diffraction characteristics, and time accumulation rules are calculated independently, ensuring that each three-dimensional time field matrix accurately reflects the sound wave propagation characteristics sensed by that sensor, achieving a strict correspondence between the sensor and the time field matrix.

[0087] It should be noted that the value stored in each voxel in the three-dimensional time field matrix directly represents the shortest physical propagation time required for a partial discharge signal emitted from the corresponding spatial physical location of that voxel to reach the designated reference sensor. This time value is determined by the anisotropic propagation, diffraction, and refraction processes experienced by the signal in the complex medium inside the transformer. Its calculation is based on the sound velocity tensor distribution and boundary treatment rules of each region in the semantic voxel field. The resulting spatial-temporal distribution field completely maps the physical propagation delay of the signal from any point to the sensor.

[0088] Optionally, anisotropic fast-margin solution based on semantic template adaptive switching is used to generate a three-dimensional time field matrix. By solving the equations, the distance from any point in space to the target sensor is calculated. The shortest "physical flight time" scalar field The core of this step lies in dynamically adjusting the discretization solution template (Stencil) of the partial differential equation using the fixed semantic tag IDs obtained in the aforementioned embodiments. The specific implementation process is as follows:

[0089] S5.1: Discretized definition of the anisotropic equation. Wavefront propagation is defined to follow the anisotropic Eikonal equation as follows: Where T is the arrival time field and D is the final output calibrated velocity tensor matrix (i.e., the target sound speed tensor matrix). It refers to the time gradient. In a conventional FMM (Fast Moving Model), only the six adjacent voxels (up, down, left, right, front, and back) are considered. However, in anisotropic media (winding), the wave propagation direction is not collinear with the gradient direction, and a wider computational template (such as a 26-neighborhood or larger) must be introduced to capture the characteristics of oblique propagation.

[0090] S5.2: Semantic-driven adaptive selection of computational templates (Stencils). In each iteration of wavefront advancement, the semantic label of the voxel P to be updated is examined. .

[0091] The trigger condition for setting mode A (isotropic, high-speed mode) is as follows: or The corresponding algorithm strategy is to adopt the standard 6-neighborhood Godunov difference scheme. The computational cost of this mode is only solving a simple quadratic equation, with a computational complexity of O(1).

[0092] The trigger condition for setting mode B (anisotropic, high-precision mode) is as follows: The corresponding algorithm strategy is to adopt the Wide-Stencil Anisotropic Update format. It searches the 26 neighboring nodes around point P to find the simplex that best matches the principal velocity direction (the principal axis vector calculated in the aforementioned embodiment). This mode requires multiple matrix-vector multiplications and iterative searches, with a computational cost 10-20 times that of mode A, but it only triggers in the winding region.

[0093] Setting mode C (the trigger condition for obstacle mode is...) The corresponding algorithm strategy is to skip updates (set to ∞), forcing the wavefront to detour.

[0094] S5.3: Wavefront propulsion based on narrow band heap to move the target sensor voxel Time set The rest are set as .Will Add it to a min-heap. Proceed in a loop as follows: pop the voxel with the shortest time from the top of the heap. Mark as frozen; iterate through All neighbors ;like If the object is not frozen and is not an obstacle, then the following judgment logic is executed: Based on Given a semantic ID, select either pattern A or pattern B to calculate the new arrival time. .like Then update It updates the position of the data in the heap. The corresponding termination conditions are: when the heap is empty, or when all voxels are frozen.

[0095] S5.4: Generation and storage of three-dimensional time-field maps. For each partial discharge sensor... Repeat the above process to generate N three-dimensional time field matrices. Any point in the matrix The value represents the partial discharge signal emitted from that point, reaching the sensor after refraction, diffraction, and speed change. Precise physical time.

[0096] Step S110: When a partial discharge signal of the transformer is detected, the partial discharge location result of the transformer is determined based on multiple three-dimensional time field matrices and the observation arrival time difference data corresponding to each of the multiple sensors. The observation arrival time difference data is used to indicate the time difference between the corresponding sensor and the reference sensor in capturing the partial discharge signal.

[0097] In this step, upon detecting a partial discharge signal from the transformer, the spatial location of the discharge event is estimated based on multiple pre-generated three-dimensional time field matrices and the observed time difference of arrival data between each sensor and the reference sensor. The three-dimensional time field matrix records the physical propagation time required for a signal emitted from any point in space to reach each sensor, while the observed time difference of arrival data reflects the temporal differences in signal reception between each sensor and the reference sensor during actual detection. By comparing the measured time difference with the pre-calculated time difference for corresponding candidate points in the three-dimensional time field matrix, the spatial location that best matches the measured data can be identified, thus directly determining the coordinates of the partial discharge. The entire process relies entirely on the mathematical correspondence between the time field matrix and the time difference data, without involving any additional physical model corrections or signal feature analysis.

[0098] In one optional embodiment, upon detecting a partial discharge signal from the transformer, the partial discharge location result of the transformer is determined based on multiple three-dimensional time field matrices and the observed arrival time difference data corresponding to each of the multiple sensors. This includes: using any voxel within the non-obstacle voxel range of the three-dimensional semantic voxel matrix as a candidate point, reading the theoretical arrival time difference data of the candidate point corresponding to each of the multiple sensors from the multiple three-dimensional time field matrices; constructing a residual function, where the residual function is the sum of squares of the differences between the observed arrival time difference data and the theoretical arrival time difference data; based on the residual function, traversing all voxels within the non-obstacle voxel range of the three-dimensional semantic voxel matrix, calculating the residual function value corresponding to each voxel; and determining the partial discharge location result based on the voxel with the smallest residual function value among all voxels.

[0099] In this embodiment, the theoretical time difference of arrival (TDOA) data is the numerical value corresponding to each candidate point in multiple three-dimensional time field matrices. By using the non-obstacle objects in the three-dimensional semantic voxel matrix of the transformer as the only valid candidate positioning space, and combining the TDOA data observed by multiple sensors with the theoretical TDOA data corresponding to each candidate voxel, a residual function representing the time difference error in the form of a sum of squares is constructed. All candidate points are systematically traversed within the range of the non-obstacle objects, and their residual function values ​​are calculated. Finally, the voxel with the smallest residual is selected as the partial discharge location, realizing an adaptive positioning mechanism guided by quantified error and starting from the physical model. This process fully integrates the semantic information of the transformer's internal structure and the anisotropic propagation characteristics of sound waves, effectively avoiding positioning drift caused by ignoring material heterogeneity and multipath interference. The positioning result no longer relies on human experience or single-path assumptions, but automatically converges to the physical location that best matches the measured data through least-squares optimization, significantly improving positioning accuracy, robustness, and repeatability. This effectively avoids the problem of low positioning accuracy caused by the lack of an error quantification evaluation mechanism.

[0100] In one optional embodiment, the partial discharge localization result is determined based on the voxel with the smallest residual function value among all voxels, including: taking the voxel with the smallest residual function value among all voxels as the initial discharge localization result; constructing a trilinear interpolation function in a predetermined neighborhood centered on the initial discharge localization result to fit the local residual surface; solving for the minimum point in a predetermined continuous domain of the trilinear interpolation function to obtain the target localization coordinates; and using the target localization coordinates as the partial discharge localization result.

[0101] In this embodiment, based on multiple three-dimensional time field matrices generated from the three-dimensional semantic voxel matrix and the target sound velocity tensor matrix, and the sensor observation arrival time difference data, the residual function values ​​are first calculated by traversing the discrete voxel space, and the voxel corresponding to the minimum value is selected as the initial discharge location result. Subsequently, within a predetermined neighborhood of this initial result, the residual values ​​of neighboring voxels are continuously fitted using a trilinear interpolation function to construct a residual surface model reflecting the local energy minimum trend. Within this continuous domain, its minimum point is accurately solved using mathematical optimization methods, thereby obtaining the target location coordinates that exceed the limitations of voxel discrete resolution, which serve as the final partial discharge location. This method effectively overcomes the problem of limited positioning accuracy caused by voxel modeling, achieving sub-voxel-level high-precision positioning without changing the original mesh structure, significantly improving the spatial positioning accuracy and engineering practicality of transformer partial discharge signals.

[0102] Optionally, the partial discharge location result of the transformer can be obtained by further minimizing the residuals of the physical time map. Combining the TDoA (Time Difference of Arrival) data of the partial discharge signal captured in real time during transformer operation, the spatial coordinates of the partial discharge occurrence can be accurately located in multiple (e.g., N) three-dimensional time field maps generated in the aforementioned embodiment using a global optimization algorithm. The specific implementation process is as follows:

[0103] S6.1: Real-time TDoA extraction of partial discharge signal. The algorithm input is set as: when partial discharge occurs in the transformer, the sensor array... The captured pulse signals are analyzed using conventional signal processing algorithms (such as the cumulative energy method) to extract the absolute times of the signals captured by each sensor. Select the sensor with the highest signal-to-noise ratio as the reference sensor (e.g., ), calculate the arrival time difference of other sensors relative to the reference sensor (i.e., observe the arrival time difference data): .

[0104] S6.2: Constructing a residual objective function based on the spectrum (core algorithm). For any voxel candidate point P in space, if it is a true partial discharge source, then the pre-calculated time difference of the spectrum from P to each sensor should be exactly equal to the measured time difference.

[0105] Construct a spatial fitness function of the following form : ,in, This represents the theoretical time difference of arrival data. The values ​​are read directly from the three-dimensional time field matrix (used to indicate the shortest physical time required for the partial discharge signal emitted from the corresponding physical component location to reach the corresponding reference sensor). This indicates the time when the reference sensor actually detected the partial discharge signal.

[0106] S6.3: Sub-Voxel level search from coarse to fine. This consists of two stages. Stage one involves a global mesh search, traversing all non-obstacle objects defined in the 3D semantic mesh model, and calculating the center of each voxel. Find the voxel coordinates with the smallest residual. (Since it's a table lookup operation, traversing tens of millions of voxels can be completed in milliseconds); sub-voxel interpolation optimization is performed in stage two, due to the voxel resolution... (e.g., 10mm) limits positioning accuracy, requiring continuous domain optimization; in Within a 3×3×3 neighborhood, construct The function is modeled using trilinear interpolation or quadratic surface fitting; the extreme points of the fitted surface are found within a small local range using the Newton-Raphson method or simplex method; the final floating-point coordinates are output. Theoretically, the accuracy can reach the 1mm level.

[0107] S6.4: Location Confidence Assessment. To prevent misleading information, the system automatically calculates the Hessian matrix or curvature of the objective function near the extreme points. The criteria are as follows: if the extreme point is steep → marked as "high confidence" (precise location); if the extreme point is as flat as a "frying pan" → marked as "low confidence" (possible multiple solutions or interference).

[0108] In the 3D model, Centered on the target, a semi-transparent "error probability cloud" is rendered to intuitively and visually display the possible range of positioning deviations.

[0109] Through steps S102 to S110, the three-dimensional semantic mesh model of the transformer is transformed into a three-dimensional semantic voxel matrix carrying semantic labels of physical components. Based on the semantic labels corresponding to each voxel, a target sound velocity tensor matrix reflecting the differences in acoustic properties of different physical components is constructed, making the sound velocity characteristics of each voxel directionally dependent, thereby accurately characterizing the anisotropic propagation characteristics of sound waves in key components such as winding structures. Furthermore, according to the semantic labels, corresponding arrival time strategy templates are matched for each voxel, and the three-dimensional time field matrices corresponding to multiple sensors are dynamically generated in combination with the target sound velocity tensor matrix, realizing high-precision spatial modeling of the shortest propagation time from the partial discharge source to the sensor. After detecting the partial discharge signal, the multiple three-dimensional time field matrices and the actual arrival time difference data of the sensor are combined, and joint positioning calculation is performed through an adaptive wave velocity compensation mechanism, effectively eliminating the problems of propagation path misjudgment and time error accumulation caused by the use of uniform scalar sound velocity in traditional methods, thereby significantly improving the accuracy and robustness of partial discharge positioning. Therefore, it can solve the technical problem of low positioning accuracy caused by ignoring the anisotropic propagation characteristics of sound waves in transformer winding structure in related technologies, and achieve accurate characterization of the non-uniform and direction-sensitive propagation behavior of sound waves in complex heterogeneous structures, realize high-precision and adaptive spatial positioning of partial discharge sources, and improve the reliability of transformer operation status monitoring and fault diagnosis efficiency.

[0110] Power transformers are core hub equipment in the power grid, and their internal insulation condition directly affects the safe and stable operation of the power system. Partial discharge is a major sign of transformer insulation degradation and a primary cause of insulation breakdown. Currently, time difference of arrival (TDoA) based on ultra-high frequency (UHF) or ultrasonic (AE) sensor arrays is the main method for on-site detection and location of partial discharge sources. The basic principle of TDoA location is to construct a nonlinear equation system by measuring the absolute time difference of the partial discharge signal arriving at sensors at different known locations, and then to determine the spatial coordinates of the discharge source. However, the location accuracy of this method is highly dependent on the accurate estimation of the "signal propagation path" and "signal propagation speed".

[0111] Currently, the calculation methods for transformer partial discharge localization mainly fall into three categories: The first related technical solution is an analytical geometric method based on the assumption of a homogeneous medium (the mainstream commercial method). This solution assumes that the inside of the transformer is a homogeneous, isotropic medium space (usually simplified to pure transformer oil), and that the partial discharge signal propagates in a straight line from the source to the sensor at a fixed speed (e.g., 1400 m / s for ultrasound in oil). The algorithm obtains the result by solving the time difference equation. The second related technical solution is a path search method based on simplified geometric obstacle avoidance. To solve the non-line-of-sight (NLOS) propagation problem caused by core obstruction, some improved techniques introduce simple obstacle models. The principle is to establish a simplified three-dimensional model of the transformer (usually containing only a cylindrical core), dividing the space into discrete grids. Using Dijkstra's algorithm or A... The algorithm searches for the shortest broken-line path around the core, thereby correcting the distance parameter in the TDoA equation. In this scheme, the core is treated as an impassable obstacle, and other areas are treated as uniform oil. Related technical scheme three is a fingerprint database matching method based on full-wave simulation (FDTD / FEM). This scheme mainly utilizes the Finite Differential Time Domain Method (FDTD) or the Finite Element Method (FEM) to perform refined electromagnetic / acoustic field simulation of the transformer, simulating signal reflection, refraction, and attenuation in complex structures. A massive "position-waveform" fingerprint database is pre-established, and the location is determined by matching the on-site waveform with the fingerprint database.

[0112] However, the above-mentioned technical solutions have the following problems. Regarding technical solution one, although this type of solution achieves partial discharge localization to a certain extent, it still has the following unresolved technical pain points (i.e., the problems this invention aims to solve) when facing the high-precision localization requirements of transformers with complex structures in the field: The first defect is that it temporarily ignores the "anisotropic" propagation characteristics of the winding region (the core pain point). This type of solution (whether it's the uniform assumption of technical solution one or the simple obstacle avoidance of technical solution two) usually treats the winding as a "solid copper block" or a "uniform mixed medium," assigning it a single equivalent sound velocity. However, a real transformer winding is a complex structure composed of alternating layers of copper wire, insulating paper, and oil channels. Physical experiments show that sound waves propagate mainly through the copper medium (velocity approximately 3800-4000 m / s) along the conductor direction (axial / tangential), while propagation through the insulating layer (radial) requires passing through multiple layers of oil-paper interfaces (velocity attenuates to 1400-1600 m / s and the path is tortuous). The related technologies ignore this velocity tensor characteristic, resulting in a 20%-40% velocity error in the calculation of any signal path passing through the winding. This is the fundamental physical reason why the current positioning accuracy is difficult to break through the 10cm bottleneck. The second drawback is the reliance on ideal models and the lack of a fault-tolerant mechanism for modeling errors (an engineering pain point). Related technical solutions two and three heavily rely on the accuracy of the transformer's geometric model. However, on-site drawings are often lacking, or the models obtained through external scanning (such as the applicant's previous work) have dimensional errors of several centimeters. The algorithms in these related technologies often use "hard boundary" calculations (i.e., either oil or iron). Once there is a slight deviation between the model boundary and the actual physical location (e.g., 20mm), the signal calculation path may abruptly change from "through" to "blocking / diffusing," causing drastic discontinuous jumps in the calculation results and extremely poor system robustness. The third drawback is the lack of "in-situ calibration" capability for on-site physical parameters. The sound velocity parameters used in these related technologies are mostly empirical values ​​(e.g., the oil velocity is 1420m / s from a table). However, in reality, uneven oil temperature distribution inside the transformer and different degrees of insulation aging can cause the actual wave velocity to deviate from the empirical value.

[0113] In summary, the technology lacks a closed-loop mechanism—that is, using the existing sensor array installed on-site as known conditions to reverse-engineer and correct the actual physical parameters of the current equipment. This results in a model that is always fixed and cannot adapt to the individual differences of transformers.

[0114] To address the aforementioned problems, based on the above embodiments and optional embodiments, the present invention proposes an optional implementation method. Figure 3 This is a flowchart of an optional transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity according to an embodiment of the present invention, as follows: Figure 3 As shown, the method includes:

[0115] Step S1: Spatial voxelization and discrete semantic mapping of the 3D semantic mesh model. The generated continuous geometric surface model of the transformer is transformed into a discrete 3D matrix (i.e., a 3D semantic voxel matrix) that can be used by a computer to calculate wave velocity fields, and the "identity information" of the geometric components is fixed into each spatial mesh point. The specific implementation process is the same as in the previous embodiment, and will not be repeated here.

[0116] Step S2: Construct a semantically based non-uniform anisotropic sound velocity tensor field to obtain the initial sound velocity tensor matrix. Based on the three-dimensional semantic voxel matrix output in Step S1... This transforms abstract semantic IDs into specific physical sound speed attributes (i.e., physical sound speed characteristics). The core principle is that, for windings, instead of assigning a single numerical value, a "velocity ellipsoid" model is constructed; and for tap changers and the core, specific equivalent medium attributes are assigned. The specific implementation process is the same as in the aforementioned embodiments and will not be repeated here.

[0117] Step S3: For the initial sound velocity tensor matrix mentioned above, boundary fuzzing and tolerance processing based on morphological gradient are used to obtain the first sound velocity tensor matrix.

[0118] Because the geometric model generated in the previous steps may have ±10... A 30mm spatial error (originating from SLAM drift or inversion parameter estimation bias) means that directly using sharp "hard boundaries" may cause the calculated propagation path to erroneously refract or reflect at the model edges. This step aims to construct a "physical property transition zone" to prevent algorithm crashes due to minute boundary displacements and improve system robustness. The specific implementation process is the same as in the aforementioned embodiments and will not be repeated here.

[0119] Step S4: For the first sound velocity tensor matrix, in-situ calibration of the sound velocity field based on sensor array mutual measurement constraints is used to obtain the target sound velocity tensor matrix. Using sensors installed on the transformer tank wall (whose precise coordinates have been locked via SLAM scanning) as "known anchor points," several "detection sound paths" traversing the transformer's interior are constructed. By comparing the measured time-of-flight (ToF) of these paths with the model-calculated time of flight, the global sound velocity scaling factor for the entire field is corrected in reverse. The specific implementation process is the same as in the aforementioned embodiments and will not be repeated here.

[0120] Step S5: Anisotropic fast-moving solution based on semantic template adaptive switching generates a 3D time field matrix. By solving the equation, the distance from any point in space to the target sensor is calculated. The shortest "physical flight time" scalar field The core of this step lies in dynamically adjusting the discretization solution template (Stencil) of the partial differential equation using the semantic tag ID fixed in step S1. The specific implementation process is the same as in the previous embodiment, and will not be repeated here.

[0121] Step S6: Based on the residual minimization solution of the physical time map, the partial discharge location result of the transformer is obtained. Combining the TDoA (Time Difference of Arrival) data of the partial discharge signal captured in real time during transformer operation, the spatial coordinates of the partial discharge occurrence are accurately located in the N three-dimensional time field maps generated in step S5 through a global optimization algorithm. The specific implementation process is the same as the aforementioned embodiments, and will not be repeated here.

[0122] It should be noted that, in response to the aforementioned deficiencies in related technologies, this embodiment aims to provide an adaptive compensation method for transformer partial discharge velocity based on a semantic voxel field anisotropic propagation model. This embodiment does not rely on ideal drawing models or general full-wave simulations. Instead, it addresses the specific physical structure of the transformer by constructing a "semantic-physical" mapping mechanism and an in-situ calibration algorithm to specifically solve the following technical problems: 1) Solving the problem of severe distortion in velocity estimation caused by the anisotropic propagation characteristics in the transformer winding structure (core objective). This embodiment constructs a sound velocity tensor field based on local principal axial quantities, which can dynamically calculate the effective velocity according to the actual propagation direction of the signal (whether it travels along the line or crosses laterally). This eliminates the systematic bias caused by a single velocity assumption and significantly improves the calculation accuracy of signals passing through the winding region. 2) Solving the problem of poor algorithm robustness and easy jumps when relying on "imperfect geometric models" for simulation. This embodiment introduces a boundary fuzzification and tolerance processing mechanism based on morphological gradients to construct a physical property transition zone at the interface between oil and windings, and between oil and tap changers. This allows the algorithm to "accommodate" modeling errors, ensuring the continuity and stability of the positioning results even with a certain tolerance in the geometric model. 3) It solves the problem of mismatch between theoretical simulation parameters and the actual physical state (temperature, aging) of the equipment on site. This embodiment proposes an in-situ calibration method based on mutual measurement constraints of sensor arrays. Using sensors at known locations as "spatial anchor points," the global sound velocity field parameters are reverse-engineered and corrected using measured signals, achieving "instant calibration" and ensuring that the physical parameters of the model are highly consistent with the actual state of the equipment at the current moment. 4) It solves the problem of large computational load in solving complex non-uniform medium fields, making it difficult to meet the real-time positioning requirements on site. In related technologies, finite element simulation (FEM) or time-domain difference (FDTD) calculations for complex media are extremely time-consuming and cannot be run in real time on portable devices. This embodiment designs a semantically driven adaptive template fast-moving algorithm (Adaptive FMM). By recognizing the semantic ID of voxels, it automatically switches to a high-speed calculation mode in pure oil areas and only enables a high-precision anisotropic calculation mode in complex areas such as windings. While maintaining accuracy, computational efficiency was improved by several orders of magnitude, meeting the on-site requirement of "results within seconds" for partial discharge detection. 5) Solving the problem of obstruction and scattering interference of partial discharge signals by complex internal components such as tap changers. Related technologies often ignore tap changers or simply treat them as iron cores, resulting in inaccurate calculation of signal paths passing through this area. This embodiment accurately describes the physical characteristics of tap changers as "semi-transmitting and high attenuation" through envelope voxelization and penalty factor modeling. It effectively suppresses the weight of low-quality direct wave paths passing through complex insulation components, prioritizing the algorithm to find stronger and more reliable diffraction paths, thus improving the reliability of the positioning results.

[0123] This embodiment achieves at least one of the following effects: 1) It breaks through the bottleneck of modeling the "anisotropic" propagation of transformer windings, significantly improving positioning accuracy. Related technologies typically simplify the windings to a single scalar sound velocity (e.g., equivalent to a mixed medium), failing to reflect the speed difference between sound wave propagation along the conductor (fast) and propagation through the insulation layer (slow), resulting in a path calculation error of over 20% across the windings. This embodiment innovatively constructs a sound velocity tensor model based on local principal axis quantities. The algorithm can dynamically calculate the accurate effective wave velocity based on the angle between the signal propagation direction and the winding texture direction. 2) It possesses "engineering fault tolerance" for imperfect geometric models, exhibiting extremely strong algorithm robustness. Obstacle avoidance algorithms in related technologies rely on absolutely precise "hard boundaries." Once the externally inverted model has a small size deviation (e.g., SLAM error), it easily leads to erroneous jumps between "direct" and "diffraction" in path calculation, causing unstable positioning results. This embodiment introduces boundary fuzzification processing based on morphological gradients. A physical property transition zone was constructed at the interface between oil and winding, and between oil and tap changer, and an uncertainty penalty factor was introduced for complex components. 3) "In-situ adaptive calibration" of physical parameters was achieved, solving the problem of the disconnect between theoretical parameters and actual conditions. Related technologies rely on looking up tables to obtain empirical values ​​of sound velocity, which cannot adapt to changes in oil temperature, insulation aging, or fastening force during actual transformer operation, resulting in "rigid" model parameters. This embodiment establishes a closed-loop calibration mechanism based on mutual measurement constraints of sensor arrays. Using sensors at known locations as anchor points, the global sound velocity field parameters are deduced and corrected using measured signals. 4) A "semantic-driven adaptive computing architecture" is adopted, balancing high accuracy and real-time performance. Related technologies mainly use full-wave simulation (FEM / FDTD), which has high accuracy but extremely long computation time (hours); analytical methods are fast but have low accuracy, making it difficult to achieve both. This embodiment designs an adaptive template fast-moving algorithm (AdaptiveFMM). Using the semantic ID in S1 as a "switch", the high-speed template is automatically used in the pure oil area, and the high-precision template is only enabled in complex areas. 5) This solution addresses the positioning failure problem caused by non-line-of-sight (NLOS) propagation. In related technologies, the TDoA equations solution often fails to establish a correct linear equation when the signal is blocked by the iron core, leading to no solution or convergence to the wrong location. This embodiment employs a residual minimization search strategy based on a physical time map. The pre-calculated time map already contains all diffraction and refraction information.

[0124] This embodiment also provides a transformer partial discharge location device based on adaptive compensation for partial discharge wave velocity. This device is used to implement the above embodiments and preferred embodiments, and details already described will not be repeated. As used below, the terms "module" and "device" can refer to a combination of software and / or hardware that performs a predetermined function. Although the devices described in the following embodiments are preferably implemented in software, hardware implementation, or a combination of software and hardware, is also possible and contemplated.

[0125] According to an embodiment of the present invention, an apparatus embodiment for implementing the above-described transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity is also provided. Figure 4 This is a schematic diagram of a transformer partial discharge location device based on adaptive compensation of partial discharge wave velocity according to an embodiment of the present invention, as shown below. Figure 4 As shown, the above-mentioned transformer partial discharge location device based on adaptive compensation of partial discharge wave velocity includes: a conversion module 400, a sound velocity tensor determination module 402, an arrival time strategy determination module 404, a three-dimensional time field matrix generation module 406, and a partial discharge location module 408, wherein:

[0126] The conversion module 400 is used to convert the three-dimensional semantic mesh model of the transformer into a three-dimensional semantic voxel matrix. The three-dimensional semantic mesh model consists of multiple sub-mesh, which correspond one-to-one with multiple physical components inside the transformer. Each sub-mesh carries a semantic label for the corresponding physical component, which is used to identify the device identifier and device attributes of the corresponding physical component. The value of each voxel in the semantic voxel matrix is ​​used to indicate the semantic label of the corresponding physical component.

[0127] The sound velocity tensor determination module 402 is used to determine the target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix, wherein the value of each voxel in the target sound velocity tensor matrix is ​​used to indicate the physical sound velocity characteristics of the corresponding physical component.

[0128] The arrival time strategy determination module 404 is used to determine the arrival time strategy template corresponding to each voxel included in the target sound velocity tensor matrix according to the semantic tags identified by the three-dimensional semantic voxel matrix. The arrival time strategy template is used to indicate the determination strategy of the arrival time of the partial discharge signal at the corresponding voxel position.

[0129] The three-dimensional time field matrix generation module 406 is used to generate multiple three-dimensional time field matrices corresponding to multiple sensors on the transformer surface based on the target sound velocity tensor matrix and according to their respective arrival time strategy templates. The multiple sensors correspond one-to-one with the multiple three-dimensional time field matrices. The value of each voxel in the three-dimensional time field matrix is ​​used to indicate the shortest physical time required for the partial discharge signal emitted from the corresponding physical component location to reach the corresponding reference sensor.

[0130] The partial discharge location module 408 is used to determine the partial discharge location result of the transformer based on multiple three-dimensional time field matrices and the observation arrival time difference data corresponding to multiple sensors when a partial discharge signal of the transformer is detected. The observation arrival time difference data is used to indicate the time difference between the corresponding sensor and the reference sensor when they capture the partial discharge signal.

[0131] It should be noted that the above modules can be implemented by software or hardware. For example, for the latter, it can be implemented in the following ways: the above modules can be located in the same processor; or the above modules can be located in different processors in any combination.

[0132] It should be noted that the aforementioned conversion module 400, sound velocity tensor determination module 402, arrival time strategy determination module 404, three-dimensional time field matrix generation module 406, and partial discharge location module 408 correspond to steps S102 to S110 in the embodiments. The instances and application scenarios implemented by the above modules and corresponding steps are the same, but are not limited to the content disclosed in the above embodiments. It should be noted that the above modules, as part of the device, can run in a computer terminal.

[0133] It should be noted that the optional or preferred implementation methods of this embodiment can be found in the relevant descriptions in the embodiments, and will not be repeated here.

[0134] The aforementioned transformer partial discharge location device based on adaptive compensation of partial discharge wave velocity may further include a processor and a memory. The aforementioned conversion module 400, sound velocity tensor determination module 402, arrival time strategy determination module 404, three-dimensional time field matrix generation module 406, partial discharge location module 408, etc., are all stored in the memory as program modules, and the processor executes the aforementioned program modules stored in the memory to realize the corresponding functions.

[0135] The processor contains a core that retrieves the corresponding program modules from memory. One or more cores may be configured. Memory may include non-persistent memory in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory includes at least one memory chip.

[0136] According to an embodiment of this application, an embodiment of a non-volatile storage medium is also provided. Optionally, in this embodiment, the non-volatile storage medium includes a stored program, wherein, when the program runs, it controls the device containing the non-volatile storage medium to execute any of the aforementioned transformer partial discharge location methods based on adaptive compensation for partial discharge velocity.

[0137] Optionally, in this embodiment, the non-volatile storage medium may be located in any computer terminal in a group of computer terminals in a computer network, or in any mobile terminal in a group of mobile terminals, and the non-volatile storage medium includes stored programs.

[0138] Optionally, during program execution, a program can be used to control the device containing the non-volatile storage medium to execute any of the steps of the transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity described above.

[0139] According to an embodiment of this application, an embodiment of a processor is also provided. Optionally, in this embodiment, the processor is used to run a program, wherein the program executes any of the above-described transformer partial discharge location methods based on adaptive compensation for partial discharge wave velocity.

[0140] According to an embodiment of this application, an embodiment of a computer program product is also provided, which, when executed on a data processing device, is adapted to execute a program that initializes the transformer partial discharge location method steps based on adaptive compensation for partial discharge wave velocity, as described above.

[0141] like Figure 5 As shown, an embodiment of the present invention provides an electronic device 10, which includes a processor, a memory, and a program stored in the memory and executable on the processor. When the processor executes the program, it implements the steps of the transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity as described above.

[0142] The order of the above embodiments of the present invention is merely for description and does not represent the superiority or inferiority of the embodiments.

[0143] In the above embodiments of the present invention, the descriptions of each embodiment have different focuses. For parts not described in detail in a certain embodiment, please refer to the relevant descriptions of other embodiments.

[0144] In the several embodiments provided in this application, it should be understood that the disclosed technical content can be implemented in other ways. The device embodiments described above are merely illustrative; for example, the division of modules described above can be a logical functional division, and in actual implementation, there may be other division methods. For example, multiple modules or components may be combined or integrated into another system, or some features may be ignored or not executed. Furthermore, the coupling or direct coupling or communication connection shown or discussed may be through some interfaces, or indirect coupling or communication connection between modules, and may be electrical or other forms.

[0145] The modules described above as separate components may or may not be physically separate. Similarly, the components shown as modules may or may not be physical modules; they may be located in one place or distributed across multiple modules. Some or all of the modules can be selected to achieve the purpose of this embodiment, depending on actual needs.

[0146] Furthermore, the functional modules in the various embodiments of the present invention can be integrated into one processing module, or each module can exist physically separately, or two or more modules can be integrated into one module. The integrated modules described above can be implemented in hardware or as software functional modules.

[0147] If the aforementioned integrated modules are implemented as software functional modules and sold or used as independent products, they can be stored in a computer-readable non-volatile storage medium. Based on this understanding, the technical solution of this invention, in essence, or the part that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a non-volatile storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods of the various embodiments of this invention. The aforementioned non-volatile storage medium includes various media capable of storing program code, such as USB flash drives, read-only memory (ROM), random access memory (RAM), portable hard drives, magnetic disks, or optical disks.

[0148] The above are merely preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principle of the present invention, and these improvements and modifications should also be considered within the scope of protection of the present invention.

Claims

1. A transformer partial discharge location method based on adaptive compensation for partial discharge wave velocity, characterized in that, include: The three-dimensional semantic mesh model of the transformer is transformed into a three-dimensional semantic voxel matrix. The three-dimensional semantic mesh model consists of multiple sub-mesh, which correspond one-to-one with multiple physical components inside the transformer. Each sub-mesh carries a semantic label for the corresponding physical component, which is used to identify the device identifier and device attributes of the corresponding physical component. The value of each voxel in the semantic voxel matrix is ​​used to indicate the semantic label of the corresponding physical component. Determine the target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix, wherein the value of each voxel in the target sound velocity tensor matrix is ​​used to indicate the physical sound velocity characteristics of the corresponding physical component; According to the semantic tags identified by the three-dimensional semantic voxel matrix, the arrival time strategy templates corresponding to each voxel included in the target sound velocity tensor matrix are determined, wherein the arrival time strategy templates are used to indicate the strategy for determining the arrival time of the partial discharge signal at the corresponding voxel position. Based on the target sound velocity tensor matrix, and according to the respective arrival time strategy templates, multiple three-dimensional time field matrices are generated for multiple sensors on the transformer surface, wherein the multiple sensors correspond one-to-one with the multiple three-dimensional time field matrices; the value of each voxel in the three-dimensional time field matrix is ​​used to indicate the shortest physical time required for the partial discharge signal emitted from the corresponding physical component location to reach the corresponding reference sensor. When a partial discharge signal is detected in the transformer, the partial discharge location result of the transformer is determined based on the plurality of three-dimensional time field matrices and the observation arrival time difference data corresponding to each of the plurality of sensors. The observation arrival time difference data is used to indicate the time difference between the corresponding sensor and the reference sensor in capturing the partial discharge signal.

2. The method according to claim 1, characterized in that, Determining the target sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix includes: Based on the device attributes of the multiple physical components, the initial sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix is ​​determined. The transition region in the initial sound velocity tensor matrix is ​​blurred to obtain the first sound velocity tensor matrix. The blurring process is used to smooth the sound velocity tensor of the transition region, which is the boundary region between the transformer oil and the winding in the transformer. Based on the target oil speed temperature change factor and the target winding structure aging factor, the first sound velocity tensor matrix is ​​corrected to obtain the target sound velocity tensor matrix. The target oil speed temperature change factor is used to correct the sound velocity deviation caused by oil temperature changes, and the target winding structure aging factor is used to correct the sound velocity deviation caused by insulation aging or changes in fastening force.

3. The method according to claim 2, characterized in that, In the case where the plurality of physical components include the transformer oil, tap changer, iron core, and windings, determining the initial sound velocity tensor matrix corresponding to the three-dimensional semantic voxel matrix based on the equipment attributes of the plurality of physical components includes: Based on the oil temperature parameters of the transformer oil, determine the initial velocity of sound corresponding to the transformer oil; The initial sound velocity corresponding to the tap changer is determined to be a preset first value; The initial sound velocity corresponding to the iron core is determined to be a preset second value; Based on the preset empirical sound velocities of the winding in each of the three principal directions, the predetermined sound velocity tensor of the winding is determined, wherein the three principal directions include the axial, tangential, and radial directions of the winding; The initial sound velocity of the winding is determined based on the sound wave propagation direction of the winding and the predetermined sound velocity tensor. According to the voxel positions in the three-dimensional semantic voxel matrix, the initial sound velocities corresponding to the transformer oil, the tap changer, the iron core, and the winding are arranged to obtain the initial sound velocity tensor matrix.

4. The method according to claim 2, characterized in that, The process of blurring the transition region in the initial sound velocity tensor matrix to obtain the first sound velocity tensor matrix includes: In the initial sound velocity tensor matrix, the sound velocity tensors of the voxels belonging to the transformer oil in the predetermined neighborhood of the transition region are uniformly converted into a diagonal tensor matrix, wherein the diagonal elements of the diagonal tensor matrix are the scalar sound velocities of the transformer oil. Obtain the sound velocity tensor of the voxel corresponding to the winding in the first predetermined neighborhood of the transition region in the initial sound velocity tensor matrix; Based on the Euclidean distance from the transition zone to the pure oil region and the Euclidean distance from the transition zone to the corresponding region of the winding, the oil property weight of the corresponding voxel of the transformer oil in the transition zone is determined. Based on the oil property proportion weight, the diagonal tensor matrix and the sound velocity tensor of the voxel corresponding to the winding in the first predetermined neighborhood are linearly interpolated to generate the smooth sound velocity tensor corresponding to the transition region. The sound velocity tensor in the transition region of the initial sound velocity tensor matrix is ​​replaced with the smooth sound velocity tensor to obtain the first sound velocity tensor matrix.

5. The method according to claim 4, characterized in that, The step of replacing the sound velocity tensor in the transition region of the initial sound velocity tensor matrix with the smoothed sound velocity tensor to obtain the first sound velocity tensor matrix includes: The sound velocity tensor of the transition region in the initial sound velocity tensor matrix is ​​replaced with the smooth sound velocity tensor. The first predetermined region is marked as the isolation region, and a preset uncertainty penalty factor is introduced for the voxel corresponding to the second predetermined region to obtain the first sound velocity tensor matrix. The first predetermined region is the boundary region between the transformer oil and the iron core, the second predetermined region is the minimum region covering the tap changer at the boundary position with the transformer oil, and the isolation region represents the region that does not support sound wave propagation.

6. The method according to claim 2, characterized in that, Before correcting the first sound velocity tensor matrix based on the target oil speed temperature change factor and the target winding structure aging factor to obtain the target sound velocity tensor matrix, the method further includes: The target oil speed temperature change factor and the target winding structure aging factor are obtained in the following manner: The acoustic propagation path of at least two of the multiple sensors, which are connected through the core region of the winding, is selected as the calibration path. An external acoustic calibration source is used to excite a standard pulse signal at the starting point of the calibration path, and the measured arrival time of each sensor on the calibration path is recorded. The measured arrival time refers to the time when the corresponding sensor actually receives the excitation standard pulse signal. Based on the target sound velocity tensor matrix, determine the theoretical arrival time of each sensor on the calibration path; Construct a residual objective function between the measured arrival time and the theoretical arrival time for each sensor on the calibration path, wherein the residual objective function is the sum of squares of the differences between the measured arrival time and the theoretical arrival time; With the goal of minimizing the residual objective function, the initial oil speed temperature change factor and the initial winding structure aging factor are optimized to obtain the target oil speed temperature change factor and the target winding structure aging factor.

7. The method according to any one of claims 1 to 6, characterized in that, The step of determining the arrival time strategy template corresponding to each voxel in the target sound velocity tensor matrix according to the semantic tags identified by the three-dimensional semantic voxel matrix includes: In the target sound velocity tensor matrix, the arrival time strategy template corresponding to the voxel with the semantic label "transformer oil" is determined to be the first strategy template. The first strategy template is used to indicate that, based on the diagonal sound velocity tensor of the corresponding voxel, a univariate quadratic equation is solved in six orthogonal neighborhoods using a specified difference scheme to calculate the minimum arrival time of the corresponding voxel. In the target sound velocity tensor matrix, the arrival time strategy template corresponding to the voxel with the semantic label of winding is determined to be the second strategy template. The second strategy template is used to indicate the anisotropic sound velocity tensor based on voxels, search for the simplex that best matches the main velocity direction in twenty-six neighborhoods, and obtain the minimum arrival time that satisfies the specified equation through matrix transformation and iterative solution. In the target sound velocity tensor matrix, the arrival time strategy template corresponding to the voxel with the semantic label of iron core is determined to be the third strategy template, wherein the third strategy template is used to set the arrival time of the corresponding voxel to a specified value in order to isolate the sound wave propagation.

8. The method according to any one of claims 1 to 6, characterized in that, In the event of a partial discharge signal detected in the transformer, the method for determining the partial discharge location result of the transformer based on the multiple three-dimensional time field matrices and the observation arrival time difference data corresponding to each of the multiple sensors includes: Using any voxel within the non-obstacle voxel range in the three-dimensional semantic voxel matrix as a candidate point, the theoretical arrival time difference data of the candidate point at each of the multiple three-dimensional time field matrices are read. Construct a residual function, wherein the residual function is the sum of squares of the differences between the observed time difference of arrival data and the theoretical time difference of arrival data; Based on the residual function, all voxels are traversed within the non-obstacle voxel range of the three-dimensional semantic voxel matrix, and the residual function value corresponding to each voxel is calculated. The partial discharge localization result is determined based on the voxel with the smallest residual function value among all voxels.

9. The method according to claim 8, characterized in that, The determination of the partial discharge localization result based on the voxel with the smallest residual function value among all voxels includes: The voxel with the smallest residual function value among all voxels is taken as the initial discharge localization result. A trilinear interpolation function is constructed within a predetermined neighborhood centered on the initial discharge location result to fit the local residual surface; The minimum point is solved within the predetermined continuous domain of the trilinear interpolation function to obtain the target positioning coordinates; The target location coordinates are used as the partial discharge location result.

10. An electronic device, characterized in that, The method includes one or more processors and a memory, the memory being used to store one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors cause the one or more processors to implement the transformer partial discharge location method based on adaptive compensation of partial discharge wave velocity as described in any one of claims 1 to 9.

Images