Transformer winding damage assessment method, system and intelligent terminal

Abstract

The present invention discloses a transformer winding damage assessment method, system, and intelligent terminal, relating to the technical field of transformer damage assessment. The method comprises: configuring a vibration sensor array on the transformer housing; deploying a multimodal sensing array; activating a fiber optic temperature sensor array, a high-frequency current mutual inductance array, and an online chromatographic sensor array when the vibration frequency domain energy distribution transmitted back by the vibration sensor array deviates from a preset frequency domain distribution; receiving multi-source heterogeneous sensing data; performing feature coupling on the multi-source heterogeneous sensing data to construct a three-dimensional damage feature tensor; performing damage verification and positioning, and outputting a real-time damage alarm. The present invention solves the technical problem of insufficient accuracy in transformer winding damage diagnosis and positioning, which leads to delayed fault response, in the prior art. It achieves the technical effect of accurately locating transformer winding damage and providing real-time alarms, thereby improving the comprehensiveness and timeliness of fault diagnosis.

Description

Technical Field

[0001] The present invention relates to the technical field of transformer damage assessment, and in particular to a transformer winding damage assessment method, system and intelligent terminal. Background Art

[0002] Existing transformer winding damage assessment faces multiple challenges. Traditional single-source monitoring methods, such as vibration monitoring, temperature detection, or chromatographic analysis, struggle to fully characterize the complex characteristics of winding damage due to their single data dimension and insufficient fusion capabilities. For example, relying solely on vibration signals is susceptible to mechanical noise interference and cannot correlate thermal damage or insulation degradation; single-source temperature monitoring struggles to distinguish between normal temperature rise and fault temperature rise; and the independent operation of each system leads to fragmented spatiotemporal data, making it impossible to accurately locate damage and analyze its dynamic evolution. Furthermore, existing technologies have low sensitivity to early, subtle damage characteristics and lack a collaborative analysis mechanism for multi-physics field data, resulting in delayed fault diagnosis and a high rate of misjudgment.

[0003] The existing technology has the technical problem that the transformer winding damage diagnosis and positioning accuracy is insufficient, resulting in delayed fault response. Summary of the Invention

[0004] The present application provides a transformer winding damage assessment method, system and intelligent terminal, which are used to solve the technical problem in the prior art of insufficient transformer winding damage diagnosis and positioning accuracy, resulting in delayed fault response.

[0005] In view of the above problems, the present application provides a transformer winding damage assessment method, system and intelligent terminal.

[0006] In a first aspect of the present application, a method for assessing transformer winding damage is provided, the method comprising:

[0007] A vibration sensor array is configured on the transformer housing according to a preset vibration sensor density; a multimodal sensing array is deployed according to the structural characteristics of the transformer winding, wherein the multimodal sensing array includes a fiber optic temperature sensor array, a high-frequency current mutual inductance array, and an online color spectrum sensor array; when the vibration frequency domain energy distribution transmitted back by the vibration sensor array deviates from the preset frequency domain distribution, the fiber optic temperature sensor array, the high-frequency current mutual inductance array, and the online color spectrum sensor array are activated in a graded manner according to the frequency band energy offset rate; multi-source heterogeneous sensing data collected and transmitted back by the fiber optic temperature sensor array, the high-frequency current mutual inductance array, and the online color spectrum sensor array are received; after the multi-source heterogeneous sensing data are spatially aligned with the transformer winding as a spatial constraint, a three-dimensional damage feature tensor is constructed by feature coupling the multi-source heterogeneous sensing data; damage verification and positioning are performed based on the three-dimensional damage feature tensor, and a real-time damage alarm is output.

[0008] A second aspect of the present application provides a transformer winding damage assessment system, the system comprising:

[0009] A vibration sensing array configuration module is configured to configure a vibration sensing array on the transformer housing according to a preset vibration sensing density. A multimodal sensing array deployment module is configured to deploy a multimodal sensing array according to the structural characteristics of the transformer winding, wherein the multimodal sensing array includes a fiber optic temperature sensing array, a high-frequency current mutual inductance array, and an online color spectrum sensing array. An array activation module is configured to hierarchically activate the fiber optic temperature sensing array, the high-frequency current mutual inductance array, and the online color spectrum sensing array according to the frequency band energy deviation rate when the vibration frequency domain energy distribution transmitted back by the vibration sensing array deviates from the preset frequency domain distribution. A multi-source heterogeneous sensing data acquisition module is configured to receive multi-source heterogeneous sensing data collected and transmitted back by the fiber optic temperature sensing array, the high-frequency current mutual inductance array, and the online color spectrum sensing array. A feature coupling module is configured to construct a three-dimensional damage feature tensor by feature coupling the multi-source heterogeneous sensing data after spatially and temporally aligning the multi-source heterogeneous sensing data using the transformer winding as a spatial constraint. A real-time damage alarm output module is configured to perform damage verification and location based on the three-dimensional damage feature tensor and output a real-time damage alarm.

[0010] The third aspect of the present application provides an intelligent terminal, which includes: a processor; a memory for storing instructions executable by the processor; wherein the processor is used to execute the transformer winding damage assessment method provided in the present application.

[0011] One or more technical solutions provided in this application have at least the following technical effects or advantages:

[0012] A vibration sensor array is configured on the transformer housing based on a preset vibration sensor density; a multimodal sensing array is deployed based on the structural characteristics of the transformer winding; when the vibration frequency domain energy distribution deviates from the preset frequency domain distribution, the fiber optic temperature sensor array, high-frequency current mutual inductance array, and online color spectrum sensor array are activated; multi-source heterogeneous sensing data collected and transmitted back by the fiber optic temperature sensor array, high-frequency current mutual inductance array, and online color spectrum sensor array are received; after spatially aligning the multi-source heterogeneous sensing data with the transformer winding as a spatial constraint, feature coupling is performed on the multi-source heterogeneous sensing data to construct a three-dimensional damage feature tensor; damage verification and location are performed based on the three-dimensional damage feature tensor, and real-time damage alarms are output. This achieves the technical effect of accurately locating transformer winding damage and providing real-time alarms, improving the comprehensiveness and timeliness of fault diagnosis. BRIEF DESCRIPTION OF THE DRAWINGS

[0013] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0014] Figure 1 A schematic diagram of a flow chart of a transformer winding damage assessment method provided in an embodiment of the present application;

[0015] Figure 2 A schematic diagram of the structure of a transformer winding damage assessment system provided in an embodiment of the present application.

[0016] Figure 3 A schematic diagram of the structure of a smart terminal provided in this application.

[0017] Explanation of the accompanying drawings: vibration sensing array configuration module 10, multimodal sensing array deployment module 20, array activation module 30, multi-source heterogeneous sensing data acquisition module 40, feature coupling module 50, real-time damage alarm output module 60, processor 21, memory 22, input device 23, output device 24. DETAILED DESCRIPTION

[0018] This application provides a transformer winding damage assessment method, system and intelligent terminal to solve the technical problem in the prior art of insufficient transformer winding damage diagnosis and positioning accuracy, resulting in delayed fault response.

[0019] The following will be combined with the accompanying drawings in the embodiments of this application to clearly and completely describe the technical solutions in the embodiments of this application. Obviously, the embodiments described are only part of the embodiments of this application, not all of them. Based on the embodiments in this application, all other embodiments obtained by ordinary technicians in this field without making any creative work are within the scope of protection of this application.

[0020] Example 1, as Figure 1 As shown, the present application provides a transformer winding damage assessment method, the method comprising:

[0021] Step S100: configuring a vibration sensor array on a transformer housing according to a preset vibration sensor density.

[0022] Specifically, a vibration sensor array is configured on the transformer casing according to the preset vibration sensor density. Specifically, the transformer casing is gridded according to the vibration sensor density, multiple deployment nodes are located, and vibration response analysis of the transformer winding is performed to obtain the vibration sensitivity of each deployment node. Gain balance matching is performed based on the sensitivity results, and vibration sensors are installed accordingly to complete the configuration of the vibration sensor array, thereby achieving comprehensive and sensitive capture of the vibration signal of the transformer casing.

[0023] Step S200: deploying a multimodal sensing array according to the structural characteristics of the transformer winding, wherein the multimodal sensing array includes a fiber optic temperature sensing array, a high-frequency current mutual inductance array, and an online color spectrum sensing array.

[0024] Specifically, based on the structural characteristics of the transformer winding (such as winding layout, electrical connection path, etc.), a multimodal sensing array is deployed at key positions of the winding (such as areas prone to heat generation, current concentration areas and insulating oil flow paths). The array includes a fiber optic temperature sensor array (for monitoring the temperature gradient distribution of the winding), a high-frequency current mutual inductance array (for collecting current harmonic distortion signals) and an online chromatography sensor array (for analyzing characteristic gas components in the insulating oil). Through the multimodal sensing array, multi-dimensional data collection of the winding temperature field, current spectrum and insulation degradation status is achieved, providing multi-source heterogeneous data support for subsequent damage feature coupling analysis.

[0025] Step S300: When the vibration frequency domain energy distribution transmitted back by the vibration sensing array deviates from the preset frequency domain distribution, the optical fiber temperature sensing array, the high frequency current mutual inductance array and the online color spectrum sensing array are activated in stages according to the frequency band energy deviation rate.

[0026] Specifically, when the vibration frequency domain energy distribution transmitted back by the vibration sensing array deviates from the preset frequency domain distribution, the energy offset rate of each deployed node is first calculated, and the first K nodes are selected in ascending order as candidate attenuation starting points. The K internal vibration attenuation starting points are located by reverse tracing through linear attenuation, and the fault risk space is framed after clustering based on spatial proximity; then, the spatial hierarchical scale is matched according to the frequency band energy offset rate, and the transformer winding space is divided into multi-level areas, and the multimodal sensing array is hierarchical into multi-level sub-arrays; finally, the sub-arrays at each level are activated from high to low according to the risk level, and spatial hierarchical data is collected for the winding to obtain multi-source heterogeneous sensing data containing multi-level sensing data sets, realizing hierarchical response and precise coverage from vibration anomalies to multimodal perception.

[0027] Step S400: receiving multi-source heterogeneous sensing data collected and transmitted back by the optical fiber temperature sensor array, the high-frequency current mutual inductance array, and the online chromatography sensor array.

[0028] Specifically, the data acquisition system receives multi-source heterogeneous sensing data in real time from the fiber optic temperature sensor array, high-frequency current mutual inductance array, and online chromatographic sensor array. This data includes winding temperature field distribution data (reflecting temperature gradient characteristics) collected by the fiber optic temperature sensor array, current harmonic spectrum data (reflecting current distortion characteristics) collected by the high-frequency current mutual inductance array, and characteristic gas composition data in the insulating oil (reflecting insulation degradation characteristics) collected by the online chromatographic sensor array. This data covers multi-dimensional information such as temperature, current, and insulation status, providing multimodal raw data support for subsequent spatiotemporal alignment processing and the construction of a three-dimensional damage feature tensor.

[0029] Step S500: After the multi-source heterogeneous sensing data are spatiotemporally aligned using the transformer winding as a spatial constraint, a three-dimensional damage feature tensor is constructed by feature coupling the multi-source heterogeneous sensing data.

[0030] Specifically, using the benchmark geometric model of the transformer winding as a spatial constraint, the multi-source heterogeneous sensing data (including temperature field, current spectrum, and oil chromatography data) collected by the fiber optic temperature sensor array, high-frequency current mutual inductance array, and online chromatography sensor array are first spatiotemporally aligned to ensure the consistency of the data in timestamps and spatial coordinates. Then, the first-level temperature field, current spectrum, and oil chromatography sensing data corresponding to the fault risk space are extracted from the multi-level sensing data set, and multimodal damage identification and marking are performed on each of them. The temperature gradient is calculated for the temperature data and the damage feature points are marked to generate a first-level temperature damage feature tensor. The harmonic energy ratio is calculated for the current data and the high-frequency energy area is marked to generate a first-level current damage feature tensor. The characteristic gas analysis is performed on the chromatography data and the damage area is marked to generate a first-level chromatography damage feature tensor. Finally, spatial feature coupling is performed on the damage feature tensors at all levels in the benchmark geometric model. Through layer-by-layer fusion and spatial connection, a three-dimensional damage feature tensor is constructed, which includes the physical dimensions of mechanical deformation, current distortion, and temperature gradient, as well as the spatiotemporal dimensions of time, frequency, and spatial domains, to achieve a multi-dimensional spatiotemporal integrated characterization of winding damage.

[0031] Step S600: performing damage verification and positioning based on the three-dimensional damage feature tensor, and outputting a real-time damage alarm.

[0032] Specifically, first, multi-level slicing scales and multi-level damage confidence weights are configured for the multi-level winding space, and the three-dimensional damage feature tensor is sliced ​​layer by layer along the winding space to obtain a multi-level slicing damage feature set including temperature damage, current damage, and chromatographic damage, and it is vectorized into a multi-level damage feature matrix; then, the preset damage feature vector library is used to traverse and match the multi-level damage feature matrix, and the corresponding multi-level damage intensity matrix is ​​output; finally, the damage intensity matrix is ​​linearly weighted integrated according to the multi-level damage confidence weight to obtain M winding damage quantitative features. The damage location is verified and the damage level is evaluated through the quantitative features, and alarm information containing specific damage areas and risk levels is output in real time, providing accurate fault location and risk warning for transformer operation and maintenance.

[0033] In one possible implementation, step S100 further includes:

[0034] Step S110: locating a plurality of deployment nodes based on the gridded transformer housing of the vibration sensing density.

[0035] Step S120: performing vibration response on the transformer winding to obtain multiple vibration sensitivities of the multiple deployment nodes.

[0036] Step S130: performing gain equalization matching according to the multiple vibration sensitivities, and calling and installing the vibration sensor according to the matching result, thereby completing the configuration of the vibration sensor array.

[0037] Specifically, first, the transformer casing is spatially gridded according to the preset vibration sensing density. By establishing a three-dimensional coordinate system, the casing surface is divided into uniform grid units. The vibration-sensitive area is determined by combining the winding structure characteristics and historical fault data, thereby accurately locating multiple sensor deployment nodes.

[0038] When analyzing the vibration response of transformer windings, excitations simulating operating conditions (such as current excitation or mechanical load) are applied to the windings. Vibration signals from each deployed node are collected using a laser vibrometer or accelerometer. Signal processing algorithms such as fast Fourier transform (FFT) are used to obtain parameters such as the vibration amplitude and phase of each node at different frequencies. Combined with the winding structure dynamics model, the quantitative value of each node's sensitivity to the mechanical deformation of the winding, i.e., vibration sensitivity, is calculated. This value characterizes the signal contribution of each node when monitoring winding looseness or deformation, providing data support for subsequent sensor gain matching and array optimization.

[0039] Based on the vibration sensitivity data of each deployed node, a mapping relationship between sensitivity and sensor gain is established. For nodes with high sensitivity, low-gain sensors are selected to avoid vibration signal overload and distortion. For nodes with low sensitivity, high-gain sensors are configured to improve the accuracy of weak vibration signal acquisition. By traversing the sensitivity distribution of all nodes, a particle swarm optimization (PSO) algorithm is used to globally balance sensor types (such as piezoelectric and fiber optic) with gain parameters to ensure that the vibration signal output of each node is within a uniform dynamic range. After matching, vibration sensors are installed at the corresponding nodes according to the optimization results, forming a vibration sensor array covering key areas of the transformer casing, achieving differentiated and high-precision monitoring of the winding mechanical condition.

[0040] In one possible implementation, step S300 further includes:

[0041] Step S310: Based on the frequency band energy offset rate of the vibration frequency domain energy distribution and the preset frequency domain distribution, vibration conduction attenuation prediction is performed with the multiple deployment nodes as starting points to define the fault risk space.

[0042] Step S320: Taking the fault risk space as a starting point, hierarchically transforming the multimodal sensing array into multiple levels of multimodal sub-arrays.

[0043] Step S330: hierarchically activating the multi-level multimodal sub-arrays to collect spatial hierarchical data of the transformer winding, and obtaining the multi-source heterogeneous perception data including a multi-level perception data set.

[0044] Specifically, by calculating the frequency band energy offset rate between the vibration frequency domain energy distribution transmitted by the vibration sensing array and the preset frequency domain distribution, the node energy offset rate of each deployed node is obtained and arranged in ascending order, and the top K nodes with higher offset rates are screened out as candidate attenuation starting points; based on the elastic wave propagation theory or the transformer structural dynamics model, the vibration signal of each candidate starting point is linearly attenuated and reversely traced to simulate the conduction path of the vibration energy in the winding and casing structure, and K internal vibration attenuation starting points are located; a spatial proximity clustering algorithm (such as DBSCAN) is used to perform cluster analysis on the K internal attenuation starting points, and a closed area is formed by connecting the cluster boundaries, and finally a fault risk space containing the potential damage location of the winding is framed, providing a spatial positioning basis for the subsequent hierarchical activation of the multimodal sensing array.

[0045] Starting from the fault risk space, the transformer winding space is divided into multiple levels of winding space based on the offset scale of the frequency band energy offset rate and the spatial hierarchical scale. The first-level subarray represents the core area within the fault risk space (e.g., the middle section of the winding). Within this area, high-density multimodal sensors, including fiber optic temperature sensors, high-frequency current transformers, and online chromatographic sensors, are deployed to achieve high-precision, real-time monitoring of the temperature field, current harmonics, and characteristic gases in the insulating oil. The second-level subarray represents the associated areas outside the fault risk space (e.g., adjacent windings and oil tank support structures). Basic sensors, namely fiber optic temperature sensors and high-frequency current transformers, are deployed in this area to monitor the diffusion trend of the temperature field and current distortion. By hierarchically dividing the multimodal sensing array into multiple levels of multimodal subarrays, a hierarchical monitoring system for the transformer winding space is formed, achieving differentiated perception of different risk areas.

[0046] Multi-level sub-arrays are activated from high to low risk levels, with the first-level sub-array being activated first for high-frequency, high-density data acquisition to capture temperature gradients, current harmonics, and oil chromatogram anomalies in the core area in real time. When the first-level sub-array detects damage characteristics exceeding the threshold, the second-level sub-array is automatically triggered to start and synchronously collect data from related areas, forming a multi-level perception data set that includes detailed data from the core area and trend data from the peripheral areas. Multi-source heterogeneous data fusion provides a multi-level evidence chain for damage assessment.

[0047] In one possible implementation, step S320 further includes:

[0048] Step S321: Matching a spatial layering scale according to the offset scale of the frequency band energy offset rate.

[0049] Step S322: Taking the fault risk space as a starting point, the transformer winding space is divided into multiple winding spaces according to the spatial hierarchical scale.

[0050] Step S323: hierarchically dividing the multimodal sensing array into multi-level multimodal sub-arrays according to the spatial coverage relationship between the multi-level winding space and the multimodal sensing array.

[0051] Specifically, first, based on the energy offset rate between the vibration frequency domain energy distribution and the preset frequency band distribution (for example, every 10% increase in the offset rate corresponds to an improvement in spatial monitoring accuracy by one level), a mapping relationship between the offset scale and the spatial stratification scale is established (for example, an offset rate ≤5% corresponds to the basic monitoring layer, 5% to 15% corresponds to the enhanced monitoring layer, and >15% corresponds to the core monitoring layer), and the spatial monitoring range and density corresponding to different risk levels are clarified.

[0052] Taking the fault risk space as the core starting point, the transformer winding space is divided into multi-level nested winding spaces (such as core damage area, direct impact area, and indirect impact area) from the inside to the outside according to the spatial hierarchical scale. Each level of space corresponds to a specific damage propagation probability and monitoring priority.

[0053] Finally, based on the multi-level winding space and the physical coverage range of the multimodal sensing array, the overall sensing array is divided into multi-level sub-arrays. The core damage area corresponds to the first-level sub-array, where high-density multimodal sensors are deployed (temperature, current, and color spectrum sensor spacing ≤ 10 cm); the directly affected area corresponds to the second-level sub-array, where medium-density sensors are configured (temperature and current sensor spacing ≤ 20 cm); and the indirect impact area corresponds to the third-level sub-array, where low-density basic sensors are used (temperature sensor spacing ≤ 30 cm). Through this hierarchical division, precise matching of sensor resources and risk areas is achieved, ensuring monitoring accuracy in high-risk areas and coverage efficiency in low-risk areas.

[0054] In one possible implementation, step S500 further includes:

[0055] Step S510: pre-constructing a reference geometric model of the transformer winding.

[0056] Step S520: extracting the primary temperature field sensing data, the primary current spectrum sensing data, and the primary oil chromatogram sensing data corresponding to the fault risk space from the multi-level sensing data set.

[0057] Step S530: performing multimodal damage identification marking on the primary temperature field sensing data, the primary current spectrum sensing data, and the primary oil chromatogram sensing data to obtain a primary temperature damage feature tensor, a primary current damage feature tensor, and a primary chromatogram damage feature tensor.

[0058] Step S540: performing spatial feature coupling on the first-level temperature damage feature tensor, the first-level current damage feature tensor, and the first-level chromatographic damage feature tensor in the reference geometric model to construct a first-level fusion damage feature tensor.

[0059] Step S550: Similarly, spatial feature coupling is performed on the multi-level perception data set to construct a multi-level fusion damage feature tensor.

[0060] Step S560: spatially connecting the multi-level fusion damage feature tensors to obtain the three-dimensional damage feature tensor.

[0061] Specifically, first, based on the design drawings and 3D scanning data of the transformer winding, a reference geometric model including the winding geometric dimensions, spatial coordinates and material properties is pre-built as a reference framework for multi-source data spatial alignment.

[0062] For the first-level temperature field sensing data, real-time temperature measurement data from fiber optic temperature sensors within the fault risk space is filtered to form a two-dimensional matrix consisting of spatial coordinates (X, Y, Z) and temperature values ​​(T), with a centimeter-level grid resolution. For the first-level current spectrum sensing data, the current signal of the high-frequency current mutual inductance array corresponding to the winding branch during the period of abnormal vibration is intercepted. After fast Fourier transform, the amplitude and phase information of the fundamental wave and third and fifth harmonic components are extracted to form a frequency-band-energy distribution feature vector. For the first-level oil chromatogram sensing data, the detection results of the online chromatogram sensor array at the corresponding oil circuit location in the risk space are retrieved. The concentration values ​​and unit time growth rates of characteristic gases such as hydrogen (H2) and acetylene (C2H2) are extracted to form a gas composition-time series dataset. Through this extraction process, high-fidelity raw data on temperature, current, and insulation status within the fault risk space are obtained, providing the core input for multimodal damage identification.

[0063] The temperature gradient is calculated based on the primary temperature field sensing data, and the temperature damage gradient characteristics (e.g., ≥2°C / cm) are set. The temperature field distribution is traversed to mark damage feature points where the temperature rises abnormally or the gradient changes suddenly. A temperature damage feature surface is formed by spatially connecting lines, and a primary temperature damage feature tensor is generated. Secondly, the harmonic energy ratio (e.g., the third harmonic energy ratio >5%) and the high-frequency energy ratio area (e.g., a sudden increase of 10% in the energy band >20kHz) are calculated for the primary current spectrum sensing data. Areas with significant current distortion are marked to form a current damage feature matrix and convert it into a primary current damage feature tensor. Finally, based on the insulating oil characteristic gas concentration standard (e.g., hydrogen >150ppm, acetylene >5ppm), the primary oil chromatogram sensing data is marked for chromatographic damage, the insulation degradation area is determined, and a primary chromatographic damage feature tensor is output. Through independent damage identification using multimodal data, a multi-dimensional damage feature set is constructed.

[0064] Using a pre-built transformer winding benchmark geometry model as the spatial mapping carrier, the first-level temperature damage feature tensor, the first-level current damage feature tensor, and the first-level chromatographic damage feature tensor are projected onto the corresponding coordinates of the model. Using a tensor space registration algorithm, the spatial grid of the temperature damage feature surface, the node coordinates of the current damage feature matrix, and the oil path of the chromatographic damage feature region are aligned with the physical structures of the benchmark geometry model, including the winding segments, pad structures, and oil channels. A feature cross-validation method is then used to verify the consistency of the spatially overlapping regions of the three tensors (for example, a spatial overlap of >80% between the temperature anomaly region and the current distortion region is marked as a strongly coupled feature). A tensor product operation is then used to fuse the location, type, and confidence information of the multimodal damage features to construct a first-level fused damage feature tensor that incorporates the temperature-current-chromatographic coupling relationship. This tensor intuitively presents the multi-physics field correlation characteristics of the core region damage in the spatial dimension, providing the underlying coupling foundation for subsequent multi-level feature fusion.

[0065] For the secondary perception data set, the temperature field and current spectrum data of the associated areas outside the risk space are extracted, and damage identification marks are performed on them to generate secondary temperature damage feature tensors and secondary current damage feature tensors, which are then projected to the corresponding spatial areas of the baseline geometric model (such as adjacent windings and oil tank support structures) and aligned with the outer boundaries of the primary fusion tensor. By calculating the temperature gradient diffusion coefficient and current harmonic conductivity, the spatial correlation between the secondary damage features and the primary damage features is quantified (for example, a strong correlation is determined when the temperature diffusion rate is greater than 0.5°C / min and the current distortion conduction delay is less than 20ms). The feature confidence of the multi-level tensors is fused using a weighted average algorithm to construct a secondary fusion damage feature tensor containing the coupling relationship between the core area and the associated area. Similarly, the above feature extraction, spatial alignment and coupling operation process are repeated for the outer multi-level perception data to generate a multi-level fusion damage feature tensor covering the entire space of the transformer winding, forming a multi-scale feature fusion chain from the core damage area to the peripheral impact area.

[0066] The spatial coordinates of tensors at all levels are normalized using the unified coordinate system of the reference geometric model to ensure the coordinate continuity of the first-level tensor in the core area and the second- and third-level tensors in the outer periphery in the axial (height direction), radial (radius direction), and circumferential (angular direction) directions of the winding. The Laplace pyramid algorithm is then used to perform cross-level interpolation on the multi-level tensors, generating transition feature layers (such as damage gradient attenuation coefficient and characteristic conduction delay time) at the boundary region between the first- and second-level tensors to eliminate the spatial resolution differences between different levels (such as centimeter-level for the first-level tensor and decimeter-level for the second-level tensor). Finally, through a three-dimensional tensor splicing operation, the fused tensors at all levels are expanded along the physical dimensions of mechanical deformation-current distortion-temperature gradient and the spatiotemporal dimensions of time domain-frequency domain-space domain to construct a three-dimensional damage feature tensor containing L spatial layers, M physical feature dimensions, and N time series. This tensor can fully characterize the evolution path of winding damage from local defects to global risks, providing a three-dimensional data model with multi-domain feature fusion for subsequent damage verification and positioning.

[0067] In one possible implementation, step S530 further includes:

[0068] Step S531: Calculate the temperature gradient based on the first-level temperature field sensing data to obtain the first-level temperature field distribution.

[0069] Step S532: Preset a temperature damage gradient feature, and use the temperature damage gradient feature to traverse the first-level temperature field distribution to perform damage marking, thereby obtaining a plurality of temperature damage feature points.

[0070] Step S533: spatially connecting the multiple temperature damage feature points to obtain the first-level temperature damage feature tensor.

[0071] Step S534: Similarly, the harmonic energy ratio calculation and high-frequency energy proportion area marking are performed on the primary current spectrum sensing data, and the primary current damage feature tensor is output.

[0072] Step S535: Similarly, the first-level oil chromatogram sensing data is marked with chromatogram damage, and the first-level chromatogram damage feature tensor is output.

[0073] Specifically, the high-density temperature data (sampling interval ≤ 1 second, spatial resolution ≤ 2 cm) collected by the fiber optic temperature sensor in the fault risk space is used to calculate the temperature change rate of each temperature measuring point in the directions of the x, y, and z coordinate axes through the finite difference method or Gaussian kernel smoothing algorithm, and the temperature gradient modulus and direction of each point are obtained through vector synthesis; the calculation results are mapped to the corresponding spatial grid of the reference geometric model to generate a three-dimensional temperature field distribution matrix containing temperature values, gradient vectors and spatial coordinates, which intuitively presents the temperature distribution morphology and heat conduction trend of the core area, providing basic thermal characteristic data for subsequent damage marking.

[0074] A threshold for determining temperature damage gradient characteristics is preset. Based on the primary temperature field distribution matrix, a sliding window is used to traverse the spatial grid point by point. For each grid point, the temperature gradient modulus and directional consistency with adjacent points are calculated. When the gradient modulus exceeds the threshold and the gradient direction deviation of adjacent points is within the allowable range, it is marked as an initial suspected damage point. Further, combined with time series data, points with gradient modulus values ​​that continuously exceed the threshold and show an increasing trend over multiple consecutive sampling periods are screened and confirmed as temperature damage feature points, forming a discrete feature point set containing spatial coordinates, gradient amplitude, timestamp, and damage confidence.

[0075] The Delaunay triangulation algorithm is used to spatially connect discrete temperature damage feature points. First, a Voronoi diagram is constructed in three-dimensional space, and the convex polyhedron formed by each feature point and its adjacent points is triangulated to generate a three-dimensional mesh containing the damage area boundary, gradient direction and severity. By calculating the angle between the normal vectors of adjacent triangular facets, the boundary of the area with sudden temperature gradient changes is identified to form a damage characteristic surface. The time dimension data is integrated into the three-dimensional mesh, and the timestamp and damage development rate attributes are added to each node to construct a four-dimensional tensor structure. Finally, the high-dimensional features are mapped to the preset reference geometric model coordinate system through tensor decomposition to form a first-level temperature damage feature tensor containing spatial coordinates (x, y, z), time (t), temperature gradient (▽T) and damage confidence (C), thereby realizing the mathematical representation of the spatiotemporal evolution characteristics of winding thermal damage.

[0076] Similarly, the harmonic energy ratio of the primary current spectrum sensing data is calculated, the amplitude and phase information of each frequency component is extracted, and the energy ratio of the 3rd, 5th, and 7th harmonics to the fundamental wave is calculated. At the same time, the areas where the high-frequency (such as >10kHz) energy ratio exceeds the preset threshold are marked, and the spatial points with significant current distortion are mapped to the reference geometric model to form a multi-dimensional feature matrix containing spatial coordinates, frequency distribution, harmonic energy ratio and high-frequency energy ratio. After tensor decomposition and reconstruction, the primary current damage feature tensor is output.

[0077] Similarly, the first-level oil chromatographic sensing data is marked for chromatographic damage. Based on the characteristic gas concentration threshold (such as H2, C2H2, CH4, etc. exceeding the warning value) and the gas production rate (such as the growth rate per unit time), the abnormal area of ​​the insulating oil decomposition product is mapped to the three-dimensional space of the oil circuit, and a feature tensor containing gas composition, concentration value, growth rate and spatial coordinates is constructed. Through time series analysis and spatial interpolation, the first-level chromatographic damage feature tensor is output.

[0078] In one possible implementation, step S600 further includes:

[0079] Step S610: configuring multi-level slicing scales and multi-level damage confidence weights for the multi-level winding space.

[0080] Step S620: Slice the three-dimensional damage feature tensor layer by layer along the multi-level winding space according to the multi-level slicing scale to obtain a multi-level slicing damage feature set, wherein the slicing damage features include temperature damage slicing features, current damage slicing features, and chromatographic damage slicing features.

[0081] Step S630: obtaining a multi-level damage feature matrix by vectorizing the multi-level slice damage feature set.

[0082] Step S640: using a damage feature vector library to traverse the multi-level damage feature matrix, and matching and outputting a multi-level damage intensity matrix.

[0083] Step S650: performing linear weighted integration on the multi-level damage intensity matrix according to the multi-level damage confidence weights to obtain M winding damage quantitative features as the real-time damage alarm output.

[0084] Specifically, differentiated slicing scales and damage confidence weights are configured for different levels of the multi-level winding space, based on the risk level and monitoring priority. For core damage areas (such as winding deformation areas), a millimeter-level fine slicing scale (e.g., 1-5mm) and a high confidence weight (e.g., 0.6-0.8) are set to ensure that subtle damage features are not missed. For directly affected areas (such as adjacent windings), a centimeter-level slicing scale (e.g., 1-2cm) and a medium confidence weight (e.g., 0.3-0.5) are used. For indirectly affected areas (such as the tank wall), a centimeter-level coarse slicing scale (e.g., 5-10cm) and a low confidence weight (e.g., 0.1-0.2) are set. This configuration achieves a balance between monitoring accuracy and computational efficiency, while ensuring that damage features in high-risk areas are given higher weight in subsequent quantification.

[0085] Based on the configured multi-level slicing scale, the three-dimensional damage feature tensor is sliced ​​layer by layer along the axial, radial, and circumferential dimensions of the winding space. Millimeter-level slicing (e.g., 5mm layer thickness) is used for high-risk areas, centimeter-level slicing (e.g., 1cm layer thickness) is used for medium-risk areas, and a coarser slicing scale (e.g., 5cm layer thickness) is used for low-risk areas. Each slice extracts temperature damage features (e.g., gradient amplitude and direction), current damage features (e.g., harmonic energy ratio, high-frequency component proportion), and chromatographic damage features (e.g., characteristic gas concentration, growth rate), forming a multi-dimensional feature vector containing spatial coordinates, physical eigenvalues, and damage confidence. This layered slicing strategy effectively reduces data redundancy while ensuring feature resolution in key areas, generating a multi-level slice damage feature set covering the entire winding space and providing multi-dimensional, multi-scale feature input for subsequent damage quantification.

[0086] The multidimensional features in the multi-level slice damage feature set are structured and converted. Key parameters (such as temperature gradient modulus, third harmonic energy ratio, hydrogen concentration growth rate, etc.) are extracted for the temperature damage slice feature, current damage slice feature, and chromatographic damage slice feature of each slice layer. These parameters are arranged in a preset order to form a one-dimensional feature vector. All slice layer feature vectors at the same level are horizontally spliced ​​in spatial order to form a two-dimensional damage feature matrix for that level (rows represent slice layers, and columns represent feature parameters). The above operations are performed on each level of the multi-level winding space in sequence, ultimately forming a three-dimensional damage feature matrix containing multiple levels (such as core layer, affected layer, and peripheral layer). This realizes the conversion from multimodal discrete features to structured matrix data, providing a standardized input format for subsequent damage pattern matching.

[0087] Utilizing a pre-built damage feature vector library (containing feature templates for typical mechanical damage, insulation aging, partial discharge, and other fault modes), the multi-level damage feature matrix is ​​traversed and matched row by row using the cosine similarity algorithm. For each slice layer's feature vector, its similarity score with each type of damage pattern template in the vector library is calculated, and the top K patterns with the highest similarity are selected as candidates (e.g., K = 3). A weighted voting mechanism is used to determine the damage type and confidence level for that slice layer (the weights are based on the similarity score and the prior probability of the templates in the vector library), generating a three-dimensional vector containing the damage type code, severity index, and confidence level. The matching results for all slice layers at the same level are arranged by spatial position to form a damage intensity matrix for that level. Each level of winding space is processed sequentially, ultimately outputting a multi-level damage intensity matrix containing the distribution of multi-level damage types, severity, and confidence levels, providing multi-scale, multi-dimensional damage feature matching results for subsequent damage quantification.

[0088] Based on preset multi-level damage confidence weights (e.g., a core area weight of 0.5, a directly affected area weight of 0.3, and an indirect affected area weight of 0.2), a linear weighted operation is performed on the damage severity index and confidence level in the multi-level damage intensity matrix. The damage intensity matrices for each level of the winding space are superimposed and integrated according to the hierarchical weights (e.g., core area damage intensity × 0.5 + directly affected area × 0.3 + indirect affected area × 0.2), compressing the multidimensional damage characteristics into M quantitative indicators (e.g., comprehensive damage index, characteristic diffusion rate, multi-physics field coupling degree, etc.). After standardization, these quantitative characteristics are output in real-time as numerical dashboards and spatial heat maps. These visually display the overall severity of the winding damage, its specific location, and the coupled evolution trends of various physical fields (temperature, current, insulation), providing accurate fault location and risk assessment for operation and maintenance personnel.

[0089] In one possible implementation, step S310 further includes:

[0090] Step S311: Calculating multiple node energy offset rates of the multiple deployed nodes according to the frequency band energy offset rate of the vibration frequency domain energy distribution and the preset frequency domain distribution.

[0091] Step S312: Arrange the energy offset rates of the multiple nodes in ascending order to select the first K deployed nodes as K candidate attenuation starting points.

[0092] Step S313: performing linear attenuation reverse tracing on the K candidate attenuation starting points to locate K internal vibration attenuation starting points.

[0093] Step S314: After clustering the K internal vibration attenuation starting points based on spatial proximity, the fault risk space is defined by spatially connecting the clustering results.

[0094] Specifically, the real-time vibration frequency domain energy distribution transmitted by the vibration sensor array is compared with the preset normal frequency domain distribution in each frequency band, and the energy offset rate of each deployed node in each target frequency band (such as the 100Hz-500Hz mechanical vibration sensitive frequency band) is calculated. The specific formula is: Node energy offset rate = ×100%. Through this calculation, the abnormal degree of vibration energy of each node is quantified, forming a multidimensional data set containing node number, frequency band information and offset rate value, providing a data basis for subsequent abnormal node screening and vibration source tracing.

[0095] The calculated energy offset rates of multiple nodes are sorted in ascending order from smallest to largest. Nodes with smaller energy offset rates indicate a closer-to-normal vibration energy distribution and are more likely to be located near the initial conduction path of the vibration source. The first K nodes (e.g., K = 3 or 5, depending on the density of the vibration sensor array) are selected from the sorted results as candidate attenuation starting points. These nodes serve as initial monitoring points for vibration energy conduction and are subsequently used to trace the vibration source location within the transformer. By focusing on nodes with low offset rates, the accuracy and efficiency of vibration source location are improved.

[0096] For the K candidate attenuation starting points that have been screened, a reverse tracing model is established based on the linear attenuation characteristics of vibration energy in the internal medium of the transformer (for example, assuming that the energy attenuates by α decibels per 10 cm of conduction). Taking the real-time energy offset rate of each candidate node as the starting point, the initial value position when the vibration energy has not attenuated is calculated by linear extrapolation along the reverse direction of the vibration conduction path (i.e., from the outer shell to the inside of the winding), and the corresponding K internal vibration attenuation starting points are located. For example, if the energy offset rate of a candidate node is 5%, based on the preset attenuation coefficient, it is inferred that the original excitation point 15 cm inside is the energy without attenuation, and this position is marked as an internal vibration attenuation starting point. Through this process, the vibration monitoring data on the outer shell surface is mapped to the potential damage area of ​​the transformer winding, providing an internal coordinate reference for subsequent spatial clustering.

[0097] A spatial clustering algorithm (such as DBSCAN) is used to analyze the three-dimensional coordinates (x, y, z) of K internal vibration attenuation starting points. A spatial neighborhood radius ε (e.g., 10 cm) and a minimum sample number (MinPts) (e.g., 2) are set. Starting points with a distance less than ε are grouped together to identify closely related vibration source clusters. For each cluster, boundary points are connected using a convex hull algorithm or a minimum bounding box algorithm to form a closed spatial geometric figure (such as an ellipsoid or polyhedron) as the fault risk space. This space intuitively reflects the possible range of winding damage, with its boundaries determined by the spatial distribution of the clustering results. This enables a mapping from discrete vibration source points to continuous risk areas, providing precise spatial targeting for subsequent hierarchical activation and data acquisition of the multimodal sensing array.

[0098] Embodiment 2 is based on the same inventive concept as the transformer winding damage assessment method in the above embodiment. Figure 2 As shown, the present application provides a transformer winding damage assessment system. The system and method embodiments in the present application are based on the same inventive concept. The system includes:

[0099] The vibration sensor array configuration module 10 is used to configure a vibration sensor array on the transformer housing according to a preset vibration sensor density.

[0100] The multimodal sensing array deployment module 20 is used to deploy a multimodal sensing array according to the structural characteristics of the transformer winding, wherein the multimodal sensing array includes an optical fiber temperature sensing array, a high-frequency current mutual inductance array and an online color spectrum sensing array.

[0101] The array activation module 30 is used to activate the optical fiber temperature sensor array, high-frequency current mutual inductance array and online color spectrum sensor array in a graded manner according to the frequency band energy deviation rate when the vibration frequency domain energy distribution transmitted back by the vibration sensor array deviates from the preset frequency domain distribution.

[0102] The multi-source heterogeneous sensing data acquisition module 40 is used to receive the multi-source heterogeneous sensing data collected and transmitted back by the optical fiber temperature sensor array, the high-frequency current mutual inductance array and the online chromatogram sensor array.

[0103] The feature coupling module 50 is used to construct a three-dimensional damage feature tensor by feature coupling the multi-source heterogeneous perception data after spatiotemporally aligning the multi-source heterogeneous perception data using the transformer winding as a spatial constraint.

[0104] The real-time damage alarm output module 60 is used to perform damage verification and positioning based on the three-dimensional damage feature tensor and output a real-time damage alarm.

[0105] Furthermore, the system is also used to implement the following functions:

[0106] The transformer housing is gridded according to the vibration sensing density to locate multiple deployment nodes; a plurality of vibration sensitivities of the plurality of deployment nodes are obtained by performing vibration response on the transformer winding; gain equalization matching is performed according to the plurality of vibration sensitivities, and vibration sensors are called and installed according to the matching results to complete the configuration of the vibration sensing array.

[0107] Furthermore, the system is also used to implement the following functions:

[0108] Based on the frequency band energy offset rate between the vibration frequency domain energy distribution and the preset frequency domain distribution, vibration conduction attenuation prediction is performed with the multiple deployment nodes as the starting point to define the fault risk space; with the fault risk space as the starting point, the multimodal sensing array is hierarchicalized into a multi-level multimodal sub-array; the multi-level multimodal sub-arrays are hierarchically activated to perform spatial hierarchical data collection on the transformer winding, thereby obtaining the multi-source heterogeneous sensing data including a multi-level sensing data set.

[0109] Furthermore, the system is also used to implement the following functions:

[0110] According to the offset scale of the frequency band energy offset rate, the spatial hierarchical scale is matched; starting from the fault risk space, the transformer winding space is divided into multi-level winding spaces according to the spatial hierarchical scale; according to the spatial coverage relationship between the multi-level winding space and the multimodal sensing array, the multimodal sensing array is hierarchized into multi-level multimodal sub-arrays.

[0111] Furthermore, the system is also used to implement the following functions:

[0112] Pre-constructing a reference geometric model of the transformer winding; extracting the first-level temperature field perception data, first-level current spectrum perception data and first-level oil chromatogram perception data corresponding to the fault risk space from the multi-level perception data set; performing multimodal damage identification marking on the first-level temperature field perception data, first-level current spectrum perception data and first-level oil chromatogram perception data to obtain a first-level temperature damage feature tensor, a first-level current damage feature tensor and a first-level chromatogram damage feature tensor; performing spatial feature coupling on the first-level temperature damage feature tensor, the first-level current damage feature tensor and the first-level chromatogram damage feature tensor in the reference geometric model to construct a first-level fusion damage feature tensor; similarly, performing spatial feature coupling on the multi-level perception data set to construct a multi-level fusion damage feature tensor; spatially connecting the multi-level fusion damage feature tensors to obtain the three-dimensional damage feature tensor.

[0113] Furthermore, the system is also used to implement the following functions:

[0114] A temperature gradient is calculated based on the first-level temperature field sensing data to obtain a first-level temperature field distribution; a temperature damage gradient feature is preset, and the temperature damage gradient feature is used to traverse the first-level temperature field distribution to perform damage marking to obtain multiple temperature damage feature points; the multiple temperature damage feature points are spatially connected to obtain the first-level temperature damage feature tensor; similarly, the harmonic energy ratio is calculated and the high-frequency energy proportion area is marked on the first-level current spectrum sensing data, and the first-level current damage feature tensor is output; similarly, the chromatographic damage is marked on the first-level oil chromatogram sensing data, and the first-level chromatographic damage feature tensor is output.

[0115] Furthermore, the system is also used to implement the following functions:

[0116] The multi-level winding space is configured with a multi-level slicing scale and a multi-level damage confidence weight; according to the multi-level slicing scale, the three-dimensional damage feature tensor is sliced ​​layer by layer along the multi-level winding space to obtain a multi-level slice damage feature set, wherein the slice damage features include temperature damage slice features, current damage slice features and chromatographic damage slice features; a multi-level damage feature matrix is ​​obtained by vectorizing the multi-level slice damage feature set; a damage feature vector library is used to traverse the multi-level damage feature matrix, and a multi-level damage intensity matrix is ​​matched and output; a linear weighted integration is performed on the multi-level damage intensity matrix according to the multi-level damage confidence weight to obtain M winding damage quantitative features as the real-time damage alarm output.

[0117] Furthermore, the system is also used to implement the following functions:

[0118] Based on the frequency band energy offset rate between the vibration frequency domain energy distribution and the preset frequency domain distribution, multiple node energy offset rates of the multiple deployed nodes are calculated; the multiple node energy offset rates are arranged in ascending order to screen the top K deployed nodes as K candidate attenuation starting points; linear attenuation reverse tracing is performed on the K candidate attenuation starting points to locate K internal vibration attenuation starting points; after clustering the K internal vibration attenuation starting points based on spatial proximity, the fault risk space is framed by spatially connecting the clustering results.

[0119] Example 3, Figure 3 The schematic structural diagram of the intelligent terminal provided for the transformer winding damage assessment method of the present invention shows a block diagram of an exemplary intelligent terminal suitable for implementing the embodiments of the present invention. Figure 3 The smart terminal shown is only an example and should not limit the functions and scope of use of the embodiments of the present invention. Figure 3 As shown, the intelligent terminal includes a processor 21, a memory 22, an input device 23 and an output device 24; the number of processors 21 in the intelligent terminal can be one or more. Figure 3 Taking a processor 21 as an example, the processor 21, memory 22, input device 23 and output device 24 in the smart terminal can be connected through a bus or other means. Figure 3 The bus connection is taken as an example.

[0120] It should be noted that the order in which the embodiments of the present application are presented is for illustrative purposes only and does not necessarily represent the superiority or inferiority of the embodiments. Furthermore, the foregoing descriptions of specific embodiments of this specification are provided. Furthermore, the processes depicted in the accompanying drawings do not necessarily require the specific order or sequential sequence shown to achieve the desired results. In certain embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0121] The above description is only a preferred embodiment of the present application and is not intended to limit the present application. Any modifications, equivalent replacements, improvements, etc. made within the spirit and principles of the present application should be included in the scope of protection of the present application.

[0122] This specification and drawings are merely illustrative of the present application and are intended to cover any and all modifications, variations, combinations, or equivalents within the scope of this application. Obviously, those skilled in the art may make various modifications and variations to this application without departing from the scope of this application. Thus, this application is intended to include such modifications and variations as fall within the scope of this application and its equivalents.

Claims

1. A transformer winding damage assessment method, characterized in that: The method comprises: Arrange a vibration sensor array on the transformer housing according to a preset vibration sensor density; Deploy a multimodal sensing array based on the structural characteristics of the transformer winding, wherein the multimodal sensing array includes an optical fiber temperature sensing array, a high-frequency current mutual inductance array, and an online color spectrum sensing array; When the vibration frequency domain energy distribution transmitted back by the vibration sensing array deviates from the preset frequency domain distribution, the optical fiber temperature sensing array, the high-frequency current mutual inductance array and the online color spectrum sensing array are activated in stages according to the frequency band energy deviation rate; Receiving multi-source heterogeneous sensing data collected and transmitted back by the optical fiber temperature sensor array, the high-frequency current mutual inductance array, and the online chromatography sensor array; After aligning the multi-source heterogeneous sensing data in time and space using the transformer winding as a spatial constraint, constructing a three-dimensional damage feature tensor by feature coupling the multi-source heterogeneous sensing data; Perform damage verification and positioning based on the three-dimensional damage feature tensor, and output a real-time damage alarm; A vibration sensor array is configured on the transformer housing according to a preset vibration sensor density, including: Gridding the transformer housing according to the vibration sensing density, and locating a plurality of deployment nodes; Performing vibration response on the transformer winding to obtain a plurality of vibration sensitivities of the plurality of deployment nodes; Performing gain equalization matching according to the multiple vibration sensitivities, and calling and installing vibration sensors according to the matching results to complete the configuration of the vibration sensor array; When the vibration frequency domain energy distribution returned by the vibration sensing array deviates from the preset frequency domain distribution, the optical fiber temperature sensing array, the high-frequency current mutual inductance array and the online color spectrum sensing array are activated in stages according to the frequency band energy deviation rate, including: Based on the frequency band energy offset rate between the vibration frequency domain energy distribution and the preset frequency domain distribution, vibration conduction attenuation is predicted with the multiple deployment nodes as starting points to define the fault risk space; Taking the fault risk space as a starting point, hierarchizing the multimodal sensing array into multiple levels of multimodal sub-arrays; hierarchically activating the multi-level multimodal subarrays to perform spatial hierarchical data collection on the transformer winding to obtain the multi-source heterogeneous sensing data including a multi-level sensing data set; After the multi-source heterogeneous perception data are aligned in time and space, a three-dimensional damage feature tensor is constructed by performing feature coupling on the multi-source heterogeneous perception data, including: Pre-building a reference geometric model of the transformer winding; Extracting primary temperature field perception data, primary current spectrum perception data, and primary oil chromatogram perception data corresponding to the fault risk space from the multi-level perception data set; By performing multimodal damage identification marking on the primary temperature field sensing data, the primary current spectrum sensing data, and the primary oil chromatogram sensing data, a primary temperature damage feature tensor, a primary current damage feature tensor, and a primary chromatogram damage feature tensor are obtained; Performing spatial feature coupling on the first-level temperature damage feature tensor, the first-level current damage feature tensor, and the first-level chromatographic damage feature tensor in the reference geometric model to construct a first-level fusion damage feature tensor; Similarly, spatial feature coupling is performed on the multi-level perception data set to construct a multi-level fusion damage feature tensor; The multi-level fusion damage feature tensors are spatially connected to obtain the three-dimensional damage feature tensor.

2. The transformer winding damage assessment method according to claim 1, wherein: Taking the fault risk space as a starting point, the multimodal sensing array is hierarchically divided into multiple levels of multimodal sub-arrays. The method includes: Matching the spatial layering scale according to the offset scale of the frequency band energy offset rate; Taking the fault risk space as a starting point, dividing the transformer winding space into multi-level winding spaces according to the spatial hierarchical scale; According to the spatial coverage relationship between the multi-stage winding space and the multi-modal sensing array, the multi-modal sensing array is hierarchically divided into multi-stage multi-modal sub-arrays.

3. The transformer winding damage assessment method according to claim 1, wherein: By performing multimodal damage identification marking on the primary temperature field sensing data, the primary current spectrum sensing data, and the primary oil chromatogram sensing data, a primary temperature damage feature tensor, a primary current damage feature tensor, and a primary chromatogram damage feature tensor are obtained. The method includes: Performing temperature gradient calculation based on the first-level temperature field sensing data to obtain the first-level temperature field distribution; Preset a temperature damage gradient feature, and use the temperature damage gradient feature to traverse the first-level temperature field distribution to perform damage marking, thereby obtaining a plurality of temperature damage feature points; spatially connecting the plurality of temperature damage feature points to obtain the first-level temperature damage feature tensor; Similarly, the harmonic energy ratio is calculated and the high-frequency energy ratio area is marked on the primary current spectrum sensing data, and the primary current damage feature tensor is output; Similarly, the first-level oil chromatogram sensing data is marked with chromatogram damage, and the first-level chromatogram damage feature tensor is output.

4. The transformer winding damage assessment method according to claim 2, wherein: Performing damage verification and positioning based on the three-dimensional damage feature tensor and outputting a real-time damage alarm, the method includes: configuring multi-level slicing scales and multi-level damage confidence weights for the multi-level winding space; According to the multi-level slicing scale, the three-dimensional damage feature tensor is sliced ​​layer by layer along the multi-level winding space to obtain a multi-level slice damage feature set, wherein the slice damage feature set includes temperature damage slice features, current damage slice features, and chromatographic damage slice features; Obtaining a multi-level damage feature matrix by vectorizing the multi-level slice damage feature set; Using a damage feature vector library to traverse the multi-level damage feature matrix, matching and outputting a multi-level damage intensity matrix; The multi-level damage intensity matrix is ​​linearly weighted integrated according to the multi-level damage confidence weights to obtain M winding damage quantitative features as the real-time damage alarm output.

5. The transformer winding damage assessment method according to claim 1, wherein: Based on the frequency band energy offset rate between the vibration frequency domain energy distribution and the preset frequency domain distribution, vibration conduction attenuation prediction is performed with the multiple deployment nodes as starting points to define a fault risk space, the method comprising: Calculating a plurality of node energy offset rates of the plurality of deployed nodes according to a frequency band energy offset rate between the vibration frequency domain energy distribution and a preset frequency domain distribution; Arrange the energy offset rates of the multiple nodes in ascending order to select the top K deployed nodes as K candidate attenuation starting points; Performing linear attenuation reverse tracing on the K candidate attenuation starting points to locate K internal vibration attenuation starting points; After clustering the K internal vibration attenuation starting points based on spatial proximity, the fault risk space is defined by spatially connecting the clustering results.

6. Transformer winding damage assessment system, characterized in that, The system is used to implement the transformer winding damage assessment method according to any one of claims 1 to 5, and the system includes: A vibration sensor array configuration module, configured to configure a vibration sensor array on the transformer housing according to a preset vibration sensor density; A multimodal sensing array deployment module, configured to deploy a multimodal sensing array according to the structural characteristics of the transformer winding, wherein the multimodal sensing array includes an optical fiber temperature sensing array, a high-frequency current mutual inductance array, and an online chromatographic sensing array; An array activation module is used to activate the optical fiber temperature sensor array, high-frequency current mutual inductance array, and online color spectrum sensor array in a hierarchical manner according to the frequency band energy deviation rate when the vibration frequency domain energy distribution transmitted back by the vibration sensor array deviates from the preset frequency domain distribution; A multi-source heterogeneous sensing data acquisition module is used to receive the multi-source heterogeneous sensing data collected and transmitted back by the optical fiber temperature sensor array, the high-frequency current mutual inductance array, and the online chromatogram sensor array; a feature coupling module, configured to construct a three-dimensional damage feature tensor by performing feature coupling on the multi-source heterogeneous sensing data after spatiotemporally aligning the multi-source heterogeneous sensing data using the transformer winding as a spatial constraint; The real-time damage alarm output module is used to perform damage verification and positioning based on the three-dimensional damage feature tensor and output a real-time damage alarm.

7. An intelligent terminal, characterized in that: The intelligent terminal includes: processor; a memory for storing instructions executable by the processor; Wherein, the processor is used to execute the transformer winding damage assessment method described in any one of claims 1 to 5.

Images