A method and system for condition monitoring of oil-immersed transformers

Abstract

This invention discloses a method and system for monitoring the condition of an oil-immersed transformer, relating to the field of online transformer monitoring. The method includes: injecting nano-magnetic fluid into the transformer oil; arranging a magnetic field excitation device and an acoustic sensor array outside the oil tank; using periodic pulsed magnetic field excitation to induce magnetostrictive vibration in the nano-magnetic fluid; acquiring dual-channel signals to obtain magnetic susceptibility distribution data sequences and acoustic signal sequences; establishing a magneto-acoustic coupling characteristic spectrum of core defects; predicting core defects; generating core defect prediction results; and outputting transformer control commands based on the comparison deviation with standard core conditions. This application solves the technical problems of existing transformer core condition monitoring, which relies primarily on indirect characterization, suffers from delayed defect perception, and struggles to accurately identify and locate early minor faults. It achieves the technical effects of directly sensing changes in the core magnetic circuit, providing strong real-time defect perception, and enabling accurate identification and location of early minor faults.

Description

Technical Field

[0001] This application relates to the field of online transformer monitoring, and in particular to a method and system for monitoring the condition of an oil-immersed transformer. Background Technology

[0002] The health status of the core of an oil-immersed transformer is a core factor determining the long-term reliable operation of a power transformer and is of great significance to the safety and stability of the power grid. Currently, monitoring of transformer core defects commonly employs methods such as external vibration detection, no-load current analysis, and dissolved gas analysis in the oil. These methods indirectly determine the presence of core faults by collecting data on tank surface vibration, electrical parameters, or the composition of gases in the oil. However, existing methods are all non-contact, indirect monitoring methods. They cannot directly sense changes in the magnetic circuit near the core and are easily affected by load current, environmental vibration, and oil propagation attenuation. They are also difficult to identify early localized defects, and their location accuracy and detection sensitivity are significantly limited.

[0003] At present, transformer core condition monitoring suffers from technical problems such as relying mainly on indirect characterization, delayed defect perception, and difficulty in accurately identifying and locating early minor faults. Summary of the Invention

[0004] This application provides a method and system for monitoring the condition of an oil-immersed transformer. It employs a method involving injecting a core-shell structured nano-magnetic fluid with high magnetic susceptibility and stable dispersion into the transformer oil. A magnetic field excitation device, a magnetic field sensor array, and an acoustic sensor array are deployed outside the oil tank. The magnetic fluid is excited by a periodic pulsed magnetic field to generate magnetostrictive vibrations, while simultaneously acquiring magnetic susceptibility distribution data and vibration acoustic signals within the oil channels. Based on the dual-channel signals, a magneto-acoustic coupling characteristic spectrum of core defects is constructed. Core defects are predicted based on this spectrum, and the prediction results are compared with the standard core condition. Corresponding transformer control commands are output based on the deviation. This approach solves the technical problems of existing transformer core condition monitoring methods, which rely primarily on indirect characterization, suffer from delayed defect perception, and struggle to accurately identify and locate early, minor faults. It achieves the technical effects of directly sensing changes in the core magnetic circuit, providing real-time defect perception, and enabling accurate identification and location of early, minor faults.

[0005] This application provides a condition monitoring method for an oil-immersed transformer, comprising: injecting a nano-magnetic fluid into transformer oil, the nano-magnetic fluid containing core-shell structured magnetic particles with high magnetic susceptibility and whose surfaces are insulated to maintain dispersion stability; arranging a magnetic field excitation device and an acoustic sensor array outside the transformer tank, using periodic pulsed magnetic field excitation to induce magnetostrictive vibration in the nano-magnetic fluid, and acquiring dual-channel signal acquisition through the magnetic field sensor array and acoustic sensor array to obtain a magnetic susceptibility distribution data sequence within the oil channel and an acoustic wave signal sequence generated by the magnetic fluid vibration; establishing a magnetic-acoustic coupling feature map of core defects based on the magnetic susceptibility distribution data sequence and the acoustic wave signal sequence; predicting core defects based on the magnetic-acoustic coupling feature map, generating core defect prediction results, and outputting transformer control commands based on the comparison deviation between the core defect prediction results and the standard core condition.

[0006] In one possible implementation, the nano-magnetic fluid is excited by a periodic pulsed magnetic field to induce magnetostrictive vibrations, and the following process is performed: a Helmholtz coil is used to generate a periodic pulsed magnetic field perpendicular to the direction of the transformer core laminations; the frequency of the pulsed magnetic field is set to be lower than the magnetostrictive intrinsic frequency of the silicon steel sheets in the core and higher than the dominant frequency of the transformer operating noise, so as to preferentially excite the magnetostrictive effect of the nano-magnetic fluid; through the periodic pulsed magnetic field excitation, the nano-magnetic fluid particles dispersed in the oil undergo periodic deformation, generating detectable magnetostrictive vibrations.

[0007] In a possible implementation, the following process is performed: the duty cycle of the periodic pulsed magnetic field excitation is set to a ratio range that optimizes the signal-to-noise ratio of the magnetohydrodynamic vibration signal.

[0008] In a possible implementation, based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence, a magnetic-acoustic coupling feature map of core defects is established, and the following processing is performed: the magnetic susceptibility distribution data sequence is decomposed in the time-space domain to extract the magnetic susceptibility gradient feature vector characterizing the abnormal distribution of the magnetic field in the oil passage; the acoustic signal sequence is subjected to wavelet packet transform to extract the acoustic energy feature vector of vibration energy distribution in different frequency bands; the magnetic susceptibility gradient feature vector and the acoustic energy feature vector are fused in a multi-modal manner to construct a magnetic-acoustic coupling feature matrix characterizing the local stress and structural anomalies of the core; the magnetic-acoustic coupling feature matrix is ​​subjected to dimensionality reduction and feature enhancement processing to construct a magnetic-acoustic coupling feature map.

[0009] In a possible implementation, the magnetic susceptibility gradient feature vector and the acoustic energy feature vector are fused in a multimodal manner, and the following processing is performed: the magnetic susceptibility gradient feature vector and the acoustic energy feature vector are weighted and aligned using an attention mechanism to generate a fused feature vector; the fused feature vector is subjected to convolutional analysis to extract deep spatial correlation features and form a magnetic-acoustic coupling feature matrix; the magnetic-acoustic coupling feature matrix is ​​dimensionality reduced by principal component analysis to generate a magnetic-acoustic coupling feature map.

[0010] In a possible implementation, based on the magnetic-acoustic coupling feature map, core defect prediction is performed to generate core defect prediction results. The following processing is then performed: real-time magnetic-acoustic coupling features are extracted based on the magnetic-acoustic coupling feature map and input into a pre-trained defect classification model; the type probability distribution and defect location coordinates of the core defect are calculated through the defect classification model, and the core defect prediction results are generated comprehensively.

[0011] In a possible implementation, the defect classification model is used to calculate the probability distribution of core defect types and the coordinates of defect locations. A core defect prediction result is then generated by combining these components, and the following processing is performed: The defect classification model is a multi-task learning model, comprising a model sharing layer, a classification task branch, and a regression task branch; the input magnetic-acoustic coupling features are encoded based on the model sharing layer to generate basic encoded features; the classification task branch calculates the probability distribution of defect types based on the basic encoded features; the regression task branch calculates the coordinates of defect locations based on the basic encoded features; and the probability distribution of defect types and the coordinates of defect locations are integrated to generate the core defect prediction result.

[0012] In a possible implementation, the following processing is performed: the acoustic sensor array is arranged in a tetrahedral topology on the four sides of the transformer tank, and each sensor is spatially symmetrically distributed to form an array topology structure capable of three-dimensional sound source localization.

[0013] In a possible implementation, the following processing is also performed: establishing a correlation model between the concentration of the nano-magnetic fluid and dielectric loss, and periodically monitoring the dielectric constant of the transformer oil; when the dielectric loss change exceeds a preset threshold according to the correlation model, the supplementary injection process is automatically triggered.

[0014] This application also provides an oil-immersed transformer condition monitoring system, comprising: a nano-magnetic fluid injection module for injecting nano-magnetic fluid into transformer oil, wherein the nano-magnetic fluid contains core-shell structured magnetic particles with high magnetic susceptibility and whose surfaces are insulated to maintain dispersion stability; a dual-channel signal acquisition module for arranging a magnetic field excitation device and an acoustic sensor array outside the transformer tank, thereby causing the nano-magnetic fluid to generate magnetostrictive vibration through periodic pulsed magnetic field excitation, and acquiring dual-channel signals through the magnetic field sensor array and the acoustic sensor array to obtain a magnetic susceptibility distribution data sequence in the oil channel and an acoustic wave signal sequence generated by the magnetic fluid vibration; a magnetic-acoustic coupling feature map establishment module for establishing a magnetic-acoustic coupling feature map of core defects based on the magnetic susceptibility distribution data sequence and the acoustic wave signal sequence; and a core defect prediction module for predicting core defects based on the magnetic-acoustic coupling feature map, generating a core defect prediction result, and outputting transformer control commands based on the comparison deviation between the core defect prediction result and the standard core condition.

[0015] This application proposes a method and system for monitoring the condition of an oil-immersed transformer. First, a nano-magnetic fluid containing core-shell magnetic particles is injected into the transformer oil. These particles exhibit high magnetic susceptibility and maintain dispersion stability after surface insulation modification. Next, a magnetic field excitation device and an acoustic sensor array are arranged outside the transformer tank. Periodic pulsed magnetic field excitation induces magnetostrictive vibration in the nano-magnetic fluid. Dual-channel signal acquisition using both the magnetic field and acoustic sensor arrays acquires the magnetic susceptibility distribution data sequence within the oil channels and the acoustic signal sequence generated by the magnetic fluid vibration. Based on these data, a magneto-acoustic coupling characteristic map of core defects is established. Finally, core defect prediction is performed based on this map, generating a predicted result. The transformer control command is output based on the deviation between the predicted result and the standard core condition. Through this process, the proposed method and system achieve the technical effects of directly sensing changes in the core magnetic circuit, providing strong real-time defect detection, and accurately identifying and locating early, minor faults. Attached Figure Description

[0016] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings of the embodiments of the present invention will be briefly described below. Flowcharts are used in this application to illustrate the operations performed by the system according to the embodiments of the present application. It should be understood that the preceding or following operations are not necessarily performed precisely in sequence. Instead, various steps can be processed in reverse order or simultaneously as needed. Furthermore, other operations can be added to these processes, or one or more steps can be removed from these processes.

[0017] Figure 1 This is a flowchart illustrating a condition monitoring method for an oil-immersed transformer provided in an embodiment of this application.

[0018] Figure 2 This is a schematic diagram of the structure of an oil-immersed transformer condition monitoring system provided in an embodiment of this application.

[0019] Figure labeling: 10 nano-magnetic fluid injection module, 20 dual-channel signal acquisition module, 30 magneto-acoustic coupling feature map establishment module, 40 iron core defect prediction module. Detailed Implementation

[0020] To further illustrate the technical means and effects of the present invention in achieving its intended purpose, the following detailed description of the specific implementation methods, structures, features, and effects of the present invention, in conjunction with the accompanying drawings and preferred embodiments, is provided below.

[0021] This application provides a method for condition monitoring of an oil-immersed transformer, such as... Figure 1 As shown, the method includes:

[0022] Step S100: Inject nano-magnetic fluid into transformer oil. The nano-magnetic fluid contains magnetic particles with a core-shell structure. The magnetic particles have high magnetic susceptibility and can maintain dispersion stability after their surfaces are insulated.

[0023] Specifically, a magnetic core of magnetite (Fe3O4) is prepared by chemical co-precipitation, and then a core-shell structure is formed by coating the particle surface with a silica or polyimide insulating layer using a sol-gel method. The particle size of the magnetic particles is controlled at 10-30 nanometers, and the magnetic susceptibility is not less than 10. -3 For each mole per cubic meter, oleic acid or silane coupling agent is used as the surface modifier. The mixture is slowly injected at a ratio of 0.5%-1% of the transformer oil volume and circulated for at least 30 minutes. This nano-magnetic fluid can maintain a long-term, uniform dispersion in the transformer oil without agglomeration or sedimentation.

[0024] Step S200: A magnetic field excitation device and an acoustic sensor array are arranged outside the transformer tank. The nano-magnetic fluid is excited by a periodic pulsed magnetic field to induce magnetostrictive vibration. Dual-channel signal acquisition is performed using both the magnetic field and acoustic sensor arrays to obtain the magnetic susceptibility distribution data sequence within the oil channel and the acoustic wave signal sequence generated by the magnetofluid vibration. The duty cycle of the periodic pulsed magnetic field excitation is set to a range that optimizes the signal-to-noise ratio of the magnetofluid vibration signal. The acoustic sensor array is arranged in a tetrahedral topology on the four sides of the transformer tank, with each sensor spatially symmetrically distributed, forming an array topology capable of three-dimensional sound source localization.

[0025] Specifically, a magnetic field excitation device is installed around the outer wall of the transformer tank. Magnetic field sensor arrays and acoustic sensor arrays are deployed at corresponding spatial positions outside the tank. The acoustic sensors are four piezoelectric ultrasonic sensors, fixed to the four sides of the tank in a tetrahedral topology to ensure spatial symmetry and meet the array geometry requirements of the three-dimensional sound source localization algorithm. The magnetic field sensors are a high-sensitivity Hall sensor array, uniformly distributed along the height and width of the tank to collect the magnetic susceptibility distribution signal within the oil passages. The magnetic field excitation device outputs a periodic pulsed magnetic field. By measuring the signal-to-noise ratio (SNR) of the vibration signal under different duty cycles, the interval with the highest SNR is selected as the working duty cycle. Under excitation, the nano-magnetic fluid undergoes magnetostrictive vibration. The magnetic field sensor array collects the spatial magnetic field changes in real time and converts them into a magnetic susceptibility distribution data sequence. The acoustic sensor array synchronously collects the vibration sound wave signal and forms a sound wave signal sequence. Both signals are collected synchronously using a unified clock.

[0026] In one possible implementation, the nano-magnetic fluid is excited by a periodic pulsed magnetic field to induce magnetostrictive vibration. Step S200 further includes step S210, which uses a Helmholtz coil to generate a periodic pulsed magnetic field perpendicular to the direction of the transformer core laminations. Specifically, two sets of coaxially parallel circular coils are selected to form a Helmholtz coil structure. The distance between the two sets of coils is equal to the effective radius of the coil. They are installed on opposite outer walls of the transformer tank. The central axis of the coil is perpendicular to the plane of the core laminations. The coil is wound with enameled copper flat wire, and the number of turns is determined according to the required magnetic field strength. A pulsed drive current is provided by a DC switching power supply in conjunction with an H-bridge inverter circuit to ensure that the magnetic field uniformity in the central region of the coil is not less than 95%, thereby forming a stable, uniform periodic pulsed magnetic field perpendicular to the direction of the core laminations in the transformer oil passage region.

[0027] Step S220: The frequency of the pulsed magnetic field is set to be lower than the magnetostrictive intrinsic frequency of the silicon steel core sheet and higher than the dominant frequency of the transformer's operating noise, so as to preferentially excite the magnetostrictive effect of the nano-magnetic fluid. Specifically, the magnetostrictive intrinsic frequency of the target transformer's silicon steel core sheet is first obtained through no-load vibration testing, and then the dominant frequency of the transformer's operating noise is obtained through noise spectrum analysis. The excitation frequency of the pulsed magnetic field is set in a specific range, making it lower than the magnetostrictive intrinsic frequency of the silicon steel core sheet itself, while being higher than the dominant frequency of the transformer's operating noise during normal operation. The specific value is determined based on the principle of being far away from the intrinsic frequency and the dominant noise frequency, so that the signal spectrum and the interference spectrum are completely separated, thereby improving the signal-to-noise ratio of the magnetofluid vibration signal.

[0028] In step S230, the periodic pulsed magnetic field excitation causes the nano-magnetic fluid particles dispersed in the oil to undergo periodic deformation, generating detectable magnetostrictive vibrations. Specifically, under the action of the periodic pulsed magnetic field with optimized frequency and direction, the magnetic moment inside the nano-magnetic fluid particles periodically flips and changes orientation with the magnetic field, causing the particles to undergo periodic stretching deformation along the magnetic field direction. The vibration frequency is consistent with the excitation frequency of the pulsed magnetic field. This deformation propagates outward through the insulating oil medium in the form of elastic waves, forming an acoustic signal that can be picked up by an acoustic sensor. The amplitude of the vibration signal collected by the acoustic sensor is directly related to the magnetic fluid concentration, magnetic susceptibility, and local magnetic field uniformity, and can be used to reflect the magnetic field distribution inside the oil passage and the state of the iron core structure.

[0029] Step S300: Based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence, establish a magnetic-acoustic coupling characteristic map of the core defect.

[0030] Specifically, the magnetic susceptibility distribution data sequence is smoothed over time and reconstructed by spatial interpolation to remove power frequency interference and random noise. The acoustic signal sequence is bandpass filtered and denoised to retain the effective frequency band signals related to magnetostrictive vibration. The two signals are synchronized on the time and space axes. Through feature extraction, multimodal fusion, dimensionality reduction and enhancement, a magneto-acoustic coupling feature map that can simultaneously reflect abnormal states such as local stress concentration, structural deformation and local short circuit in the iron core is constructed. Different gray-scale or color regions in the map correspond to the magneto-acoustic coupling feature intensity at different positions in the iron core, which can intuitively reflect the defect location and degree of abnormality.

[0031] In one possible implementation, based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence, a magnetic-acoustic coupling feature map of core defects is established. Step S300 further includes step S310, which decomposes the magnetic susceptibility distribution data sequence into a time-space domain and extracts a magnetic susceptibility gradient feature vector characterizing the abnormal distribution of the magnetic field within the oil passage. Specifically, a three-dimensional spatial interpolation algorithm is used to reconstruct the discrete magnetic susceptibility data collected by the magnetic field sensor array into a grid, forming a three-dimensional magnetic susceptibility distribution field in the oil passage region inside the transformer. In the spatial domain, the rate of change of magnetic susceptibility between adjacent grid points is calculated to obtain the magnetic susceptibility gradient value. In the time domain, the gradient distributions at each sampling time are sequentially stitched together to form a high-dimensional magnetic susceptibility gradient feature vector. The dimension of the feature vector corresponds to the number of spatial grids and the number of sampling points, and is used to characterize the degree of spatial distortion of the magnetic field.

[0032] Step S320: Perform wavelet packet transform on the acoustic signal sequence to extract acoustic energy feature vectors representing the vibration energy distribution in different frequency bands. Specifically, wavelet basis functions are selected to perform multi-level wavelet packet decomposition on the acoustic signal acquired by the acoustic sensor array, dividing the signal into multiple non-overlapping frequency sub-bands. The signal energy within each frequency sub-band is calculated and normalized. The energy of each sub-band is combined in frequency band order to form an acoustic energy feature vector. The value of each element in the vector corresponds to the magnitude of vibration energy within the corresponding frequency band. Different defect locations and types will lead to significant differences in energy distribution.

[0033] Step S330 involves multimodal fusion of the magnetic susceptibility gradient feature vector and the acoustic energy feature vector to construct a magneto-acoustic coupling feature matrix characterizing local stress and structural anomalies in the iron core. Specifically, the magnetic susceptibility gradient feature vector and the acoustic energy feature vector are jointly characterized at the feature layer. Through feature weighting, spatial alignment, and correlation encoding, the two heterogeneous features are mapped to the same feature space, forming a two-dimensional feature matrix. The rows of the matrix correspond to spatial locations or time series, and the columns correspond to different feature dimensions, achieving deep coupling between magnetic field information and acoustic wave information. This allows the feature matrix to simultaneously contain the correlation information between magnetic circuit anomalies and vibration anomalies, improving the accuracy and robustness of defect identification.

[0034] Step S340 involves dimensionality reduction and feature enhancement processing of the magnetic-acoustic coupling feature matrix to construct a magnetic-acoustic coupling feature map. Specifically, a combination of linear dimensionality reduction and nonlinear feature enhancement algorithms is used to remove redundant information and interference components from the magnetic-acoustic coupling feature matrix, highlighting feature components highly correlated with core defects. The processed two-dimensional feature matrix is ​​then converted into a visual image to generate a magnetic-acoustic coupling feature map. This map possesses spatial location correspondence and feature intensity characterization capabilities, and can be directly used as input to a defect classification model, enabling intuitive expression and quantitative identification of defects.

[0035] In one possible implementation, the magnetic susceptibility gradient feature vector and the acoustic energy feature vector are fused in a multimodal manner. Step S330 further includes step S331, which uses an attention mechanism to weight and align the magnetic susceptibility gradient feature vector and the acoustic energy feature vector to generate a fused feature vector. Specifically, a feature weighting network cascaded with CBAM channel attention and spatial attention is constructed: First, the magnetic susceptibility gradient feature vector and the acoustic energy feature vector are reshaped into two-dimensional feature maps and concatenated along the channel dimension. The channel attention module learns the channel weights through global average pooling and global max pooling, and then through two fully connected layers. The spatial attention module learns the spatial weights through channel dimension mean and maximum pooling, and then through a 3×3 convolutional layer. The two types of weights are multiplied sequentially with the input features to achieve defect-sensitive feature enhancement and noise suppression. Finally, the weighted features are flattened into a one-dimensional fused feature vector.

[0036] Step S332 involves performing convolutional analysis on the fused feature vector to extract deep spatial correlation features, forming a magnetic-acoustic coupling feature matrix. Specifically, a one-dimensional convolutional neural network is used to scan the local receptive field of the fused feature vector, and multi-layer convolutional kernels are used to extract spatial correlation features at different scales. A non-linear activation function is used, and the output features are reshaped to form a two-dimensional magnetic-acoustic coupling feature matrix. Each element in the matrix corresponds to the coupling response value between a specific spatial location and the feature dimension, which is used to characterize the spatial distribution pattern of anomalies inside the iron core.

[0037] Step S333: Dimensionality reduction of the magnetic-acoustic coupling feature matrix is ​​performed using principal component analysis to generate a magnetic-acoustic coupling feature map. Specifically, principal component analysis is used to calculate the eigenvalues ​​and eigenvectors of the magnetic-acoustic coupling feature matrix. Principal components with a cumulative contribution rate of over 95% are selected to map the high-dimensional feature matrix to a low-dimensional space, achieving data dimensionality reduction and redundant information removal. The dimensionality-reduced feature matrix is ​​then subjected to grayscale stretching or color mapping to generate a two-dimensional visualized magnetic-acoustic coupling feature map. Highlighted areas in the map correspond to abnormal magnetic-acoustic coupling response areas, directly corresponding to the location of core defects.

[0038] Step S400: Based on the magnetic-acoustic coupling feature spectrum, perform core defect prediction, generate core defect prediction results, and output transformer control commands based on the comparison deviation between the core defect prediction results and the standard core condition.

[0039] Specifically, the magnetic-acoustic coupling feature map is input into a pre-trained defect classification model. The model encodes, classifies, and locates the map features, and outputs whether the core has defects, the specific type of defects, the spatial coordinates of the defects, and the confidence level. The difference between this result and the preset standard core state features is calculated. When the deviation exceeds the set threshold, it is determined that the core is abnormal, and control commands including load reduction operation, enhanced oil circulation monitoring, shutdown maintenance, and local fixed-point detection are output.

[0040] In one possible implementation, based on the magnetic-acoustic coupling feature map, core defect prediction is performed to generate core defect prediction results. Step S400 further includes step S410, extracting real-time magnetic-acoustic coupling features based on the magnetic-acoustic coupling feature map and inputting them into a pre-trained defect classification model. Specifically, the magnetic-acoustic coupling feature map is normalized to a value range of 0 to 1, keeping the map size consistent with the model input requirements. The normalized map data is input into the pre-trained deep learning defect classification model in pixel matrix form. The model input layer size strictly matches the map matrix size to ensure that real-time features can be correctly received by the model and used for forward inference.

[0041] Step S420: Using the defect classification model, the probability distribution of core defect types and the coordinates of defect locations are calculated, and a core defect prediction result is generated. Specifically, the defect classification model simultaneously performs classification and regression tasks. The classification branch outputs probability values ​​for multiple preset defect types, and the regression branch outputs the coordinates of the defects in the three-dimensional space inside the transformer. The defect type with the highest probability, its corresponding confidence level, and the three-dimensional coordinates are integrated to form a complete defect prediction result that includes defect type, confidence level, and spatial location.

[0042] In one possible implementation, the defect classification model calculates the probability distribution of the core defect type and the defect location coordinates, and comprehensively generates the core defect prediction result. Step S420 further includes step S421, where the defect classification model is a multi-task learning model, comprising a model sharing layer, a classification task branch, and a regression task branch. Specifically, the multi-task learning model consists of a shared convolutional feature extraction layer, a fully connected branch for the classification task, and a fully connected branch for the regression task. The model sharing layer is used to extract the general low-level features of the magneto-acoustic coupling feature map. The two branches share the feature encoding results and respectively complete the classification and regression tasks. During the training phase, the model simultaneously optimizes the classification loss and the regression loss, and during the inference phase, it simultaneously outputs the classification result and the location result.

[0043] Step S422: Based on the model shared layer, the input magneto-acoustic coupling features are fundamentally encoded to generate fundamental encoded features. Specifically, the model shared layer consists of multiple convolutional layers, pooling layers, and normalization layers. Convolutional operations extract local features such as edges, textures, and gradients from the map; pooling layers reduce the feature dimensionality; and batch normalization stabilizes the feature distribution, ultimately outputting fundamental encoded features of fixed dimensions. These features simultaneously contain coupling information of magnetic field anomalies and vibration anomalies, which are used in classification and regression branches.

[0044] Step S423: The classification task branch calculates the probability distribution of defect types based on the basic coding features. Specifically, the classification task branch consists of multiple fully connected layers and activation functions, mapping the basic coding features to output neurons corresponding to the defect types. The output is transformed into a probability distribution through a normalized exponential function. The output results include probability values ​​for various types such as normal, loose core, local short circuit, lamination deformation, and corrosion.

[0045] Step S424: The regression task branch calculates the defect location coordinates based on the basic coding features. Specifically, the regression task branch consists of multiple fully connected layers and a linear activation function, mapping the basic coding features into three continuous numerical outputs, corresponding to the horizontal, vertical, and height coordinates of the transformer's internal spatial coordinate system, respectively. The output values ​​directly correspond to the actual spatial location of the defect inside the tank.

[0046] Step S425: Integrate the defect type probability distribution and defect location coordinates to generate a core defect prediction result. Specifically, select the defect type with the highest output probability value from the classification task branch as the identification result, use the corresponding probability as the confidence level, and use the three-dimensional coordinates output from the regression task branch as the defect location. Combine the above information into a structured output result as the final defect prediction result.

[0047] In one possible implementation, the method further includes step S500: establishing a correlation model between the concentration of the nano-magnetic fluid and dielectric loss, and periodically monitoring the dielectric constant of the transformer oil. Specifically, the dielectric loss tangent of the transformer oil at different nano-magnetic fluid concentrations is tested through indoor experiments, and a polynomial correlation model between the concentration and dielectric parameters is fitted. An online dielectric parameter sensor is installed in the transformer oil circulation pipeline to automatically collect dielectric loss data at fixed time intervals, which is used to determine the attenuation, sedimentation, or failure state of the magnetic fluid in the oil.

[0048] Step S600: When the dielectric loss change exceeds a preset threshold as determined by the correlation model, a supplementary injection process is automatically triggered. Specifically, the real-time monitored dielectric loss is substituted into the correlation model to obtain the current equivalent magnetic fluid concentration. The deviation between the current concentration and the initial set concentration is calculated. When the dielectric loss rises above the initial value by more than the preset threshold, the system drives the automatic injection device to start, injecting nano-magnetic fluid into the transformer oil at a set flow rate until the dielectric loss returns to the normal range, thus ensuring that the concentration of the monitoring magnetic fluid is maintained within the effective working range.

[0049] This application employs a method of injecting a core-shell structured nano-magnetic fluid with high magnetic susceptibility and stable dispersion into transformer oil after insulation modification. A magnetic field excitation device, a magnetic field sensor array, and an acoustic sensor array are deployed outside the oil tank. The magnetic fluid is excited by a periodic pulsed magnetic field to generate magnetostrictive vibration, while simultaneously acquiring magnetic susceptibility distribution data and vibration acoustic signals within the oil channels. Based on the dual-channel signals, a magneto-acoustic coupling characteristic spectrum of core defects is constructed. Core defects are predicted based on this spectrum, and the prediction results are compared with the standard core condition. Corresponding transformer control commands are output based on the deviation. This approach solves the technical problems of existing transformer core condition monitoring, which relies primarily on indirect characterization, suffers from delayed defect perception, and struggles to accurately identify and locate early minor faults. It achieves the technical effects of directly sensing changes in the core magnetic circuit, providing real-time defect perception, and enabling accurate identification and location of early minor faults.

[0050] In the above text, refer to Figure 1 A method for monitoring the condition of an oil-immersed transformer according to an embodiment of the present invention has been described in detail. Next, reference will be made to... Figure 2 A condition monitoring system for an oil-immersed transformer according to an embodiment of the present invention is described.

[0051] An oil-immersed transformer condition monitoring system according to an embodiment of the present invention addresses the technical problems of existing transformer core condition monitoring, which relies primarily on indirect characterization, suffers from delayed defect detection, and struggles to accurately identify and locate early-stage minor faults. The system achieves the technical effects of directly sensing changes in the core magnetic circuit, providing strong real-time defect detection, and enabling accurate identification and location of early-stage minor faults. The oil-immersed transformer condition monitoring system includes: a nano-magnetic fluid injection module 10, a dual-channel signal acquisition module 20, a magneto-acoustic coupling feature map establishment module 30, and a core defect prediction module 40.

[0052] The module 10 injects nano-magnetic fluid into transformer oil. The nano-magnetic fluid contains core-shell magnetic particles with high magnetic susceptibility and maintains dispersion stability after surface insulation modification. The dual-channel signal acquisition module 20 arranges a magnetic field excitation device and an acoustic sensor array outside the transformer tank. It induces magnetostrictive vibration in the nano-magnetic fluid through periodic pulsed magnetic field excitation and acquires dual-channel signals through the magnetic field sensor array and acoustic sensor array to obtain the magnetic susceptibility distribution data sequence in the oil channel and the acoustic wave signal sequence generated by the magnetic fluid vibration. The magnetic-acoustic coupling feature map establishment module 30 establishes a magnetic-acoustic coupling feature map of core defects based on the magnetic susceptibility distribution data sequence and the acoustic wave signal sequence. The core defect prediction module 40 predicts core defects based on the magnetic-acoustic coupling feature map, generates core defect prediction results, and outputs transformer control commands based on the deviation between the core defect prediction results and the standard core condition.

[0053] The specific configuration of the dual-channel signal acquisition module 20 is described in detail below: As mentioned above, the nano-magnetic fluid is excited by a periodic pulsed magnetic field to generate magnetostrictive vibration. The dual-channel signal acquisition module 20 may further include: a periodic pulsed magnetic field generation unit for generating a periodic pulsed magnetic field with a direction perpendicular to the direction of the transformer core laminations using a Helmholtz coil; a frequency setting unit for setting the frequency of the pulsed magnetic field to be lower than the magnetostrictive intrinsic frequency of the silicon steel sheet of the core and higher than the main frequency of the transformer operating noise, so as to preferentially excite the magnetostrictive effect of the nano-magnetic fluid; and a periodic deformation generation unit for causing the nano-magnetic fluid particles dispersed in the oil to undergo periodic deformation through the periodic pulsed magnetic field excitation, thereby generating detectable magnetostrictive vibration.

[0054] The dual-channel signal acquisition module 20 may further include: the duty cycle of the periodic pulsed magnetic field excitation is set to a ratio range that optimizes the signal-to-noise ratio of the magnetohydrodynamic vibration signal.

[0055] The detailed description of the specific configuration of the magnetic-acoustic coupling feature map establishment module 30 is as follows: As mentioned above, based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence, a magnetic-acoustic coupling feature map of the core defect is established. The magnetic-acoustic coupling feature map establishment module 30 may further include: a time-space domain decomposition unit for performing time-space domain decomposition on the magnetic susceptibility distribution data sequence to extract the magnetic susceptibility gradient feature vector characterizing the abnormal distribution of the magnetic field in the oil passage; a wavelet packet transform unit for performing wavelet packet transform on the acoustic signal sequence to extract the acoustic energy feature vector of the vibration energy distribution in different frequency bands; a multi-modal fusion unit for performing multi-modal fusion on the magnetic susceptibility gradient feature vector and the acoustic energy feature vector to construct a magnetic-acoustic coupling feature matrix characterizing the local stress and structural anomalies of the core; and a feature enhancement processing unit for performing dimensionality reduction and feature enhancement processing on the magnetic-acoustic coupling feature matrix to construct a magnetic-acoustic coupling feature map.

[0056] The method involves multimodal fusion of the magnetic susceptibility gradient feature vector and the acoustic energy feature vector. The multimodal fusion unit may further include: a weighted alignment subunit for weighted alignment of the magnetic susceptibility gradient feature vector and the acoustic energy feature vector using an attention mechanism to generate a fused feature vector; a convolutional analysis subunit for performing convolutional analysis on the fused feature vector to extract deep spatial correlation features and form a magnetic-acoustic coupling feature matrix; and a dimensionality reduction subunit for reducing the dimensionality of the magnetic-acoustic coupling feature matrix through principal component analysis to generate a magnetic-acoustic coupling feature map.

[0057] The core defect prediction module 40 is described in detail below: As mentioned above, the core defect prediction module 40 performs core defect prediction based on the magnetic-acoustic coupling feature map and generates a core defect prediction result. The core defect prediction module 40 may further include: a real-time magnetic-acoustic coupling feature extraction unit for extracting real-time magnetic-acoustic coupling features based on the magnetic-acoustic coupling feature map and inputting them into a pre-trained defect classification model; and a defect classification unit for calculating the type probability distribution and defect location coordinates of the core defect through the defect classification model and generating a comprehensive core defect prediction result.

[0058] The defect classification unit calculates the probability distribution of core defect types and the coordinates of defect locations using the defect classification model, and then generates a core defect prediction result. The defect classification unit may further include: a basic encoding subunit, where the defect classification model is a multi-task learning model containing a model sharing layer, a classification task branch, and a regression task branch, and performs basic encoding on the input magnetic-acoustic coupling features based on the model sharing layer to generate basic encoded features; a defect type probability distribution calculation subunit, where the classification task branch calculates the defect type probability distribution based on the basic encoded features; a defect location coordinate calculation subunit, where the regression task branch calculates the defect location coordinates based on the basic encoded features; and a core defect prediction result generation subunit, where the defect type probability distribution and defect location coordinates are integrated to generate a core defect prediction result.

[0059] The dual-channel signal acquisition module 20 may further include: the acoustic sensor array is arranged in a tetrahedral topology on the four sides of the transformer tank, and each sensor is spatially symmetrically distributed to form an array topology structure capable of three-dimensional sound source localization.

[0060] The system may further include: a correlation model establishment module for establishing a correlation model between the concentration of the nano-magnetic fluid and the dielectric loss, and periodically monitoring the dielectric constant of the transformer oil; and a supplementary injection process triggering module for automatically triggering the supplementary injection process when the change in dielectric loss exceeds a preset threshold based on the correlation model.

[0061] The oil-immersed transformer condition monitoring system provided in this embodiment of the invention can execute the oil-immersed transformer condition monitoring method provided in any embodiment of the invention, and has the corresponding functional modules and beneficial effects of the method.

[0062] Although this application makes various references to certain modules in the system according to the embodiments of this application, any number of different modules can be used and run on user terminals and / or servers. The various units and modules included are only divided according to functional logic, but are not limited to the above division, as long as the corresponding functions can be achieved; in addition, the specific names of each functional unit are only for easy distinction between each other and are not used to limit the scope of protection of this invention.

[0063] The above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention in any way. Although the present invention has been disclosed above with reference to preferred embodiments, it is not intended to limit the present invention. Any person skilled in the art can make some modifications or alterations to the above-disclosed technical content to create equivalent embodiments without departing from the scope of the present invention. Any modifications, equivalent changes, and alterations made to the above embodiments based on the technical essence of the present invention without departing from the scope of the present invention shall still fall within the scope of the present invention.

Claims

1. A method for condition monitoring of an oil-immersed transformer, characterized in that, The method includes: Nano-magnetic fluid is injected into transformer oil. The nano-magnetic fluid contains magnetic particles with a core-shell structure. The magnetic particles have high magnetic susceptibility and can maintain dispersion stability after their surfaces are modified with insulation. A magnetic field excitation device and an acoustic sensor array are arranged outside the transformer tank. The nano-magnetic fluid is excited by periodic pulsed magnetic field to generate magnetostrictive vibration. The magnetic field sensor array and the acoustic sensor array are used for dual-channel signal acquisition to obtain the magnetic susceptibility distribution data sequence in the oil channel and the sound wave signal sequence generated by the magnetic fluid vibration. Based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence, a magnetic-acoustic coupling characteristic spectrum of iron core defects is established; Based on the magnetic-acoustic coupling feature spectrum, core defects are predicted, core defect prediction results are generated, and transformer control commands are output based on the comparison deviation between the core defect prediction results and the standard core condition.

2. The condition monitoring method for an oil-immersed transformer as described in claim 1, characterized in that, Magnetostrictive vibrations in nanofluids are induced by periodic pulsed magnetic field excitation, including: A periodic pulsed magnetic field with a direction perpendicular to the direction of the transformer core laminations is generated using a Helmholtz coil; The frequency of the pulsed magnetic field is set to be lower than the magnetostrictive eigenfrequency of the iron core silicon steel sheet and higher than the main frequency of the transformer operating noise, so as to preferentially excite the magnetostrictive effect of the nano-magnetic fluid. The periodic pulsed magnetic field excitation causes the nano-magnetic fluid particles dispersed in the oil to undergo periodic deformation, generating detectable magnetostrictive vibrations.

3. The condition monitoring method for an oil-immersed transformer as described in claim 1, characterized in that, The duty cycle of the periodic pulsed magnetic field excitation is set to a ratio range that optimizes the signal-to-noise ratio of the magnetohydrodynamic vibration signal.

4. The condition monitoring method for an oil-immersed transformer as described in claim 1, characterized in that, Based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence, a magnetic-acoustic coupling characteristic spectrum of core defects is established, including: The magnetic susceptibility distribution data sequence is decomposed in the time-space domain to extract the magnetic susceptibility gradient feature vector that characterizes the abnormal distribution of the magnetic field in the oil passage. Wavelet packet transform is performed on the acoustic signal sequence to extract acoustic energy feature vectors of vibration energy distribution in different frequency bands; The magnetic susceptibility gradient feature vector and the acoustic energy feature vector are fused in a multi-modal manner to construct a magnetic-acoustic coupling feature matrix characterizing local stress and structural anomalies in the iron core. The magnetic-acoustic coupling feature matrix is ​​subjected to dimensionality reduction and feature enhancement processing to construct a magnetic-acoustic coupling feature map.

5. The condition monitoring method for an oil-immersed transformer as described in claim 4, characterized in that, Multimodal fusion of the magnetic susceptibility gradient feature vector and the acoustic energy feature vector includes: An attention mechanism is used to weight and align the magnetic susceptibility gradient feature vector and the acoustic energy feature vector to generate a fused feature vector; The fused feature vectors are subjected to convolutional analysis to extract deep spatial correlation features, forming a magnetic-acoustic coupling feature matrix. Principal component analysis is used to reduce the dimensionality of the magnetic-acoustic coupling feature matrix, generating a magnetic-acoustic coupling feature map.

6. The condition monitoring method for an oil-immersed transformer as described in claim 1, characterized in that, Based on the magnetic-acoustic coupling feature map, core defect prediction is performed, and core defect prediction results are generated, including: Real-time magnetic-acoustic coupling features are extracted based on the magnetic-acoustic coupling feature map and input into a pre-trained defect classification model; The defect classification model is used to calculate the probability distribution of core defect types and the coordinates of defect locations, and then a comprehensive prediction result for core defects is generated.

7. The condition monitoring method for an oil-immersed transformer as described in claim 6, characterized in that, Using the defect classification model, the probability distribution of core defect types and the coordinates of defect locations are calculated, and a comprehensive prediction result for core defects is generated, including: The defect classification model is a multi-task learning model, which includes a model sharing layer, a classification task branch, and a regression task branch. Based on the model shared layer, the input magnetic-acoustic coupling features are encoded to generate basic encoded features; Based on the basic coding features, the probability distribution of defect types is calculated from the classification task branch. The defect location coordinates are calculated based on the basic coding features from the regression task branch. By integrating the probability distribution of the defect type and the coordinates of the defect location, a core defect prediction result is generated.

8. The condition monitoring method for an oil-immersed transformer as described in claim 1, characterized in that, The acoustic sensor array is arranged in a tetrahedral topology on the four sides of the transformer tank, with each sensor spatially symmetrically distributed, forming an array topology capable of three-dimensional sound source localization.

9. The condition monitoring method for an oil-immersed transformer as described in claim 1, characterized in that, The method further includes: A correlation model between the concentration of the nano-magnetic fluid and dielectric loss was established, and the dielectric constant of the transformer oil was periodically monitored. When the dielectric loss change exceeds a preset threshold as determined by the correlation model, the supplementary injection process is automatically triggered.

10. A condition monitoring system for an oil-immersed transformer, characterized in that, The system is used to implement the condition monitoring method for an oil-immersed transformer according to any one of claims 1-9, and the system comprises: A nano-magnetic fluid injection module is used to inject nano-magnetic fluid into transformer oil. The nano-magnetic fluid contains magnetic particles with a core-shell structure. The magnetic particles have high magnetic susceptibility and can maintain dispersion stability after their surfaces are insulated. The dual-channel signal acquisition module is used to arrange a magnetic field excitation device and an acoustic sensor array outside the transformer tank. It uses periodic pulsed magnetic field excitation to induce magnetostrictive vibration in the nano-magnetic fluid, and performs dual-channel signal acquisition through the magnetic field sensor array and the acoustic sensor array to obtain the magnetic susceptibility distribution data sequence in the oil channel and the acoustic wave signal sequence generated by the magnetofluid vibration. The magneto-acoustic coupling feature map establishment module is used to establish a magneto-acoustic coupling feature map of iron core defects based on the magnetic susceptibility distribution data sequence and the acoustic signal sequence. The core defect prediction module is used to predict core defects based on the magnetic-acoustic coupling feature spectrum, generate core defect prediction results, and output transformer control commands based on the comparison deviation between the core defect prediction results and the standard core condition.

Images