Fault diagnosis method and device for transformer bushing, electronic equipment and storage medium

Abstract

The invention discloses a fault diagnosis method and device for a transformer bushing, electronic equipment and a storage medium, and belongs to the technical field of transformers, and the method comprises the steps: constructing a multivariable time sequence matrix corresponding to the transformer bushing according to the pressure signal data and characteristic gas concentration data of the transformer bushing, calculating a time delay covariance matrix corresponding to the multivariable time sequence matrix; carrying out eigenvalue decomposition on the time delay covariance matrix to obtain time sequence features corresponding to the transformer bushing, and carrying out feature fusion on the time sequence features and physical parameter features corresponding to the transformer bushing to obtain corresponding fusion features; and inputting the fusion features into a preset transformer bushing fault recognition model, and outputting a fault recognition result of the transformer bushing and a corresponding fault type, so that the problem of insufficient diagnosis sensitivity of a single-variable time sequence analysis method in the prior art can be solved.

Description

Technical Field

[0001] This invention relates to the field of transformer technology, and in particular to a method, apparatus, electronic device, and storage medium for diagnosing faults in transformer bushings. Background Technology

[0002] As a key component of the power system, the operating status of transformer bushings directly affects the safety and stability of the power grid. In recent years, with the continuous growth of power load and the accelerated construction of smart grids, the accurate identification of transformer bushing defects has become an important issue in ensuring the reliability of power transmission and transformation equipment.

[0003] However, existing bushing defect diagnosis technologies have significant shortcomings. Traditional univariate time-series analysis methods struggle to effectively capture the complex coupling effects of pressure fluctuations and gas diffusion, resulting in insufficient diagnostic sensitivity. This severely impacts the early warning capability of transformer bushing defects, increasing power grid operation risks and maintenance costs. The high false alarm rate of traditional methods fails to provide reliable data for equipment condition assessment and preventative maintenance, reducing power system operating efficiency and increasing the risk of sudden accidents, posing a potential threat to the safe and stable operation of the power grid. Summary of the Invention

[0004] This invention provides a method, apparatus, electronic device, and storage medium for diagnosing faults in transformer bushings, which can solve the problem of insufficient diagnostic sensitivity of existing single-variable time-series analysis methods.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for diagnosing faults in transformer bushings, comprising:

[0006] Acquire multivariate monitoring data of transformer bushings; wherein, the multivariate monitoring data includes: pressure signal data and characteristic gas concentration data;

[0007] Based on the pressure signal data and characteristic gas concentration data, a multivariate time series matrix corresponding to the transformer bushing is constructed, and the time delay covariance matrix corresponding to the multivariate time series matrix is ​​calculated.

[0008] The time delay covariance matrix is ​​decomposed into eigenvalues ​​to obtain the time-series features corresponding to the transformer bushing. The time-series features and the physical parameter features corresponding to the transformer bushing are then fused to obtain the corresponding fused features. The physical parameter features include porosity and activation energy.

[0009] The fused features are input into a preset transformer bushing fault identification model, so that the transformer bushing fault identification model can identify faults in the transformer bushing based on the fused features, determine whether the transformer bushing has a fault and identify the corresponding fault type, and output the fault identification result and the corresponding fault type of the transformer bushing.

[0010] As a preferred embodiment, calculating the time delay covariance matrix corresponding to the multivariate time series matrix includes:

[0011] Based on the multivariate time series matrix and the preset gas-liquid two-phase flow pressure fluctuation model, the time delay covariance matrix corresponding to the multivariate time series matrix is ​​calculated, and the time delay covariance matrix is ​​corrected according to the preset equivalent diffusion tensor.

[0012] As a preferred embodiment, the transformer bushing fault identification model is a deep belief network containing several layers of restricted Boltzmann machines;

[0013] The energy function of the restricted Boltzmann machine is:

[0014]

[0015] Among them, E l (v, h) is the energy function of the restricted Boltzmann machine, where v and h are the visible layer unit and the hidden layer unit, respectively; n v n is the number of visible layer cells; h v is the number of hidden layer units; i The state of the i-th visible layer unit; h j This represents the state of the j-th hidden layer unit; For connection weights; and γ is the bias term; Γ is the gas generation rate. g The weighting coefficients of (T, p); v g denoted as the characteristic gas concentration corresponding to the g-th visible layer unit.

[0016] As a preferred option, it also includes:

[0017] The multivariate monitoring data and the fused features are input into the transformer bushing fault identification model, and the state vector trajectory in the transformer bushing fault identification model is extracted.

[0018] Preset disturbance variables are added to the transformer bushing fault identification model, and the disturbance state vector trajectory after the disturbance variables are added to the transformer bushing fault identification model is extracted.

[0019] The state difference between the state vector trajectory and the disturbance state vector trajectory is calculated, and then the disturbance response capability of the transformer bushing fault identification model is evaluated based on the state difference.

[0020] Based on the above embodiments, another embodiment of the present invention provides a fault diagnosis device for transformer bushings, including: a multivariate acquisition module, a time delay covariance matrix construction module, a feature extraction and fusion module, and a fault diagnosis module;

[0021] The multivariable acquisition module is used to acquire multivariable monitoring data of the transformer bushing; wherein, the multivariable monitoring data includes: pressure signal data and characteristic gas concentration data;

[0022] The time delay covariance matrix construction module is used to construct a multivariate time series matrix corresponding to the transformer bushing based on the pressure signal data and the characteristic gas concentration data, and to calculate the time delay covariance matrix corresponding to the multivariate time series matrix.

[0023] The feature extraction and fusion module is used to perform eigenvalue decomposition on the time delay covariance matrix to obtain the time-series features corresponding to the transformer bushing, and to fuse the time-series features and the physical parameter features corresponding to the transformer bushing to obtain the corresponding fused features; wherein, the physical parameter features include: porosity and activation energy;

[0024] The fault diagnosis module is used to input the fused features into a preset transformer bushing fault identification model, so that the transformer bushing fault identification model can identify faults in the transformer bushing based on the fused features, determine whether the transformer bushing has a fault and identify the corresponding fault type, and output the fault identification result and the corresponding fault type of the transformer bushing.

[0025] As a preferred embodiment, calculating the time delay covariance matrix corresponding to the multivariate time series matrix includes:

[0026] Based on the multivariate time series matrix and the preset gas-liquid two-phase flow pressure fluctuation model, the time delay covariance matrix corresponding to the multivariate time series matrix is ​​calculated, and the time delay covariance matrix is ​​corrected according to the preset equivalent diffusion tensor.

[0027] As a preferred embodiment, the transformer bushing fault identification model is a deep belief network containing several layers of restricted Boltzmann machines;

[0028] The energy function of the restricted Boltzmann machine is:

[0029]

[0030] Among them, E l(v, h) is the energy function of the restricted Boltzmann machine, where v and h are the visible layer unit and the hidden layer unit, respectively; n v n is the number of visible layer cells; h v is the number of hidden layer units; i The state of the i-th visible layer unit; h j This represents the state of the j-th hidden layer unit; For connection weights; and γ is the bias term; Γ is the gas generation rate. g The weighting coefficients of (T, p); v g denoted as the characteristic gas concentration corresponding to the g-th visible layer unit.

[0031] As a preferred option, it also includes: a model perturbation response capability assessment module;

[0032] The model disturbance response capability evaluation module is used to input the multivariate monitoring data and the fused features into the transformer bushing fault identification model, and extract the state vector trajectory in the transformer bushing fault identification model.

[0033] Preset disturbance variables are added to the transformer bushing fault identification model, and the disturbance state vector trajectory after the disturbance variables are added to the transformer bushing fault identification model is extracted.

[0034] The state difference between the state vector trajectory and the disturbance state vector trajectory is calculated, and then the disturbance response capability of the transformer bushing fault identification model is evaluated based on the state difference.

[0035] Based on the above embodiments, another embodiment of the present invention provides an electronic device, the device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the transformer bushing fault diagnosis method described in the above embodiments of the invention.

[0036] Based on the above embodiments, another embodiment of the present invention provides a storage medium, the storage medium including a stored computer program, wherein, when the computer program is running, it controls the device where the storage medium is located to execute the fault diagnosis method for transformer bushings described in the above embodiments of the invention.

[0037] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0038] This invention provides a fault diagnosis method for transformer bushings, which acquires multivariate monitoring data of the transformer bushings. The multivariate monitoring data includes pressure signal data and characteristic gas concentration data. Based on the pressure signal data and characteristic gas concentration data, a multivariate time-series matrix corresponding to the transformer bushing is constructed, and the time delay covariance matrix corresponding to the multivariate time-series matrix is ​​calculated. Eigenvalue decomposition is performed on the time delay covariance matrix to obtain the time-series features corresponding to the transformer bushing. The time-series features and the physical parameter features corresponding to the transformer bushing are then fused to obtain corresponding fused features. The fused features are input into a preset transformer bushing fault identification model, so that the transformer bushing fault identification model identifies faults in the transformer bushing based on the fused features, determines whether a fault exists in the transformer bushing, identifies the corresponding fault type, and outputs the fault identification result and the corresponding fault type. Compared with the univariate time series analysis of existing technologies, this invention performs multivariate time series analysis on pressure signal data and characteristic gas concentration data, and combines it with a transformer bushing fault identification model. This effectively captures the complex coupling effect of pressure fluctuation and gas diffusion, enabling fault identification of transformer bushings and improving the accuracy and sensitivity of transformer bushing fault diagnosis. Attached Figure Description

[0039] Figure 1 This is a schematic flowchart of a fault diagnosis method for transformer bushings provided in an embodiment of the present invention;

[0040] Figure 2 This is a schematic diagram showing a comparative analysis of the diagnostic accuracy of the MTSA-DBN model;

[0041] Figure 3 This is a comparison chart of the main trends of the MTSA-DBN model;

[0042] Figure 4 This is a magnified view of the low concentration of the MTSA-DBN model;

[0043] Figure 5 This is a heatmap showing the performance differences of the MTSA-DBN model;

[0044] Figure 6 This is a performance comparison of the MTSA-DBN model under different working conditions;

[0045] Figure 7 This is a stability analysis diagram of the MTSA-DBN model;

[0046] Figure 8 This is a multidimensional performance evaluation graph of the MTSA-DBN model;

[0047] Figure 9This is a schematic diagram of the structure of a fault diagnosis device for transformer bushings provided in an embodiment of the present invention. Detailed Implementation

[0048] To make the objectives, technical solutions, and advantages of this application clearer, the technical solutions of this application will be clearly and completely described below with reference to the accompanying drawings of the embodiments. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0049] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains; the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the application; the terms “comprising” and “having”, and any variations thereof, in the specification, claims, and foregoing description of the drawings are intended to cover non-exclusive inclusion.

[0050] In the description of the embodiments of this application, technical terms such as "first" and "second" are used only to distinguish different objects and should not be construed as indicating or implying relative importance or implicitly specifying the number, specific order, or primary and secondary relationship of the indicated technical features. In the description of the embodiments of this application, "multiple" means two or more, unless otherwise explicitly defined.

[0051] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0052] In the description of the embodiments in this application, the term "and / or" is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, and B existing alone. Additionally, the character " / " in this document generally indicates that the preceding and following related objects have an "or" relationship.

[0053] In the description of the embodiments of this application, the terms "multiple" and "several" refer to two or more (including two), similarly, "multiple groups" refer to two or more (including two groups), and "multiple pieces" refer to two or more (including two pieces).

[0054] In the description of the embodiments of this application, unless otherwise expressly specified and limited, technical terms such as "installation," "connection," "joining," and "fixing" should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral part; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; they can refer to the internal communication of two components or the interaction between two components. For those skilled in the art, the specific meaning of the above terms in the embodiments of this application can be understood according to the specific circumstances.

[0055] Example 1

[0056] Please refer to Figure 1 To address the problem of insufficient diagnostic sensitivity in existing univariate time series analysis methods, an embodiment of the present invention provides a flowchart of a fault diagnosis method for transformer bushings, comprising the following specific steps:

[0057] S1. Acquire multivariate monitoring data of the transformer bushing; wherein, the multivariate monitoring data includes: pressure signal data and characteristic gas concentration data;

[0058] Specifically, the first step is to collect data from the transformer bushing monitoring system, including the pressure signal p(t) and the characteristic gas concentration C. s The historical time series data (multivariate monitoring data) of (t) were processed, and preprocessing operations such as time synchronization, missing value imputation and noise filtering were performed.

[0059] S2. Based on the pressure signal data and characteristic gas concentration data, construct the multivariate time series matrix corresponding to the transformer bushing, and calculate the time delay covariance matrix corresponding to the multivariate time series matrix;

[0060] Preferably, the step of calculating the time delay covariance matrix corresponding to the multivariate time series matrix includes: calculating the time delay covariance matrix corresponding to the multivariate time series matrix based on the multivariate time series matrix and a preset gas-liquid two-phase flow pressure fluctuation model, and correcting the time delay covariance matrix based on a preset equivalent diffusion tensor.

[0061] Specifically, after acquiring the multivariate monitoring data, a system is constructed that includes pressure p(t) and characteristic gas concentration C. g The multivariate time series matrix X(t) (where g is the gas type, such as H2 or C2H2):

[0062]

[0063] In the formula, Indicates time window T win The matrix consists of a multivariate time series matrix composed of pressure signals and the concentrations of two typical combustible gases; where tdiff and t react These are the diffusion characteristic time and the reaction characteristic time, respectively, determined by setting T. win =max(10t) diff 10t react This ensures that the signal contains complete evolutionary information.

[0064] Among them, T win The time window length is selected to meet the requirement of the feature time scale ratio Π.

[0065] T win =max{10τ diff ,10τ reac t};

[0066] In the formula, τ diff τ represents the characteristic time of gas diffusion. react This indicates the characteristic time of the cracking reaction of the insulating medium.

[0067] After constructing the multivariate time series matrix, multivariate time series analysis (MTSA) is performed:

[0068] Because the gas release process inside the bushing is accompanied by disturbance of the insulating oil, it constitutes a typical gas-liquid two-phase flow system. According to multiphase flow theory, this system exhibits weak but measurable pressure fluctuations, and its covariance structure can be identified and quantified through multivariate time series analysis (MTSA) to capture the early micro-defect evolution characteristics. Based on the gas-liquid two-phase flow pressure fluctuation model, the time delay covariance matrix ∑(τ) is calculated:

[0069]

[0070] Where τ is the time delay parameter, Let be the mean of the observed sequence, and N be the total length of the time series. To enhance the physical interpretation of the characteristic directions, an equivalent diffusion tensor D is introduced. eff Furthermore, a generalized eigenvalue decomposition method is used to correct the traditional covariance spectrum structure, ensuring that the eigenvectors reflect the coupling perturbation characteristics of the gas dispersion and liquid medium response process.

[0071] (Σ(τ)+λD eff )v i =λ i v i ;

[0072] Where λ is the regularization parameter, λ i For generalized eigenvalues, v i The eigenvectors are used to ensure that they reflect the physical characteristics of the diffusion process.

[0073] S3. Perform eigenvalue decomposition on the time delay covariance matrix to obtain the time-series features corresponding to the transformer bushing, and fuse the time-series features and the physical parameter features corresponding to the transformer bushing to obtain the corresponding fused features; wherein, the physical parameter features include: porosity and activation energy;

[0074] Specifically, after calculating the time delay covariance matrix, feature extraction and fusion are performed:

[0075] The MTSA (Multivariate Time Series Analysis) module described above first constructs a multivariate time series matrix: Secondly, a covariance matrix with time delay is constructed to reflect the lag coupling in the time series; then, an equivalent diffusion tensor is introduced to correct the covariance spectrum.

[0076] Then, eigenvalue decomposition is performed on the correction matrix to extract the principal eigenvectors as temporal evolution features. The extracted temporal features include: the main direction of gas concentration change (e.g., cracking trend), pressure-concentration coupling change patterns, and evolutionary path information before abrupt changes (crucial for early diagnosis). Then, the temporal features F extracted by MTSA are... time Physical parameters P of the insulating dielectric pyrolysis model phys (such as porosity φ, activation energy E) a The data structure features of the evolutionary process are fused with those of other data sources to obtain fused features. This fused feature retains the data structure features of the evolutionary process while embedding the physical prior knowledge of the pyrolysis behavior, achieving an effective unification of feature representation and diagnostic modeling in terms of interpretability and generalization ability. Specifically, the pyrolysis gas model of the insulating medium borrows from the material thermal diffusion equation and combines it with the gas generation rate under temperature and pressure conditions to form a coupled differential model, providing physical consistency constraints for deep network training. The resulting fused feature is F:

[0077]

[0078] Among them, W p The weight matrix for the physical parameters, This is a feature splicing operation.

[0079] S4. Input the fused features into a preset transformer bushing fault identification model so that the transformer bushing fault identification model can identify faults in the transformer bushing based on the fused features, determine whether the transformer bushing has a fault and identify the corresponding fault type, and output the fault identification result and the corresponding fault type of the transformer bushing.

[0080] Preferably, the transformer bushing fault identification model is a deep belief network containing several layers of restricted Boltzmann machines; the energy function of the restricted Boltzmann machine is:

[0081]

[0082] Among them, E l (v, h) is the energy function of the restricted Boltzmann machine, where v and h are the visible layer unit and the hidden layer unit, respectively; n v n is the number of visible layer cells; h v is the number of hidden layer units; i The state of the i-th visible layer unit; h j This represents the state of the j-th hidden layer unit; For connection weights; and γ is the bias term; Γ is the gas generation rate. g The weighting coefficients of (T,p); v g denoted as the characteristic gas concentration corresponding to the g-th visible layer unit.

[0083] Specifically, the steps for transformer fault identification include:

[0084] (1) Deep Belief Network (DBN) Structure Design:

[0085] This invention employs a deep belief network (DBN) constructed from stacked multilayer Restricted Boltzmann Machines (RBMs) to model high-order features of pressure gas-related signals. The energy function of each RBM layer incorporates physical prior design to extract latent feature representations related to early pyrolysis signs from the original multivariate time-series data. Considering the diffusion dynamics of pyrolysis gases in insulating media, the energy function of each Restricted Boltzmann Machine (RBM) layer is designed, as shown in the equation above.

[0086] This invention introduces a rate constraint term based on the diffusion dynamics of pyrolysis gas in insulating media into the RBM energy function, and constructs a physical prior-driven deep probabilistic model to ensure that the learned features are consistent with the generation mechanism and transport behavior of pyrolysis gas, thereby improving the accuracy and interpretability of anomaly detection.

[0087] (2) Model training:

[0088] CD-K (Contrastive Divergence-K-step) is an efficient approximation algorithm for training Restricted Boltzmann Machines (RBMs) to maximize the probability of training samples. The pre-training objects are the RBM modules of each layer in the DBN (i.e., the unsupervised weights from the visible layer to the hidden layer); the input data is the concatenation of temporal features and physical parameters extracted by MTSA.

[0089] The pre-training process can be described as follows: (1) Input the input feature V into the RBM and calculate the hidden layer activation value h; (2) Reconstruct the input based on h to obtain (3) Calculate the reconstruction error; (4) Compare the difference between the initial input and the reconstruction result and update the gradient; (5) Perform K-step Gibbs sampling to iteratively optimize the parameters until convergence; (6) After each layer is trained, continue training the next layer (layer-by-layer pre-training).

[0090] The contrastive divergence algorithm (CD-k) is used for layer-by-layer pre-training to minimize the corrected reconstruction error L. recon :

[0091]

[0092] In the formula, the second term is the residual of the diffusion equation of the pyrolysis gas in the insulating medium, and μ is the equilibrium parameter.

[0093] Based on pre-training, a softmax classification layer is added for supervised fine-tuning to optimize the cross-entropy loss function L. CE :

[0094]

[0095] Where, λ reg y is the regularization coefficient, W is the network weight, and y is the network weight. ic Let i be the true label of the i-th sample in class c. Let be the predicted probability of the i-th sample in class c.

[0096] (3) Model application:

[0097] The fused features are input into a deep belief network (DBN) model to complete the defect identification task and output the diagnostic results of the transformer bushing. The trained DBN (Transformer Bushing Fault Identification Model) plays the core role of fault identification reasoning in this invention. Its purpose is to classify and judge whether there are defects in the transformer bushing and their types based on the fused feature vector of time-series features and physical parameters.

[0098] Preferably, the method further includes: inputting the multivariate monitoring data and the fused features into the transformer bushing fault identification model, and extracting the state vector trajectory in the transformer bushing fault identification model; adding a preset disturbance variable to the transformer bushing fault identification model, and extracting the disturbance state vector trajectory after adding the disturbance variable; calculating the state difference between the state vector trajectory and the disturbance state vector trajectory, and then evaluating the disturbance response capability of the transformer bushing fault identification model based on the state difference.

[0099] Specifically, the present invention also includes the following steps:

[0100] S5. Diagnostic sensitivity analysis:

[0101] To evaluate the diagnostic system's response to minor disturbances (such as slight changes in gas concentration), the maximum Lyapunov exponent is further introduced to quantitatively analyze the sensitivity of the system's state trajectory to initial conditional disturbances. The Lyapunov exponent is a core indicator describing the stability of nonlinear dynamic systems, used to measure the rate at which the system amplifies or decays to initial small disturbances.

[0102] The "assessment and diagnosis system" mentioned here refers to the trained DBN model (Deep Belief Network). This system uses the fused feature vectors as input to automatically determine the state of the transformer bushing and output a classification label or probability. However, the "diagnostic sensitivity analysis" here aims not only to "classify" it, but also to further introduce the Lyapunov index to evaluate its stability and sensitivity, and to test whether the model has the ability to amplify early anomalies under small perturbations.

[0103] The steps are described as follows:

[0104] (1) Construct the state vector trajectory X(t) from the temporal features output by the deep feature model, such as gas concentration, pressure, etc.

[0105] The deep feature model refers to the fully trained DBN model, which is responsible for inputting the raw monitoring data (pressure / gas concentration) and the composite features fused by MTSA analysis and the physical model into the network to form a high-order feature vector representation, and then outputting a classification through Softmax. Furthermore, although gas concentration and pressure can be directly collected, the state vector trajectories used for Lyapunov assessment are not the original signals, but rather embedded spatial trajectories extracted from the model. Therefore, these states are not "directly collected" raw values, but rather evolutionary representation paths extracted and constructed by the deep model.

[0106] (2) Introduce a small perturbation δX(0) at a certain moment to construct adjacent initial states;

[0107] (3) Evolve the two state trajectories X(t) and X′(t), and record their difference δX(t);

[0108] (4) Calculate the perturbation growth rate λ max :

[0109]

[0110] Here, X(t) represents the pressure-gas state vector after time t. t is the evolution time (used for the Lyapunov exponent). A growth rate greater than 0 indicates the system is sensitive to disturbances and possesses the ability to amplify early anomalies; a growth rate less than 0 indicates that disturbances are absorbed by system stability, suggesting that anomalies are not easily detected. Furthermore, the essence of state evolution is: inputting an initial state (a combination of gas concentration / pressure features) into the model, performing model inference along the time step, and observing how its path changes in the feature space. To further evaluate the diagnostic system's response to early, subtle anomalies, the maximum Lyapunov exponent is introduced as a stability evaluation index. By introducing perturbations into the feature space extracted by the deep model and tracking their evolution trajectory, the system's ability to amplify perturbations is determined, thus verifying the model's sensitivity to identifying minor anomaly trends in actual operation.

[0111] S6. Interpretability Analysis:

[0112] To further enhance the interpretability and controllability of the system, a Jacobian matrix J is constructed based on the partial derivative relationships between physical variables to reveal the sensitive coupling path between pressure and characteristic gas concentration. The Jacobian matrix is ​​a first-order partial derivative describing the system's multi-input, multi-output response relationship. Its construction is as follows: Assume the system has multiple output features and multiple input variables (such as pressure, gas concentration, and pyrolysis model parameters); calculate the partial derivative of each output function with respect to each input variable to obtain the complete Jacobian matrix, whose elements represent the "sensitivity of the output response to the input variables." By constructing the Jacobian matrix for gas concentration, pressure, and physical parameters, key variable paths affecting diagnostic accuracy can be identified, significantly improving the model's physical interpretability and controllability, and providing important reference for model deployment and engineering practice.

[0113] Combining the dielectric pyrolysis model, the Jacobian matrix J is obtained:

[0114]

[0115] Among them, F i Let φ represent the i-th output function of the model (corresponding to the i-th output feature), and φ be the porosity.

[0116] S7. Integration of material damage evolution models:

[0117] This invention integrates a material damage evolution model, incorporating key variables such as temperature, gas concentration, and stress into the modeling expression of damage rate. This not only enhances the diagnostic model's ability to explain the causes of defects but also provides a quantitative basis for predicting the future deterioration trend of bushings, effectively improving the scientific rigor and practicality of transformer operating condition monitoring. The model is not a standalone module, but is embedded in the aforementioned MTSA-DBN overall diagnostic framework as parameter input and constraints. The specific integration method is as follows: (1) Variables involved in the model (such as activation energy QQQ, temperature TT, gas concentration ∑Cg\sumC_g∑Cg) can be fused into the deep network input as physical prior parameters; (2) During the DBN training phase, residual terms related to the fracture / diffusion model have been introduced into the reconstruction loss function; similarly, the prediction rate of the material damage model can be used as a constraint term to participate in the loss control during the training process, thereby strengthening the physical consistency of the deep model output with the actual damage trend; (3) During the diagnostic phase, if the DBN output is "defect exists" or "fracture trend is enhanced", the model can be mapped in reverse to estimate the current damage accumulation rate, which can be used as the input basis for subsequent risk assessment or life prediction modules.

[0118] The process of bushing insulation performance degradation can be described as follows:

[0119]

[0120] in, Let A be the damage variable, σ be the material's intrinsic constant, and σ be the definite material constant. vM For von Mises stress, σ y Where is the yield strength, m is the stress exponent, Q is the activation energy, R is the gas constant, T is the temperature, and B is the coefficient of gas concentration contributing to the damage.

[0121] In a specific embodiment, the above-described fault diagnosis process is simulated to further illustrate the advanced nature of the fault diagnosis method. The simulation environment uses the MATLAB / Simulink platform, and the hardware configuration includes an Intel Xeon Gold 6248 processor (2.5GHz), 256GB DDR4 RAM, and 1TB NVMe SSD storage. The simulation model includes key components such as a data preprocessing module, a multivariate time series analysis (MTSA) module, and a deep belief network (DBN) module, which can realistically reflect the dynamic correlation characteristics of transformer bushing pressure-gas parameters and the defect diagnosis process. The simulation parameter settings refer to the technical specifications and actual operating data of large-scale power transformer online monitoring systems to ensure that the simulation results have practical significance.

[0122] Please refer to Figure 2 This is a schematic diagram illustrating the comparative analysis of the diagnostic accuracy of the MTSA-DBN model, as shown below. Figure 2As shown, this embodiment compares the performance of the MTSA-DBN model with traditional MTSA and DBN methods in diagnosing five typical defect types. This figure, generated based on MATLAB code, uses a grouped bar chart format to enhance readability. From... Figure 2 As can be observed, the MTSA-DBN model exhibits significant advantages across all defect types. Particularly in partial discharge detection, the MTSA-DBN achieves an accuracy of 95.68%, a 15.15 percentage point improvement over the traditional MTSA and a 9.77 percentage point improvement over DBN alone. This result fully demonstrates the superiority of the proposed fusion model in extracting pressure-gas correlation features, particularly its significantly enhanced sensitivity to weak discharge characteristics.

[0123] The diagnostic sensitivity of the MTSA-DBN model under different hydrogen concentrations was further explored in the simulation environment provided in the examples. Figure 3-5 The model's performance was fully demonstrated. Figure 3 This is a main trend comparison chart of the MTSA-DBN model, showing the diagnostic sensitivity comparison of MTSA-DBN, MTSA, and DBN across the entire concentration range. Figure 4 This is a magnified low-concentration image of the MTSA-DBN model. Figure 5 This is a heatmap showing the performance differences between the MTSA-DBN model and other models. The results indicate that the diagnostic sensitivity of the MTSA-DBN model in the low concentration region (0-50 ppm) increased from 0.68 to 0.93, an increase of 36.76%, significantly higher than the other two methods. This result verifies the advantages of the proposed fusion model in early weak defect identification and solves the problem of insufficient sensitivity of traditional methods under low concentration conditions. Figure 4-5 Error analysis was further amplified in the low-concentration region (0-60 ppm). The MTSA-DBN method showed an average sensitivity of 0.81 and a standard deviation of only 0.02 within this range, demonstrating high stability and reliability. This result indicates that the MTSA-DBN model in this embodiment not only improves diagnostic sensitivity but also enhances its robustness under complex operating conditions, effectively addressing the problem of insufficient generalization ability of traditional methods when facing diverse fault modes.

[0124] To further verify the cross-dimensional feature fusion effect of the MTSA-DBN model, this embodiment conducted a multi-dimensional performance evaluation. Please refer to... Figure 6-8 , Figure 6 For the performance comparison of the MTSA-DBN model under different working conditions, Figure 7 The graph shows the stability analysis of the MTSA-DBN model. Figure 8 This is a multidimensional performance evaluation graph for the MTSA-DBN model.

[0125] Figure 6 The performance of MTSA-DBN compared with other methods under four operating conditions is presented. The results show that MTSA-DBN performs best in all conditions, with an average feature extraction efficiency of 85.79%, which is 15.92 percentage points higher than traditional methods. This result fully demonstrates that the feature fusion mechanism proposed in this invention effectively solves the challenges of cross-dimensional feature alignment and temporal correlation modeling. Figure 7 and 8 Stability analysis further confirms the advantages of MTSA-DBN. MTSA-DBN achieved an average performance of 85.19% under four operating conditions, with a standard deviation of only 0.17, while the traditional method achieved an average performance of 68.71% with a standard deviation of 1.62. This demonstrates that MTSA-DBN is not only higher performing but also more stable, providing a reliable basis for equipment condition assessment and preventative maintenance.

[0126] In summary, the simulation results of this embodiment fully verify the superiority of the MTSA-DBN fusion model in transformer bushing defect diagnosis. This model significantly improves the accuracy and diagnostic sensitivity of early defect identification by effectively capturing the complex coupling effect of pressure fluctuations and gas diffusion. Simultaneously, by extracting high-order features using deep learning technology, it overcomes the problem of insufficient generalization ability of traditional methods when facing diverse fault modes. The introduction of feature fusion mechanisms and physical constraints effectively solves the challenges of cross-dimensional feature alignment and temporal correlation modeling, improving the overall performance of the diagnostic system. These innovative solutions provide reliable technical support for transformer bushing condition monitoring and fault prevention, effectively improving the operating efficiency and safety of power systems.

[0127] Therefore, this invention provides a fault diagnosis method for transformer bushings. The MTSA-DBN fusion model of this invention achieves accurate modeling and efficient diagnosis of the complex physical processes inside the transformer bushing through the above steps. This method not only considers key physical mechanisms such as gas-liquid two-phase flow and insulation medium decomposition, but also extracts high-order features through deep learning technology, achieving an organic combination of research object modeling and advanced algorithms. This method significantly improves the accuracy and diagnostic sensitivity of early defect identification, provides reliable technical support for transformer bushing condition monitoring and fault prevention, and effectively improves the operating efficiency and safety of power systems. Through this invention, the following beneficial effects can be achieved:

[0128] (1) By combining multivariate time series analysis and deep belief network, the complex coupling effect of pressure fluctuation and gas diffusion was effectively captured, and the diagnostic sensitivity was improved.

[0129] (2) High-order features are extracted by using deep learning models, which overcomes the problem of insufficient generalization ability of traditional methods when facing diverse fault modes and improves the adaptability of the model under complex working conditions.

[0130] (3) By introducing feature fusion mechanism and physical constraints, the challenges of cross-dimensional feature alignment and temporal correlation modeling are solved, and the overall performance of the diagnostic system is improved.

[0131] (4) The introduction of diagnostic sensitivity analysis and interpretability analysis improved the accuracy of the model under low-concentration gas leakage conditions, providing a reliable basis for equipment condition assessment and preventive maintenance.

[0132] (5) By integrating the material damage evolution model, the model’s adaptability to dynamic changes in the power grid operating environment is enhanced, and the stability and reliability of the diagnostic results are improved.

[0133] Example 2

[0134] Please refer to Figure 9 This is a schematic diagram of a fault diagnosis device for transformer bushings provided in an embodiment of the present invention. The device includes: a multivariate acquisition module, a time delay covariance matrix construction module, a feature extraction and fusion module, and a fault diagnosis module.

[0135] The multivariable acquisition module is used to acquire multivariable monitoring data of the transformer bushing; wherein, the multivariable monitoring data includes: pressure signal data and characteristic gas concentration data;

[0136] The time delay covariance matrix construction module is used to construct a multivariate time series matrix corresponding to the transformer bushing based on the pressure signal data and the characteristic gas concentration data, and to calculate the time delay covariance matrix corresponding to the multivariate time series matrix.

[0137] The feature extraction and fusion module is used to perform eigenvalue decomposition on the time delay covariance matrix to obtain the time-series features corresponding to the transformer bushing, and to fuse the time-series features and the physical parameter features corresponding to the transformer bushing to obtain the corresponding fused features; wherein, the physical parameter features include: porosity and activation energy;

[0138] The fault diagnosis module is used to input the fused features into a preset transformer bushing fault identification model, so that the transformer bushing fault identification model can identify faults in the transformer bushing based on the fused features, determine whether the transformer bushing has a fault and identify the corresponding fault type, and output the fault identification result and the corresponding fault type of the transformer bushing.

[0139] Preferably, the step of calculating the time delay covariance matrix corresponding to the multivariate time series matrix includes: calculating the time delay covariance matrix corresponding to the multivariate time series matrix based on the multivariate time series matrix and a preset gas-liquid two-phase flow pressure fluctuation model, and correcting the time delay covariance matrix based on a preset equivalent diffusion tensor.

[0140] Preferably, the transformer bushing fault identification model is a deep belief network containing several layers of restricted Boltzmann machines;

[0141] The energy function of the restricted Boltzmann machine is:

[0142]

[0143] Among them, E l (v, h) is the energy function of the restricted Boltzmann machine, where v and h are the visible layer unit and the hidden layer unit, respectively; n v n is the number of visible layer cells; h v is the number of hidden layer units; i The state of the i-th visible layer unit; h j This represents the state of the j-th hidden layer unit; For connection weights; and γ is the bias term; Γ is the gas generation rate. g The weighting coefficients of (T, p); v g denoted as the characteristic gas concentration corresponding to the g-th visible layer unit.

[0144] Preferably, it also includes: a model perturbation response capability assessment module;

[0145] The model disturbance response capability evaluation module is used to input the multivariate monitoring data and the fused features into the transformer bushing fault identification model, and extract the state vector trajectory in the transformer bushing fault identification model.

[0146] Preset disturbance variables are added to the transformer bushing fault identification model, and the disturbance state vector trajectory after the disturbance variables are added to the transformer bushing fault identification model is extracted.

[0147] The state difference between the state vector trajectory and the disturbance state vector trajectory is calculated, and then the disturbance response capability of the transformer bushing fault identification model is evaluated based on the state difference.

[0148] It should be noted that the device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate, and the components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Furthermore, in the accompanying drawings of the device embodiments provided by this invention, the connection relationships between modules indicate that they have communication connections, which can be specifically implemented as one or more communication buses or signal lines. Those skilled in the art can understand and implement this without any creative effort.

[0149] Those skilled in the art will clearly understand that, for convenience and brevity, the specific working process of the device described above can be referred to the corresponding process in the foregoing method embodiments, and will not be repeated here.

[0150] Example 3

[0151] Accordingly, embodiments of the present invention provide an electronic device, the device including a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the transformer bushing fault diagnosis method described in the above embodiments of the invention.

[0152] The electronic device may be a desktop computer, laptop, handheld computer, or cloud server, etc. The device may include, but is not limited to, a processor and a memory.

[0153] The processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. A general-purpose processor can be a microprocessor or any conventional processor. The processor is the control center of the device, connecting various parts of the device via various interfaces and lines.

[0154] Example 4

[0155] Accordingly, embodiments of the present invention provide a storage medium, the storage medium including a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to execute the fault diagnosis method for transformer bushings described in the above embodiments of the invention.

[0156] The memory can be used to store the computer program. The processor implements various functions of the device by running or executing the computer program stored in the memory and calling data stored in the memory. The memory may mainly include a program storage area and a data storage area. The program storage area may store the operating system, at least one application program required for a function, etc.; the data storage area may store data created based on the use of the mobile phone, etc. In addition, the memory may include high-speed random access memory, and may also include non-volatile memory, such as hard disk, RAM, plug-in hard disk, smart media card (SMC), secure digital (SD) card, flash card, at least one disk storage device, flash memory device, or other volatile solid-state storage device.

[0157] The storage medium is a computer-readable storage medium, and the computer program is stored in the computer-readable storage medium. When executed by a processor, the computer program can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable file, or some intermediate form. The computer-readable medium can include: any entity or device capable of carrying the computer program code, recording media, USB flash drive, portable hard drive, magnetic disk, optical disk, computer memory, read-only memory (ROM), random access memory (RAM), electrical carrier signals, telecommunication signals, and software distribution media, etc. It should be noted that the content included in the computer-readable medium can be appropriately added or removed according to the requirements of legislation and patent practice in the jurisdiction. For example, in some jurisdictions, according to legislation and patent practice, the computer-readable medium does not include electrical carrier signals and telecommunication signals.

[0158] The above description represents the preferred embodiments of the present invention. It should be noted that those skilled in the art can make various improvements and modifications without departing from the principles of the present invention, and these improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. A method of diagnosing a fault of a transformer bushing, characterized by, include: Acquire multivariate monitoring data of transformer bushings; wherein, the multivariate monitoring data includes: pressure signal data and characteristic gas concentration data; Based on the pressure signal data and characteristic gas concentration data, a multivariate time series matrix corresponding to the transformer bushing is constructed, and the time delay covariance matrix corresponding to the multivariate time series matrix is ​​calculated. The time delay covariance matrix is ​​decomposed into eigenvalues ​​to obtain the time-series features corresponding to the transformer bushing. The time-series features and the physical parameter features corresponding to the transformer bushing are then fused to obtain the corresponding fused features. The physical parameter features include porosity and activation energy. The fused features are input into a preset transformer bushing fault identification model, so that the transformer bushing fault identification model can identify faults in the transformer bushing based on the fused features, determine whether the transformer bushing has a fault and identify the corresponding fault type, and output the fault identification result and the corresponding fault type of the transformer bushing.

2. The method of diagnosing a failure of a transformer bushing according to claim 1, characterized by, The calculation of the time delay covariance matrix corresponding to the multivariate time series matrix includes: Based on the multivariate time series matrix and the preset gas-liquid two-phase flow pressure fluctuation model, the time delay covariance matrix corresponding to the multivariate time series matrix is ​​calculated, and the time delay covariance matrix is ​​corrected according to the preset equivalent diffusion tensor.

3. The fault diagnosis method for transformer bushings as described in claim 1, characterized in that, The transformer bushing fault identification model is a deep belief network containing several layers of restricted Boltzmann machines; The energy function of the restricted Boltzmann machine is: Among them, E l (v,h) is the energy function of the restricted Boltzmann machine, where v and h are the visible layer unit and the hidden layer unit, respectively; n v n is the number of visible layer cells; h v is the number of hidden layer units; i The state of the i-th visible layer unit; h j This represents the state of the j-th hidden layer unit; For connection weights; and γ is the bias term; Γ is the gas generation rate. g The weighting coefficients of (T,p); v g denoted as the characteristic gas concentration corresponding to the g-th visible layer unit.

4. The fault diagnosis method for transformer bushings as described in claim 1, characterized in that, Also includes: The multivariate monitoring data and the fused features are input into the transformer bushing fault identification model, and the state vector trajectory in the transformer bushing fault identification model is extracted. Preset disturbance variables are added to the transformer bushing fault identification model, and the disturbance state vector trajectory after the disturbance variables are added to the transformer bushing fault identification model is extracted. The state difference between the state vector trajectory and the disturbance state vector trajectory is calculated, and then the disturbance response capability of the transformer bushing fault identification model is evaluated based on the state difference.

5. A fault diagnosis device for transformer bushings, characterized in that, include: The module includes a multivariate acquisition module, a time delay covariance matrix construction module, a feature extraction and fusion module, and a fault diagnosis module. The multivariable acquisition module is used to acquire multivariable monitoring data of the transformer bushing; wherein, the multivariable monitoring data includes: pressure signal data and characteristic gas concentration data; The time delay covariance matrix construction module is used to construct a multivariate time series matrix corresponding to the transformer bushing based on the pressure signal data and the characteristic gas concentration data, and to calculate the time delay covariance matrix corresponding to the multivariate time series matrix. The feature extraction and fusion module is used to perform eigenvalue decomposition on the time delay covariance matrix to obtain the time-series features corresponding to the transformer bushing, and to fuse the time-series features and the physical parameter features corresponding to the transformer bushing to obtain the corresponding fused features; wherein, the physical parameter features include: porosity and activation energy; The fault diagnosis module is used to input the fused features into a preset transformer bushing fault identification model, so that the transformer bushing fault identification model can identify faults in the transformer bushing based on the fused features, determine whether the transformer bushing has a fault and identify the corresponding fault type, and output the fault identification result and the corresponding fault type of the transformer bushing.

6. The fault diagnosis device for transformer bushings as described in claim 5, characterized in that, The calculation of the time delay covariance matrix corresponding to the multivariate time series matrix includes: Based on the multivariate time series matrix and the preset gas-liquid two-phase flow pressure fluctuation model, the time delay covariance matrix corresponding to the multivariate time series matrix is ​​calculated, and the time delay covariance matrix is ​​corrected according to the preset equivalent diffusion tensor.

7. The fault diagnosis device for transformer bushings as described in claim 5, characterized in that, The transformer bushing fault identification model is a deep belief network containing several layers of restricted Boltzmann machines; The energy function of the restricted Boltzmann machine is: Among them, E l (v,h) is the energy function of the restricted Boltzmann machine, where v and h are the visible layer unit and the hidden layer unit, respectively; n v n is the number of visible layer cells; h v is the number of hidden layer units; i The state of the i-th visible layer unit; h j This represents the state of the j-th hidden layer unit; For connection weights; and γ is the bias term; Γ is the gas generation rate. g The weighting coefficients of (T,p); v g denoted as the characteristic gas concentration corresponding to the g-th visible layer unit.

8. The fault diagnosis device for transformer bushings as described in claim 5, characterized in that, Also includes: Model perturbation response capability assessment module; The model disturbance response capability evaluation module is used to input the multivariate monitoring data and the fused features into the transformer bushing fault identification model, and extract the state vector trajectory in the transformer bushing fault identification model. Preset disturbance variables are added to the transformer bushing fault identification model, and the disturbance state vector trajectory after the disturbance variables are added to the transformer bushing fault identification model is extracted. The state difference between the state vector trajectory and the disturbance state vector trajectory is calculated, and then the disturbance response capability of the transformer bushing fault identification model is evaluated based on the state difference.

9. An electronic device, characterized in that, It includes a processor, a memory, and a computer program stored in the memory and configured to be executed by the processor, wherein the processor executes the computer program to implement the fault diagnosis method for transformer bushings as described in any one of claims 1 to 4.

10. A storage medium, characterized in that, The storage medium includes a stored computer program, wherein, when the computer program is executed, it controls the device where the storage medium is located to perform the fault diagnosis method for transformer bushings as described in any one of claims 1 to 4.

Images