A method and device for evaluating the insulation state of an insulation bushing of a converter transformer

Abstract

The application discloses a kind of converter transformer insulating bushing insulation state evaluation method and device, applied to insulating bushing technical field, by collecting the electrical parameter and environmental parameter of insulating bushing, obtain first data set, each parameter in first data set is analyzed to electric stress and thermal stress, obtain analysis result, based on analysis result, construct state evaluation index system;Obtain the historical electrical parameter and historical environmental parameter of insulating bushing, obtain second data set, based on second data set, the data of each parameter corresponding to first data set is carried out data enhancement, obtain third data set, according to each evaluation index in state evaluation index system, the weight vector of each evaluation index is obtained by weighting;According to third data set and weight vector, the insulation state of insulating bushing is evaluated using integrated algorithm, obtain evaluation result, by the above method can realize data self enhancement, avoid single model deviation, improve the accuracy of evaluation result.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of insulation bushing, in particular to a method and device for evaluating insulation state of converter transformer insulation bushing. BACKGROUND

[0002] The core converter equipment (such as converter valve and converter transformer) in a DC converter station which bears the function of AC-DC conversion is facing more severe operation challenges than AC system. In a high-voltage direct-current transmission system, as the core hub connecting the converter transformer and the valve hall, the valve-side bushing of the converter transformer directly determines the operation safety of the DC system. The current engineering adopts a "gas-liquid-solid structure" triple protection system: SF6 (sulfur hexafluoride) gas or transformer oil forms the main insulation medium; the oil-impregnated insulation paper core optimizes the axial electric field distribution; and the epoxy resin casting shell provides mechanical support and environmental protection. In the long-term operation of the high-voltage direct-current bushing, the composite insulation structure is threatened by multiple aging factors: high-frequency harmonics cause abnormal enhancement of oil-paper insulation partial discharge, accelerating material degradation; load fluctuations cause significant temperature difference between the epoxy resin shell and the internal core, inducing interfacial separation to form micro defects; harmful chemical substances generated by the decomposition of the insulation medium erode the metal components, while the carbon particles generated by the decomposition accumulate between the paper layers; and the moisture and salt spray in the environment penetrate the material microcracks to form conductive channels. The synergistic effect of these electrical, thermal, chemical and environmental factors eventually leads to a significant decrease in the axial flashover voltage and radial breakdown strength of the bushing, seriously threatening the safety and reliability of the DC system.

[0003] In the face of such severe challenges, the existing method for evaluating the insulation state of the converter transformer insulation bushing cannot continue to meet the growing requirements for evaluation accuracy due to the inability to consider multiple influencing factors and the deviation of a single model. SUMMARY

[0004] To solve the above technical problems, the embodiments of the present application provide a method and device for evaluating the insulation state of a converter transformer insulation bushing to solve the technical problem that the existing technology cannot meet the requirement of accurate evaluation for the converter transformer insulation bushing.

[0005] The first aspect of the embodiments of the present application provides a method for evaluating the insulation state of a converter transformer insulation bushing, which comprises:

[0006] Collecting electrical parameters and environmental parameters of the insulation bushing to obtain a first data set, performing electrical stress and thermal stress analysis on each parameter in the first data set to obtain an analysis result;

[0007] Based on the analysis result, each parameter in the electrical parameters and the environmental parameters is fused to obtain a state evaluation index system, wherein the state evaluation index system includes a plurality of evaluation indexes;

[0008] obtain a second data set by acquiring historical electrical parameters and historical environmental parameters of the insulating sleeve, perform data enhancement on data corresponding to each parameter of the first data set based on the second data set, and obtain a third data set;

[0009] obtain a weight vector of each evaluation index based on the influence degree of each evaluation index in the state evaluation index system on the insulation state, and evaluate the insulation state of the insulating sleeve by using an ensemble algorithm according to the third data set and the weight vector to obtain an evaluation result.

[0010] The insulating state of the insulating sleeve of the converter transformer is comprehensively analyzed by collecting multiple parameters, instead of using a single electrical parameter for analysis, to provide a more accurate evaluation result; through transfer learning, self-data enhancement is realized to make subsequent evaluation more accurate; in addition, the ensemble algorithm makes up for the deviation vacancy of a single model, thereby realizing accurate evaluation of the insulation state.

[0011] In a possible implementation form of the first aspect, the data corresponding to each parameter of the first data set is subjected to data enhancement based on the second data set to obtain a third data set, including:

[0012] align each data in the first data set and the second data set in a heterogeneous dimension to obtain a first feature matrix;

[0013] perform heterogeneous distribution matching on the first feature matrix according to the first feature matrix to obtain a distribution-aligned feature representation, perform knowledge transfer through a domain adaptation technology according to the distribution-aligned feature representation, and obtain the third data set.

[0014] The first data set is supplemented with historical data through heterogeneous dimension alignment, and migration failure is avoided; the distribution difference between the first data set and the second data set is eliminated through heterogeneous distribution matching, data is self-enhanced, the domain failure problem is solved, and the accuracy of subsequent evaluation is improved.

[0015] In a possible implementation form of the first aspect, aligning each data in the first data set and the second data set in a heterogeneous dimension to obtain a first feature matrix includes:

[0016] extract target domain private features and target domain public features from the first data set, and extract source domain private features and source domain public features from the second data set;

[0017] generate source domain support samples through mapping according to the source domain public features;

[0018] generate target domain support samples through mapping according to the target domain public features;

[0019] The target domain private features, the target domain public features, the source domain private features, the source domain public features, the source domain support samples and the target domain support samples are combined based on a preset structured mode to obtain a first feature matrix.

[0020] In a possible implementation manner of the first aspect, according to the first feature matrix, heterogeneous distribution matching is performed on the first feature matrix to obtain distribution-aligned feature representation, and according to the distribution-aligned feature representation, knowledge migration is performed through a domain adaptation technology to obtain a third data set, including:

[0021] According to the first feature matrix, a quantization difference in the first feature matrix is calculated through a maximum mean discrepancy method.

[0022] Based on the posterior probability estimation, a hyperparameter in the maximum mean discrepancy method is adjusted to obtain an optimized hyperparameter.

[0023] According to the quantization difference and the optimized hyperparameter, the distribution of the first feature matrix is adjusted to obtain the distribution-aligned feature representation.

[0024] According to the distribution-aligned feature representation, knowledge migration is performed through the domain adaptation technology to obtain the third data set.

[0025] In a possible implementation manner of the first aspect, according to the state evaluation index system, a weight vector of each evaluation index is obtained through combined weighting, including:

[0026] N evaluation indexes are obtained from the state evaluation index system, wherein N is greater than 0 and is an integer;

[0027] According to the N evaluation indexes, a second weight vector corresponding to the N evaluation indexes is generated through an analytic hierarchy process and a coefficient of variation method;

[0028] The covariance of the N evaluation indexes is calculated, and a conflict matrix is established according to the covariance;

[0029] According to the conflict matrix, a Nash equilibrium solution of the second weight vector is solved through a balancing model to obtain a third weight vector;

[0030] The third weight vector is subjected to consistency checking, and after the checking, the third weight vector is output as a first weight vector to obtain the weight vector of each evaluation index.

[0031] The subjective weight and the objective weight are generated through the analytic hierarchy process and the coefficient of variation method, and the Nash equilibrium solution is obtained based on the conflict matrix, so that the weight optimization is realized, and the accuracy of the subsequent evaluation result is improved.

[0032] In a possible implementation manner of the first aspect, the second weight vector comprises: a third weight vector and a fourth weight vector; the second weight vector corresponding to the N evaluation indexes is generated by an analytic hierarchy process and a coefficient of variation method according to the N evaluation indexes, comprising:

[0033] Based on the analytic hierarchy process, the relative importance of the N evaluation indexes is compared by a scale method to generate a judgment matrix;

[0034] According to the judgment matrix, the subjective weight of the N evaluation indexes is calculated by a geometric mean algorithm to obtain the third weight vector corresponding to the N evaluation indexes;

[0035] Based on the coefficient of variation method, the variation index coefficient of the N evaluation indexes is calculated to obtain the fourth weight vector corresponding to the N evaluation indexes.

[0036] In a possible implementation manner of the first aspect, the insulation state of the insulating sleeve is evaluated by using an ensemble algorithm according to the third data set and the weight vector to obtain an evaluation result, comprising:

[0037] The third data set is subjected to feature fusion to obtain a first feature vector;

[0038] The first feature vector is subjected to game theory weighting according to the weight vector to obtain a second feature vector;

[0039] The second feature vector is input into a base learning machine for prediction to obtain a plurality of aging characteristic parameters, wherein the plurality of aging characteristic parameters comprise an aging probability, an aging index and a remaining life;

[0040] The plurality of aging characteristic parameters are subjected to aging trajectory fusion by a stacked ensemble model to obtain the evaluation result, wherein the base learning machine comprises a first base model, a second base model and a third base model.

[0041] The ensemble algorithm of the present application reduces the risk of overfitting, makes up for the vacancy of single model bias, and realizes accurate prediction of the insulation state of the sleeve.

[0042] To solve the same technical problem, a second aspect of an embodiment of the present application provides an insulating sleeve insulation state evaluation device of a converter transformer, comprising a parameter acquisition module, an index establishment module, a data enhancement module and a state evaluation module, wherein,

[0043] The parameter acquisition module is used to acquire electrical parameters and environmental parameters of the insulating sleeve to obtain a first data set, and to analyze each parameter in the first data set for electrical stress and thermal stress to obtain an analysis result;

[0044] The index establishing module is configured to fuse each of the electrical parameters and the environmental parameters based on the analysis result to obtain a state evaluation index system, wherein the state evaluation index system comprises a plurality of evaluation indexes;

[0045] The data enhancement module is configured to obtain historical electrical parameters and historical environmental parameters of the insulating sleeve to obtain a second data set, and perform data enhancement on data corresponding to each parameter of the first data set based on the second data set to obtain a third data set.

[0046] The state evaluation module is configured to obtain a weight vector of each evaluation index by combining weighting according to the state evaluation index system, and evaluate the insulation state of the insulating sleeve by using an integrated algorithm based on the third data set and the weight vector to obtain an evaluation result.

[0047] The third aspect of the embodiment of the present application provides a computer device, comprising:

[0048] The memory is configured to store the computer program.

[0049] The processor is configured to implement the steps of the converter transformer insulating sleeve insulation state evaluation method according to the first aspect when executing the computer program.

[0050] The fourth aspect of the embodiment of the present application provides a storage medium, and the storage medium stores a computer program, and the computer program is executed by the processor to implement the steps of the converter transformer insulating sleeve insulation state evaluation method according to the first aspect.

[0051] The technical scheme of the present application has the following advantages:

[0052] The present application analyzes and fuses each of the electrical parameters and the environmental parameters by the electrical parameters and the environmental parameters of the insulating sleeve to obtain a state evaluation index system, determines a weight vector of each evaluation index in the state evaluation index system, analyzes the insulation state of the converter transformer insulating sleeve by collecting multiple parameters, instead of using a single electrical parameter for analysis, and provides more accurate evaluation results. BRIEF DESCRIPTION OF DRAWINGS

[0053] In order to more clearly illustrate the technical solutions in the specific embodiments of the present application or the prior art, the following will briefly introduce the drawings needed to be used in the specific embodiments or the prior art description. Obviously, the drawings described below are some embodiments of the present application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0054] Figure 1 The first flow chart of the insulation state evaluation method of the converter transformer insulation bushing in the embodiment of the present application;

[0055] Figure 2 The bushing state index system based on multi-dimensional state parameters in the embodiment of the present application;

[0056] Figure 3 The bushing insulation state evaluation flow chart in the embodiment of the present application;

[0057] Figure 4 The multi-parameter data processing and insulation evaluation flow chart in the embodiment of the present application;

[0058] Figure 5 The second flow chart of the insulation state evaluation method of the converter transformer insulation bushing in the embodiment of the present application;

[0059] Figure 6 The third flow chart of the insulation state evaluation method of the converter transformer insulation bushing in the embodiment of the present application;

[0060] Figure 7 The model weight selection flow chart in the embodiment of the present application;

[0061] Figure 8 The structure block diagram of the converter transformer insulation bushing insulation state evaluation device in the embodiment of the present application;

[0062] The drawings show: 400, the converter transformer insulation bushing insulation state evaluation device; 401, the parameter acquisition module; 402, the index establishment module; 403, the data enhancement module; 404, the state evaluation module. DETAILED DESCRIPTION

[0063] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative labor are within the protection scope of the present application.

[0064] The converter transformer insulation bushing insulation state evaluation method provided by the embodiment of the present application, such as Figure 1As shown, Figure 1 The flow chart of the method for evaluating the insulation condition of the converter transformer insulation bushing comprises steps S101 to S104, and each step is specifically as follows:

[0065] S101, collecting electrical parameters and environmental parameters of the insulation bushing to obtain a first data set, and performing electrical stress and thermal stress analysis on each parameter in the first data set to obtain an analysis result.

[0066] In this embodiment, the electrical parameters include, but are not limited to, the end screen grounding current, the total harmonic content, the high-frequency partial discharge, and the dielectric loss; and the environmental parameters include, but are not limited to, the body content, the micro-water content, the pressure, the temperature, and the humidity.

[0067] It should be noted that the end screen current is directly measured by the end screen grounding line current method. The total harmonic content reflects the distortion degree of the voltage or current waveform in the power system. When collecting, the input port of the harmonic analyzer is reliably connected to the voltage and current measurement points related to the insulation bushing, and appropriate sampling frequency and recording time are set to ensure that the collected data can comprehensively and accurately reflect the harmonic condition. The expression of the total harmonic content is:

[0068]

[0069] Among them, is the effective value of the fundamental voltage; is the effective value of the hth harmonic voltage; and H is the highest harmonic number.

[0070] Similarly, the high-frequency partial discharge is collected by using a high-frequency partial discharge detector. During the detection process, appropriate coupling methods such as capacitive coupling or pulse current method coupling can be selected according to the structure and installation method of the insulation bushing, and the detector is reliably connected to the insulation bushing. The preset sampling frequency of the detector is set to capture the rapid change process of the partial discharge signal. The expression of the high-frequency partial discharge is:

[0071]

[0072] Among them, is the high-frequency PD (Partial Discharge, partial discharge) signal spectrum, which is obtained by FFT (Fast Fourier Transform); and f1~f2 is the characteristic frequency band.

[0073] The dielectric loss is one of the important parameters for measuring the insulation performance of the insulation bushing, which reflects the energy loss of the insulation material under the action of the electric field. During the collection process, appropriate test voltage and frequency are selected according to the rated voltage and actual operation of the insulation bushing for collection. The expression of the dielectric loss is:

[0074]

[0075] in, , These are the k-th sampled values ​​of voltage and current; This represents the phase difference between voltage and current.

[0076] Environmental parameters or non-electrical quantities such as gas content, trace moisture content, pressure, temperature, and humidity are measured using sensors. Because different parameters have different dimensions and orders of magnitude, direct analysis may lead to the exaggeration or neglect of the effects of some parameters. To eliminate this influence, the cleaned data needs to be normalized. The standardized expression is:

[0077]

[0078] Where x represents the measured data, namely electrical parameters and environmental parameters. min x is the minimum value of the data. max This represents the maximum value of the data.

[0079] like Figure 2 As shown, Figure 2 This is a block diagram of the bushing condition index system based on multidimensional condition parameters in an embodiment of the present invention. Based on the analysis of the influence of electrical and thermal stress on the bushing operating condition, and by integrating environmental parameters or non-electrical quantities (gas content, trace moisture content, etc.) and electrical parameters (end screen current, high-frequency partial discharge, etc.), the condition parameters are divided into indicators reflecting insulation performance and thermal performance. At the same time, combined with information such as operating conditions, maintenance history, and external environment, a condition assessment index system based on multidimensional condition parameters is constituted.

[0080] S102. Based on the analysis results, the various parameters in the electrical and environmental parameters are integrated to obtain the condition assessment index system, which includes multiple assessment indicators.

[0081] In this embodiment, by comprehensively considering the influence of multiple factors on the operating status of the bushing, these factors are categorized and integrated to form a multi-dimensional evaluation system, thereby more scientifically judging the health and operational reliability of the bushing. For example, the AHP method (Analytic Hierarchy Process) can be used to integrate environmental and electrical parameters. The basic steps are as follows:

[0082] 1) Construct a hierarchical evaluation model: Establish a three-layer structure, namely the target layer (representing the overall operating status of the bushing), the criterion layer (insulation performance, thermal performance, operating conditions, maintenance history, and external environment), and the indicator layer (specific electrical and non-electrical quantities).

[0083] 2) Construct judgment matrix: compare the importance of each element in each level with each other, and form a judgment matrix by using 1-9 scale method , which satisfies: The evaluation process integrates expert experience and data analysis.

[0084] 3) Calculate weight vector: single-layer weight calculation, calculate the geometric mean of each row element of the judgment matrix : ; the normalized weight vector ; where, ; combined weight calculation, the element of the index layer The synthetic weight of the target layer: ; where, is the weight of the criterion layer, is the weight of the index under the criterion layer.

[0085] 4) Consistency check: calculate the matrix product vector V:

[0086]

[0087] In the formula, is the judgment matrix, is the normalized weight vector.

[0088] Calculate the maximum eigenvalue :

[0089]

[0090] Where, is the first component of the product vector ; is the first component of the weight vector. Verify the consistency ratio :

[0091]

[0092] Where, CR < 0.1 (otherwise adjust the judgment matrix), RI is the random consistency index.

[0093] 5) State fusion evaluation: normalize the measured value of each index and map it to the state score ;

[0094] Weighted fusion of comprehensive health index HI:

[0095]

[0096] In the formula, is the state score, indicates the​​ a weight coefficient corresponding to the one state.

[0097] S103, historical electrical parameters and historical environmental parameters of the insulating sleeve are acquired to obtain a second data set, and data corresponding to each parameter of the first data set is subjected to data enhancement based on the second data set to obtain a third data set.

[0098] In the embodiment, the historical parameters in the second data set are used to perform knowledge migration to the first data set, so that the first data set is supplemented with historical data, and self-enhancement of the data set is realized. It should be noted that the historical parameters refer to the annotated historical electrical parameters and historical environmental parameters, i.e., source domain data.

[0099] It should be noted that the parameters can be understood as parameters such as the screen grounding current and high-frequency partial discharge in the electrical parameters, and refer to various data types, without involving specific numerical values, while the data corresponding to each parameter can be understood as specific numerical values, for example, the data corresponding to the screen grounding current and the high-frequency partial discharge.

[0100] S104, a weight vector of each evaluation index is obtained based on the influence degree of each evaluation index in the state evaluation index system on the insulation state, and an integrated algorithm is used to evaluate the insulation state of the insulating sleeve according to the third data set and the weight vector to obtain an evaluation result.

[0101] In the embodiment, the combined weighting is a multi-index weight determination method, the core of which is to fuse the advantages of the "subjective weighting method" and the "objective weighting method", and a more comprehensive and reliable index weight vector is obtained through a reasonable fusion strategy. In a multi-parameter evaluation scene (such as insulating sleeve insulation state evaluation, device health state diagnosis, etc.), a single weighting method often has limitations: the subjective weighting method (such as AHP, expert scoring method) depends on artificial experience or field knowledge, can reflect actual business needs, but may be affected by subjective preferences, leading to a disconnection between the weight and the data law; the objective weighting method (such as the coefficient of variation method, the entropy method, the principal component analysis method) calculates the weight based on the data itself characteristics (such as fluctuation degree, information entropy), which can reflect the objective law of data, but may ignore the actual physical meaning or engineering priority of the index. The combined weighting method takes the advantages of both methods, so that the optimal weight is not only consistent with the actual needs of the field, but also in line with the inherent law of data, providing a more reasonable quantitative basis for subsequent multi-index fusion evaluation (such as comprehensive health index calculation, state level judgment).

[0102] In this embodiment, S104 includes: performing feature fusion on the third dataset to obtain a first feature vector; assigning game-theoretic weights to the first feature vector according to the weight vector to obtain a second feature vector; constructing an electro-thermal aging kinetic equation based on the second feature vector, and embedding the electro-thermal aging kinetic equation into a preset loss function; calculating several aging feature parameters based on the base learner and the second feature vector through the loss function; and fusing the aging trajectory of the several aging feature parameters through a stacked ensemble model to obtain an evaluation result.

[0103] like Figure 3 As shown, Figure 3 This is a flowchart of the bushing insulation status assessment in an embodiment of the present invention; wherein, multi-source data acquisition is performed by step S101; intelligent method weighting is game theory weighting, which is used to accurately calculate the index weights and assist in decision analysis. It integrates the advantages of multiple methods and provides a reliable way to determine the index weights; Stacking integration is a stacking integration model.

[0104] In this embodiment, the third dataset is fused using a multi-feature fusion layer of an ensemble algorithm, and the first feature vector is: .

[0105] In this embodiment, game-theory weighting is implemented by the game-theory weighting layer of the ensemble algorithm, and the second feature vector is: ; The weight vector is expressed as: ;in, Features The corresponding weights are 1≤i≤n, i∈Z.

[0106] In this embodiment, the physical mechanism constraint layer of the integrated algorithm is used to construct the electro-thermal aging kinetic equation and embed it into the loss function, so that the established model conforms to physical laws. The electro-thermal aging kinetic equation is:

[0107]

[0108] Where k is the material degradation constant; is the activation energy; E is the electric field strength.

[0109] In this embodiment, the base learner training is a multi-dimensional physical pattern extractor, used to overcome the limitations of modeling from a single "electro-thermal-chemical" perspective. Its base models include the first base model LightGBM, the second base model 1D-CNN, and the third base model BiLSTM. Each base model is trained using a corresponding loss function, as shown in Table 1.

[0110] Table 1. Base Learner Training Parameter Settings

[0111]

[0112] where LightGBM is Light Gradient Boosting Machine, 1D-CNN is 1-Dimensional Convolutional Neural Network, and BiLSTM is Bidirectional Long Short-Term Memory; represent the fusion features of multiple parameters at a certain time; time sequence represents evolution and continuous trajectory in a period of time.

[0113] After obtaining the first feature vector, the first feature vector at the characteristic time is input into the LightGBM model for prediction to obtain the aging probability P GBM The first feature vector sequence in a period of time is input into the 1D-CNN model for prediction to obtain the aging index I CNN The first feature vector sequence in a period of time is input into the BiLSTM model for prediction to obtain the remaining useful life RUL LSTM Then, the output results of the above three models are fused by using the Stacking integrated layer to obtain the insulation state evaluation result.

[0114] It should be noted that the loss function of the electric-thermal aging dynamics equation constructed in the physical mechanism constraint layer can minimize the deviation between the model output and the physical law, so that the physical compliance of all base learners is trained, such as whether it conforms to the electric-thermal aging law, so as to ensure that the model will not output results that violate physical common sense, such as the higher the electric field strength, the slower the aging, the lower the temperature, and the greater the activation energy.

[0115] The Stacking integrated layer can realize physical consistency verification of multi-model prediction, adaptive weight distribution in dynamic environment, and aging trajectory fusion across time scales through the meta-learner. The primary prediction matrix is:

[0116]

[0117] In the formula, is the aging probability, is the aging index, is the remaining useful life.

[0118] In this embodiment, the physical consistency verification process is to embed a physical constraint term in the meta-learner loss function:

[0119]

[0120] wherein, is the physical rule checking function:

[0121]

[0122] wherein, α2 is the L2 regularization coefficient, which plays a role in controlling the weight complexity and preventing overfitting, and is obtained by grid search cross-validation method; λ1 is the physical constraint strength coefficient, which plays a role in adjusting the strictness of the physical rule constraint, and is obtained by the fault recall rate maximization strategy method; Y is the true state label vector; k is the electro-thermal aging equivalent coefficient, which is determined by material test.

[0123] In the embodiment, the adaptive weight distribution in the dynamic environment includes: meta-learner coefficient dynamic adjustment, the expression is:

[0124]

[0125] wherein, is the base learner sliding time window prediction variance; are the weights of P GBM , I CNN , and RUL LSTM , respectively. is the environmental adaptation factor: ; wherein, γ represents the environmental sensitivity index, which is determined by the sensitivity-stability Pareto optimization method; represents real-time temperature and humidity / load; represents the training environment.

[0126] In the embodiment, the aging trajectory fusion across time scales includes: unifying P GBM (seconds), I CNN (minutes), and RUL LSTM (hours) to a minute-level time axis as: ; the unification from hours to minutes uses the BiLSTM interpolation method.

[0127] In the embodiment, the evaluation result is calculated by the Weibull proportional risk model; the Weibull proportional risk model is:

[0128]

[0129] wherein, is the benchmark failure rate curve based on the historical failure library; is the remaining life attenuation rate.

[0130] The evaluation result is shown in Table 2:

[0131] Table 2 Risk Level Assessment Results

[0132]

[0133] The output results are: 0 (normal), 1 (caution), 2 (abnormal), 3 (serious).

[0134] In this embodiment, an integrated algorithm is used to assess the aging condition of bushing insulation. This approach offers advantages such as high accuracy, robustness, flexibility, and applicability to various scenarios. The state level is determined by combining weighted values ​​and algorithm coefficients. This effectively improves the accuracy of the assessment level while quantifying the correlation between indicators and assessment levels.

[0135] like Figure 4 As shown, Figure 4 This is a flowchart of multi-parameter data processing and insulation assessment in an embodiment of the present invention. The overall assessment process includes: establishing an index system of multi-dimensional parameters for the bushing based on actual monitoring data and simulation data (S101, S102); enhancing the data of the dataset through transfer learning (S103); selecting and optimizing model weights using intelligent methods (S104); predicting the degree of insulation aging of the bushing using an ensemble algorithm and classifying the degree (S104); and performing defect analysis on the bushing based on the obtained assessment results.

[0136] This embodiment collects and comprehensively analyzes the insulation status of the converter transformer bushings using multiple parameters, replacing the analysis using only electrical parameters, and provides more accurate evaluation results. Through transfer learning, it achieves self-data augmentation to make subsequent evaluations more accurate. In addition, the integrated algorithm makes up for the biases and gaps of the single model, thereby achieving accurate evaluation of the insulation status.

[0137] like Figure 5 As shown, Figure 5 This is the second flowchart of the method for evaluating the insulation status of converter transformer bushings in this embodiment of the invention; by Figure 5 It can be seen that S103 includes steps S201 and S202, and the specific steps are as follows:

[0138] S201. Perform heterogeneous dimension alignment on each data in the first and second datasets to obtain the first feature matrix.

[0139] In the present example, S201 comprises: extracting target domain private features and target domain public features from the first data set, and extracting source domain private features and source domain public features from the second data set; generating source domain support samples through mapping according to the source domain public features; generating target domain support samples through mapping according to the target domain public features; combining the target domain private features, the target domain public features, the source domain private features, the source domain public features, the source domain support samples and the target domain support samples based on a preset structured manner to obtain a first feature matrix.

[0140] In the present embodiment, in terms of data acquisition, different parameters have the problem of dimension misalignment. At the same time, some data that are not directly measured but known historical information can also be introduced to supplement the parameter data. This requires heterogeneous dimension alignment to solve the transfer failure problem caused by the inconsistency of the feature space dimensions of the source domain and the target domain. There are the following formulas:

[0141]

[0142] Among them, represents the source domain public features, which refers to the core parameters common to the source domain (historical / laboratory data) and the target domain (on-site monitoring), including the grounding current of the last screen, gas content, and micro water content, etc. represents the source domain private features, which refers to the parameters that only exist in the source domain and have no monitoring capability in the target domain, including laboratory destructive test data and special working condition simulation records. represents the target domain public features, which refers to the core parameters common to the target domain and the source domain. and are essentially the same, but come from the target domain, including the on-site measured current of the last screen, gas content, and micro water content, etc. represents the target domain private features, which refers to new monitoring parameters that only exist in the target domain, including newly added vibration sensor data and real-time environmental corrosion indicators; the source domain support samples can be represented as: ; the target domain support samples can be represented as: .

[0143] S202, according to the first feature matrix, performing heterogeneous distribution matching on the first feature matrix to obtain a distribution-aligned feature representation, and performing knowledge transfer through a domain adaptation technology according to the distribution-aligned feature representation to obtain a third data set.

[0144] In the embodiment, S202 comprises: calculating a quantization difference in the first feature matrix by a maximum mean difference method according to the first feature matrix; adjusting a hyperparameter in the maximum mean difference method based on a posterior probability estimation to obtain an optimized hyperparameter; and adjusting a distribution of the first feature matrix according to the quantization difference and the optimized hyperparameter to obtain a third data set.

[0145] In the embodiment, the heterogeneous distribution matching is used to solve the model failure problem caused by the distribution difference between the source domain and the target domain data, and cross-domain knowledge transfer is realized through the domain adaptation technology. MMD (Maximum Mean Discrepancy) is used to evaluate the difference between two domains:

[0146]

[0147] wherein, denotes the number of samples of the reconstructed source domain, denotes the number of samples of the reconstructed target domain. denotes the conditional distribution. denotes and the marginal distribution, denotes and the conditional distribution. denotes a Hilbert reproducing space, is a hyperparameter.

[0148] The optimized hyperparameter is:

[0149]

[0150] wherein, denotes the prior probability, and z denotes the number of fault sample sets in other working conditions. is equivalent to the coefficient, which can be obtained by a kernel function.

[0151] Heterogeneous distribution matching aims to reduce the distributional differences between the source and target domains, enabling models to better transfer knowledge from the source domain to the target domain. After heterogeneous distribution matching is completed, the target domain data can be augmented at the data, feature, and model levels. For example, when augmenting the target domain data at the data level, generative models such as Generative Adversarial Networks (GANs) or Variational Autoencoders (VAEs) can be used to generate new samples similar to the target domain data based on the matched distribution features. These new samples are then added to the target domain data to obtain a third dataset. Taking Generative Adversarial Networks (GANs) as an example, a GAN model, including a generator and a discriminator, is pre-trained using historical electrical and environmental parameters from the source domain and their corresponding annotations. During pre-training, the generator learns to generate electrical and environmental parameters corresponding to the defects, while the discriminator learns to distinguish between real and generated parameter data. Both are continuously optimized through adversarial training.

[0152] The pre-trained GAN model is then transferred to the target domain, and fine-tuned using real electrical and environmental parameters from the target domain. During fine-tuning, domain adaptation techniques can be introduced to reduce the data distribution differences between the source and target domains, thereby improving the model's transfer performance. After fine-tuning, the trained generator is used to generate the required new electrical and environmental parameters, which are then added to the second dataset to obtain the third dataset.

[0153] This embodiment supplements the first dataset with historical data through heterogeneous dimension alignment and avoids migration failure; heterogeneous distribution matching eliminates the distribution differences between the first and second datasets, realizes data self-enhancement, solves the domain failure problem, and thus improves the accuracy of subsequent evaluation.

[0154] like Figure 6 As shown, Figure 6 This is the third flowchart of the method for evaluating the insulation status of the bushing of a converter transformer in an embodiment of the present invention; wherein, S104 includes steps S301 to S305, and the specific steps are as follows:

[0155] S301. Obtain N evaluation indicators from the state evaluation indicator system; where N is greater than 0 and N is an integer.

[0156] S302. Based on the N evaluation indicators, generate the second weight vector corresponding to the N evaluation indicators using the analytic hierarchy process and the coefficient of variation method.

[0157] In the embodiment, S302 includes: generating a judgment matrix by comparing the relative importance of the N evaluation indexes with each other through a scale method based on an analytic hierarchy process; calculating the subjective weight of the N evaluation indexes through a geometric mean algorithm according to the judgment matrix to obtain a third weight vector corresponding to the N evaluation indexes; and calculating a variation index coefficient of the N evaluation indexes based on a variation coefficient method to obtain a fourth weight vector corresponding to the N evaluation indexes.

[0158] In the embodiment, the third weight vector is a subjective weight vector, and the expression is: ; wherein, is a subjective weight corresponding to the index , 1≤i≤N, i∈Z.

[0159] The fourth weight vector is an objective weight vector, and the expression is: ; wherein, is an objective weight corresponding to the index , 1≤i≤N, i∈Z. The variation index coefficient is calculated by using the variation coefficient method: ; wherein, is a standard deviation of the jth index; represents a sample mean of the jth index.

[0160] S303, calculating the covariance of the N evaluation indexes, and establishing a conflict matrix according to the covariance.

[0161] In the embodiment, the conflict matrix is used to measure the competitive relationship between indexes, can quantitatively depict the linear dependence relationship between two random variables, and can observe how different indexes are coordinated or offset by calculating the covariance. Therefore, by calculating the covariance of multiple indexes, the conflict matrix is constructed according to the obtained covariance, and the expression of the conflict matrix is:

[0162]

[0163] ; wherein, is the covariance between the indexes i and j; is a standard deviation.

[0164] S304, solving the Nash equilibrium solution of the second weight vector through a balance model according to the conflict matrix to obtain the third weight vector.

[0165] In the embodiment, the balance model is used to minimize the deviation of the subjective weight and the objective weight in the second weight vector, and the Nash equilibrium solution is:

[0166]

[0167] Wherein, λ2 is a conflict adjustment factor, usually 0.5; I is a unit matrix.

[0168] S305, the consistency of the third weight vector is checked, and after the checking, the third weight vector is output as the first weight vector, and the weight vector of each evaluation index is obtained.

[0169] In the embodiment, the Kendall coefficient consistency check is used, and the Kendall coefficient is:

[0170]

[0171] Wherein, The rank sum of the kth sample; m is the number of weight vectors. When W>0.7, the consistency check is satisfied.

[0172] The embodiment generates subjective weight and objective weight by the analytic hierarchy process and the coefficient of variation method, and obtains the Nash equilibrium solution based on the conflict matrix, realizes the weight optimization, and improves the accuracy of the subsequent evaluation result.

[0173] As shown in Figure 7 , Figure 7 The model weight selection flowchart in the embodiment of the application; wherein, the index set is composed of indexes in the index system; the initial weight vector is the second weight vector; the index conflict matrix is the conflict matrix in the embodiment; the Nash equilibrium weight is solved based on the conflict matrix, and the optimal weight, that is, the third weight vector, is output after the consistency check.

[0174] The converter transformer insulation sleeve insulation state evaluation device provided by the embodiment of the application, as shown in Figure 8 , Figure 8 The structural block diagram of the converter transformer insulation sleeve insulation state evaluation device 400 in the embodiment of the application, comprising a parameter acquisition module 401, an index establishment module 402, a data enhancement module 403 and a state evaluation module 404, wherein,

[0175] The parameter acquisition module 401 is used for acquiring the electrical parameters and environmental parameters of the insulation sleeve, obtaining a first data set, and performing electrical stress and thermal stress analysis on each parameter in the first data set to obtain an analysis result;

[0176] The index establishment module 402 is used for fusing each parameter in the electrical parameters and environmental parameters based on the analysis result to obtain a state evaluation index system, wherein the state evaluation index system comprises a plurality of evaluation indexes;

[0177] The data enhancement module 403 is used for obtaining historical electrical parameters and historical environmental parameters of the insulation sleeve to obtain a second data set, and performing data enhancement on the data corresponding to each parameter in the first data set based on the second data set to obtain a third data set;

[0178] The state evaluation module 404 is configured to obtain a weight vector of each evaluation index based on the influence degree of each evaluation index in the state evaluation index system on the insulation state, and to evaluate the insulation state of the insulation sleeve by using an integrated algorithm according to the third data set and the weight vector, to obtain an evaluation result.

[0179] The specific embodiment of the converter transformer insulation sleeve insulation state evaluation device is basically the same as the above-mentioned specific embodiment of the converter transformer insulation sleeve insulation state evaluation method, and will not be described here.

[0180] In an embodiment of the present application, a computer device is provided, which includes a memory and a processor, the memory stores a computer program, and the processor implements the above steps when executing the computer program; the computer device provided in the embodiment has similar implementation principles and technical effects to the above method embodiments, and will not be described here.

[0181] In an embodiment of the present application, a computer readable storage medium is provided, which stores a computer program, and the computer program is executed by a processor to implement the above steps; the computer readable storage medium provided in the embodiment has similar implementation principles and technical effects to the above method embodiments, and will not be described here.

[0182] The technical features of the above embodiments can be combined in any manner. To make the description concise, all possible combinations of the technical features in the above embodiments are not described, however, as long as the combinations of the technical features do not exist contradictions, they should be considered as the scope of the present application.

[0183] The above specific embodiments further illustrate the purpose, technical solutions and advantages of the present application. It should be understood that the above description is only for specific embodiments of the present application and is not intended to limit the protection scope of the present application. It is particularly pointed out that any modification, equivalent replacement, improvement, etc. made by those skilled in the art within the spirit and principles of the present application should be included in the protection scope of the present application.

Claims

1. A method for assessing the insulation condition of a converter transformer bushing, characterized in that, include: Electrical and environmental parameters of the insulating bushing are collected to obtain the first dataset. Electrical and thermal stress analyses are performed on each parameter in the first dataset to obtain the analysis results. Based on the analysis results, the electrical parameters and environmental parameters are integrated to obtain a state assessment index system, wherein the state assessment index system includes multiple assessment indicators. Historical electrical and environmental parameters of the insulating bushing are obtained to obtain a second dataset. Based on the second dataset, target domain private features and target domain public features are extracted from the first dataset, and source domain private features and source domain public features are extracted from the second dataset. According to the source domain public features, source domain support samples are generated by mapping, and target domain support samples are generated by mapping according to the target domain public features. Based on a preset structured approach, the target domain private features, the target domain public features, the source domain private features, the source domain public features, the source domain support samples, and the target domain support samples are combined to obtain a first feature matrix; Based on the first feature matrix, the quantization difference in the first feature matrix is ​​calculated using the maximum mean difference method. Based on the posterior probability estimation, the hyperparameters in the maximum mean difference method are adjusted to obtain optimized hyperparameters. Based on the quantization difference and the optimized hyperparameters, the distribution of the first feature matrix is ​​adjusted to obtain a distribution-aligned feature representation. Based on the distribution-aligned feature representation, knowledge transfer is performed using domain adaptation technology to obtain a third dataset. Based on the degree of influence of each evaluation index on the insulation state in the state evaluation index system, the weight vector of each evaluation index is obtained. According to the third dataset and the weight vector, the insulation state of the insulating bushing is evaluated by an integrated algorithm to obtain the evaluation result.

2. The method for assessing the insulation condition of converter transformer bushings as described in claim 1, characterized in that, The step of obtaining the weight vector of each evaluation index by combining and assigning weights according to the state evaluation index system includes: N evaluation indicators are obtained from the state evaluation indicator system; where N is greater than 0 and N is an integer; Based on the N evaluation indicators, a second weight vector corresponding to the N evaluation indicators is generated using the analytic hierarchy process and the coefficient of variation method. Calculate the covariance of the N evaluation indicators, and establish a conflict matrix based on the covariance; Based on the conflict matrix, the Nash equilibrium solution of the second weight vector is obtained by solving the equilibrium model, thus obtaining the optimized second weight vector. The optimized second weight vector is subjected to consistency verification. After passing the verification, the optimized second weight vector is output as the first weight vector to obtain the weight vector of each evaluation index.

3. The method for assessing the insulation condition of converter transformer bushings as described in claim 2, characterized in that, The second weight vector includes a third weight vector and a fourth weight vector; the step of generating the second weight vector corresponding to the N evaluation indicators using the analytic hierarchy process (AHP) and the coefficient of variation method includes: Based on the analytic hierarchy process (AHP), the relative importance of the N evaluation indicators is compared pairwise using scaling to generate a judgment matrix. Based on the judgment matrix, the subjective weights of the N evaluation indicators are calculated using the geometric mean algorithm to obtain the third weight vector corresponding to the N evaluation indicators; Based on the coefficient of variation method, the coefficients of variation of the N evaluation indicators are calculated to obtain the fourth weight vector corresponding to the N evaluation indicators.

4. The method for evaluating the insulation condition of converter transformer bushings as described in claim 1, characterized in that, The step of evaluating the insulation state of the insulating bushing using an ensemble algorithm based on the third dataset and the weight vector, and obtaining the evaluation result, includes: The third dataset is fused to obtain a first feature vector; Based on the weight vector, the first feature vector is weighted using game theory to obtain the second feature vector; The second feature vector is input into the base learner for prediction to obtain several aging feature parameters, including aging probability, aging index and remaining lifespan. By using a stacked ensemble model, aging trajectories are fused across several aging feature parameters to obtain evaluation results. The base learner includes a first base model, a second base model, and a third base model.

5. A device for assessing the insulation condition of a converter transformer bushing, characterized in that, The method for assessing the insulation status of converter transformer bushings as described in any one of claims 1-4 includes a parameter acquisition module, an index establishment module, a data enhancement module, and a status assessment module, wherein... The parameter acquisition module is used to acquire electrical and environmental parameters of the insulating bushing to obtain a first dataset. Electrical and thermal stress analyses are performed on each parameter in the first dataset to obtain analysis results. The index establishment module is used to integrate the electrical parameters and environmental parameters based on the analysis results to obtain a state assessment index system, wherein the state assessment index system includes multiple assessment indicators. The data augmentation module is used to obtain the historical electrical parameters and historical environmental parameters of the insulating bushing to obtain a second dataset. Based on the second dataset, the data corresponding to each parameter in the first dataset is augmented to obtain a third dataset. The state assessment module is used to obtain the weight vector of each assessment index based on the degree of influence of each assessment index on the insulation state in the state assessment index system, and to evaluate the insulation state of the insulating bushing using an integrated algorithm based on the third dataset and the weight vector to obtain the assessment result.

6. A computer device, characterized in that, include: Memory, used to store computer programs; A processor, configured to implement the method for assessing the insulation status of a converter transformer bushing as described in any one of claims 1 to 4 when executing the computer program.

7. A storage medium, characterized in that, The storage medium stores a computer program, which, when executed by a processor, implements the steps of the method for evaluating the insulation status of a converter transformer bushing as described in any one of claims 1 to 4.

Images