A transformer oil-paper insulation fault evaluation method based on characteristics of micro-impurities in transformer oil

Abstract

The application discloses a transformer oil-paper insulation fault evaluation method based on characteristics of micro impurities in transformer oil, and relates to the technical field of power equipment insulation fault diagnosis. The method comprises the following steps: an electric-thermal coupling field simulation model of a transformer oil-paper insulation system is established; micro solid impurity particles are introduced to perform a multi-gradient electric-thermal joint action simulation test to obtain impurity characteristic parameters and corresponding electrical performance parameters; a gradient boosting regression tree algorithm is used to establish a mapping correlation model; a fault evaluation model is generated based on correction of the model according to actual operation condition parameters; electrical performance evaluation values are calculated by inputting the measured impurity characteristic parameters; and an insulation failure risk grade is determined in combination with a threshold value. The application realizes accurate determination of the insulation failure risk grade of the transformer oil-paper insulation system, and improves the accuracy and adaptability of the transformer oil-paper insulation state evaluation.

Description

Technical Field

[0001] This invention relates to the field of electrical equipment insulation fault diagnosis technology, and in particular to a method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil. Background Technology

[0002] As a core piece of equipment in the power system, the stability of the transformer's operating status directly affects the grid's safe power supply capability. Inside the transformer, insulating oil and insulating paperboard together constitute the main oil-paper insulation system, undertaking the dual functions of electrical insulation and heat dissipation. With long-term operation, the transformer is subjected to multiple factors such as electric fields, thermal fields, and mechanical vibrations, causing the insulating materials to gradually age and produce solid impurity particles. These particles are suspended in the insulating oil or deposited on the surface of the insulating paperboard. Currently, the industry generally uses offline sampling analysis combined with online monitoring to assess the insulation condition of transformers. This mainly involves detecting indicators such as dissolved gases, moisture content, and large particulate contamination in the oil, using physicochemical analysis or optical counting methods to obtain relevant data, and then determining whether the equipment is within its normal operating range based on relevant standards.

[0003] However, in ultra-high voltage and long-term operating transformers, the impact mechanism of tiny solid impurities on the performance of oil-paper insulation is complex and difficult to quantify. Existing assessment techniques often fail to accurately capture the dynamic evolution of tiny particles under the influence of electro-thermal coupling fields, resulting in a lag in the prediction of potential failure risks in the insulation system. Especially when facing variable actual operating conditions, traditional assessment models lack adaptive correction capabilities and cannot accurately reflect the local electric field distortion and insulation strength decline trend caused by the accumulation of tiny impurities. This leads to a discrepancy between the insulation failure risk level determination and the actual physical state, failing to meet the urgent needs of high-reliability power grids for early fault warning and refined condition-based maintenance of transformers. Summary of the Invention

[0004] The purpose of this invention is to overcome the shortcomings of the prior art and provide a method for assessing transformer oil-paper insulation faults based on the characteristics of minute impurities in transformer oil. By constructing an electro-thermal coupled field simulation model, establishing a mapping correlation model between the characteristic parameters of minute impurities and electrical performance parameters, and adaptively correcting the model in combination with the actual operating conditions of the transformer, the method can achieve accurate assessment of the insulation failure risk of the transformer oil-paper insulation system.

[0005] To achieve the above objectives, the present invention provides a method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil, the method comprising: S1. Based on the typical structural parameters of transformer oil-paper insulation, the dielectric properties and thermophysical properties of insulating oil and insulating paperboard, establish an electro-thermal coupling field simulation model of the transformer oil-paper insulation system. S2. Introduce tiny solid impurity particles into the electro-thermal coupled field simulation model, set multiple gradients of local overheating temperature and multiple energy levels of partial discharge conditions, and conduct simulation experiments under the combined action of electro-thermal to obtain multiple sets of characteristic parameters of tiny solid impurities and corresponding electrical performance parameters of oil-paper insulation; the characteristic parameters of tiny solid impurities include at least impurity type, particle concentration, particle size distribution and morphological characteristics; the electrical performance parameters include at least breakdown voltage, partial discharge initiation voltage and surface flashover voltage; S3. Based on the acquired multiple sets of simulation data, a mapping correlation model between the characteristic parameters of minute solid impurities and the electrical performance parameters of oil-paper insulation is established using the gradient boosting regression tree algorithm. S4. Obtain the actual operating condition parameters of the transformer under test, and modify the mapping association model based on the actual operating condition parameters to generate an oil-paper insulation fault assessment model that is adapted to the operating conditions of the transformer under test. S5. Extract the characteristic parameters of tiny solid impurity particles in the insulating oil of the transformer under test, input the characteristic parameters into the oil-paper insulation fault assessment model, and calculate the electrical performance assessment value of the oil-paper insulation system of the transformer under test. S6. Based on the preset voltage threshold for the electrical performance of the oil-paper insulation, determine the insulation failure risk level of the oil-paper insulation system of the transformer under test, and output the oil-paper insulation fault risk assessment result.

[0006] Furthermore, in step S1, the typical structural parameters of the transformer oil-paper insulation include at least the type of oil-paper insulation electrode structure, the thickness of the insulation paperboard, the oil gap distance, and the size of the insulation oil channel; The dielectric properties parameters include at least the relative permittivity, volume resistivity, and dielectric loss factor. The thermophysical parameters include at least thermal conductivity, specific heat capacity, and viscosity coefficient.

[0007] Furthermore, in step S1, the establishment of the electro-thermal coupling field simulation model specifically includes: Construct a geometric structural model of the transformer oil-paper insulation system, and reconstruct the electrode structure, insulation paperboard stacking structure and oil channel distribution pattern based on typical structural parameters; Set the dielectric and thermal properties parameters of the transformer insulating oil and insulating paperboard, and set the correlation function of the dielectric parameters with temperature. Set the electric field boundary conditions and temperature field boundary conditions of the transformer oil-paper insulation system. The electric field boundary conditions include AC / DC composite voltage excitation and grounding constraint. The temperature field boundary conditions include ambient temperature and heat source distribution. The heat source distribution is determined by conductor loss and dielectric loss. The constructed geometric model is meshed, and the electric field control equation and thermal field control equation are solved by coupled iteration using the finite element method to obtain the reference electric field distribution and reference temperature field distribution of the transformer oil-paper insulation system under impurity-free conditions.

[0008] Furthermore, in solving the electric field control equation: based on the solution results of the thermal field control equation, the dielectric parameter in the electric field control equation is corrected by calculating the correlation function of the dielectric parameter changing with temperature; In solving the thermal field control equation: based on the solution results of the electric field control equation, the heat source term of the thermal field control equation is corrected by updating the heat source distribution parameters.

[0009] Furthermore, in step S2, the simulation experiment specifically includes: Set at least three temperature levels of local overheating temperature gradient and at least three energy levels of local discharge energy gradient to construct multiple sets of electro-thermal combined action conditions; In the oil gap region and / or oil-paper interface region of the electro-thermal coupled field simulation model, tiny solid impurity particles are introduced according to a preset distribution method, and initial characteristic parameters of the tiny solid impurity particles are set; the preset distribution method includes random distribution or distribution according to electric field intensity gradient; the initial characteristic parameters include at least impurity type, initial particle concentration, initial particle size distribution, and initial morphology characteristic parameters; the impurity type includes free carbon particles, fiber particles, and metal particles; The electro-thermal coupled field simulation model was run under each group of electro-thermal combined working conditions. The tiny solid impurity particles migrated and aggregated under the combined action of the electric field and the temperature field. After each simulation run, the characteristic parameters of the tiny solid impurity particles under the current working condition are collected, and the electrical performance parameters of the oil-paper insulation under the corresponding working condition are collected simultaneously.

[0010] Further, in step S3, the gradient boosting regression tree algorithm is used to establish the mapping association model, specifically including: Outlier removal and normalization were performed on the acquired simulation data to obtain a standardized simulation sample dataset. The standardized simulation sample dataset is divided into a training set and a validation set according to a preset ratio. The feature parameters of tiny solid impurities are set as the model input features, and the electrical performance parameters of oil paper insulation are set as the model output labels. A gradient boosting regression tree initial model is constructed, wherein the gradient boosting regression tree initial model uses the mean squared error as the loss function and adopts a forward step-by-step addition model to generate a CART regression tree round by round. In each iteration, the current regression tree is fitted based on the prediction residuals of the model after the previous iteration, and the node splitting rules and leaf node output weights are updated. After each round of training, the determination coefficients of the model are calculated using the validation set. When the determination coefficient Training stops when a preset threshold is reached or the number of iterations reaches a set upper limit, thus obtaining a mapping correlation model between the characteristic parameters of minute solid impurities and the electrical performance parameters of oil-paper insulation after training is completed.

[0011] Further, in step S4, based on the actual operating conditions of the transformer under test, the mapping correlation model is corrected through adaptive weighted incremental training. The adaptive weighted incremental training specifically includes: Obtain the parameter vector of the transformer under test under actual operating conditions. , where m is the number of operating condition parameters, and the operating condition parameters include at least voltage level, service life and operating oil temperature; For the j-th simulation sample data from the historical simulation experiment, obtain the simulation condition parameter vector corresponding to the simulation sample data. Calculate the weighted feature distance between the simulated operating condition parameter vector and the actual operating condition parameter vector:

[0012] in, The preset weighting coefficient for the k-th operating condition parameter is... The standard deviation of the k-th operating condition parameter across all simulation samples; the weighting coefficient The sensitivity of the electrical performance of the oil-paper insulation to various operating parameters is preset, and the higher the sensitivity, the greater the weight. Based on the weighted feature distance Determine the weight value of the j-th simulation sample data. The weight value and They are inversely proportional, and the sum of the weights of all simulated sample data is 1; Weight values ​​based on each simulation sample data The training set of the mapping correlation model is weighted, and the weighted mean square error is used as the loss function. The incremental gradient boosting regression algorithm is used to continue training on the original mapping correlation model until the error of the validation set no longer decreases, so as to obtain an oil-paper insulation fault assessment model that is adapted to the actual operating conditions of the transformer under test.

[0013] The beneficial effects of this invention are: This invention provides a method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil. It constructs an electro-thermal coupled field simulation model including typical structural parameters, dielectric properties, and thermophysical parameters. By introducing minute solid impurity particles into the model and conducting simulation experiments under multi-gradient electro-thermal combined effects, it obtains corresponding data on the characteristic parameters of minute solid impurities (including impurity type, concentration, particle size distribution, and morphological characteristics) and electrical performance parameters such as breakdown voltage and partial discharge initiation voltage. Based on this data, a nonlinear mapping correlation model between the two is established using a gradient boosting regression tree algorithm. Furthermore, the model is adaptively weighted and incrementally corrected by incorporating the actual operating parameters of the transformer under test, generating a fault assessment model adapted to specific operating conditions. By inputting the measured impurity characteristic parameters into the corrected model, the electrical performance assessment value can be accurately calculated, and the insulation failure risk level can be determined by combining it with a preset threshold. This invention achieves an interpretable quantitative mapping from microscopic impurity characteristics to macroscopic insulation state, effectively solving the problem that existing assessment techniques often struggle to accurately capture the dynamic evolution of tiny particles under electro-thermal coupling fields. This leads to a lag in the prediction of potential failure risks in insulation systems, especially when facing variable actual operating conditions. Traditional assessment models lack adaptive correction capabilities and fail to accurately reflect the local electric field distortion and insulation strength decline caused by the aggregation of tiny impurities, resulting in a discrepancy between the insulation failure risk level determination and the actual physical state. Therefore, this invention avoids misjudgments or omissions caused by insufficient generalization ability of the assessment model, significantly improving the accuracy, real-time performance, and reliability of transformer oil-paper insulation system condition assessment, and providing a scientific basis for condition-based maintenance of power equipment. Attached Figure Description

[0014] Figure 1 This is a schematic diagram of the overall process of a transformer oil paper insulation fault assessment method based on the characteristics of minute impurities in transformer oil provided by the present invention. Detailed Implementation

[0015] The technical solution of the present invention will now be clearly and completely described in conjunction with the accompanying drawings.

[0016] Example 1

[0017] like Figure 1 As shown, a method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil is described, the method comprising: S1. Based on the typical structural parameters of transformer oil-paper insulation, the dielectric properties and thermophysical properties of insulating oil and insulating paperboard, establish an electro-thermal coupling field simulation model of the transformer oil-paper insulation system. S2. Introduce tiny solid impurity particles into the electro-thermal coupled field simulation model, set multiple gradients of local overheating temperature and multiple energy levels of partial discharge conditions, and conduct simulation experiments under the combined action of electro-thermal to obtain multiple sets of characteristic parameters of tiny solid impurities and corresponding electrical performance parameters of oil-paper insulation; the characteristic parameters of tiny solid impurities include at least impurity type, particle concentration, particle size distribution and morphological characteristics; the electrical performance parameters include at least breakdown voltage, partial discharge initiation voltage and surface flashover voltage; S3. Based on the acquired multiple sets of simulation data, a mapping correlation model between the characteristic parameters of minute solid impurities and the electrical performance parameters of oil-paper insulation is established using the gradient boosting regression tree algorithm. S4. Obtain the actual operating condition parameters of the transformer under test, and modify the mapping association model based on the actual operating condition parameters to generate an oil-paper insulation fault assessment model that is adapted to the operating conditions of the transformer under test. S5. Extract the characteristic parameters of tiny solid impurity particles in the insulating oil of the transformer under test, input the characteristic parameters into the oil-paper insulation fault assessment model, and calculate the electrical performance assessment value of the oil-paper insulation system of the transformer under test. S6. Based on the preset voltage threshold for the electrical performance of the oil-paper insulation, determine the insulation failure risk level of the oil-paper insulation system of the transformer under test, and output the oil-paper insulation fault risk assessment result.

[0018] Specifically, taking a 220kV oil-immersed power transformer in a substation as the evaluation object, this paper details the implementation process of the transformer oil paper insulation fault assessment method based on the characteristics of minute impurities in transformer oil provided by this invention.

[0019] As a preferred embodiment of this example, in step S1, the typical structural parameters of the transformer oil-paper insulation include at least the type of oil-paper insulation electrode structure, the thickness of the insulation paperboard, the oil gap distance, and the size of the insulation oil channel. The dielectric properties parameters include at least the relative permittivity, volume resistivity, and dielectric loss factor. The thermophysical parameters include at least thermal conductivity, specific heat capacity, and viscosity coefficient.

[0020] In a preferred embodiment of this invention, step S1, the establishment of the electro-thermal coupling field simulation model, specifically includes: Construct a geometric structural model of the transformer oil-paper insulation system, and reconstruct the electrode structure, insulation paperboard stacking structure and oil channel distribution pattern based on typical structural parameters; Set the dielectric and thermal properties parameters of the transformer insulating oil and insulating paperboard, and set the correlation function of the dielectric parameters with temperature. Set the electric field boundary conditions and temperature field boundary conditions of the transformer oil-paper insulation system. The electric field boundary conditions include AC / DC composite voltage excitation and grounding constraint. The temperature field boundary conditions include ambient temperature and heat source distribution. The heat source distribution is determined by conductor loss and dielectric loss. The constructed geometric model is meshed, and the electric field control equation and thermal field control equation are solved by coupled iteration using the finite element method to obtain the reference electric field distribution and reference temperature field distribution of the transformer oil-paper insulation system under impurity-free conditions.

[0021] Specifically, a 220 kV oil-immersed power transformer that has been in operation for many years was selected, and its nameplate parameters (rated voltage, capacity, insulation method, etc.) and operating history data (years of operation, load conditions, upper oil temperature, etc.) were recorded. In accordance with the sampling regulations of the power industry standard DL / T 432, insulating oil samples were collected from the sampling valve at the bottom of the transformer and placed in a clean, sealed container for later use. At the same time, the actual operating parameters of the transformer were recorded, including but not limited to voltage level, years of operation, current load rate, and operating oil temperature.

[0022] Based on the actual oil-paper insulation structure of the transformer, a two-dimensional or three-dimensional geometric model is established in finite element simulation software (such as COMSOL Multiphysics or ANSYS) according to the actual dimensions. This model accurately depicts the main insulation channel width between the high-voltage and low-voltage windings, the number and thickness of each layer of insulating paperboard, the width and number of oil gaps, and the insulation distance between the electrostatic ring and the winding ends. The electrode structure is simplified to an axisymmetric cylindrical-plate electrode or ball-plate electrode model based on the actual winding-insulating paperboard configuration, which can simulate the actual electric field distribution between the transformer windings and the insulating paperboard.

[0023] After the geometric model is constructed, the above dielectric and thermal properties parameters are assigned to the insulating oil and insulating paperboard in the material library. Specifically, the typical values ​​given by the standard factory data of transformer oil and insulating paperboard or relevant national standards (such as GB / T7595) can be referred to.

[0024] Meanwhile, a correlation function for dielectric parameters as a function of temperature is set, with the range of variation covering the transformer's normal operating temperature to overheating temperature range. Typically, the relative permittivity of insulating oil decreases linearly with increasing temperature, and insulating paperboard also exhibits a similar temperature dependence. The specific variation coefficient can be obtained using values ​​known in the field or through experimental calibration.

[0025] Set electric field boundary conditions: Apply a 220kV power frequency AC voltage to the surface of the high-voltage electrode, and apply zero potential constraints to the surfaces of grounded components such as the low-voltage electrode, tank wall, and core.

[0026] Temperature field boundary conditions are set: the ambient temperature is set to 25℃, and the heat source distribution is determined based on the loss calculation under the transformer's rated load, with conductor loss and dielectric loss set according to the transformer's factory test data. Heat sources are applied separately in the geometric model according to the actual heat-generating locations (winding conductor region and insulating dielectric region).

[0027] The geometric model was meshed, and local mesh refinement was performed in the oil gap region, the oil-paper interface region, and near the electrode surface (minimum mesh size of 0.08 mm) to ensure accurate capture of electric field distortion near tiny solid particles.

[0028] The electric field control equations and thermal field control equations are solved by coupled iteration using the finite element method.

[0029] In a preferred embodiment of this invention, in solving the electric field control equation: based on the solution of the thermal field control equation, the dielectric parameter in the electric field control equation is corrected by calculating the correlation function of the dielectric parameter changing with temperature. In solving the thermal field control equation: based on the solution results of the electric field control equation, the heat source term of the thermal field control equation is corrected by updating the heat source distribution parameters.

[0030] Specifically, the initial temperature field distribution is first solved, and the dielectric parameters of each region are updated according to the temperature-dielectric parameter correlation function. Then, the electric field distribution is solved, the dielectric loss power is calculated, and the heat source term of the temperature field is updated. The iteration is repeated until the deviation between the maximum values ​​of temperature and electric field in two adjacent calculations is less than 0.1%, and the reference electric field distribution and reference temperature field distribution of the oil-paper insulation system under the condition of no impurities are obtained.

[0031] In a preferred embodiment of this invention, step S2 specifically includes the following simulation experiment: Set at least three temperature levels of local overheating temperature gradient and at least three energy levels of local discharge energy gradient to construct multiple sets of electro-thermal combined action conditions; In the oil gap region and / or oil-paper interface region of the electro-thermal coupled field simulation model, tiny solid impurity particles are introduced according to a preset distribution method, and initial characteristic parameters of the tiny solid impurity particles are set; the preset distribution method includes random distribution or distribution according to electric field intensity gradient; the initial characteristic parameters include at least impurity type, initial particle concentration, initial particle size distribution, and initial morphology characteristic parameters; the impurity type includes free carbon particles, fiber particles, and metal particles; The electro-thermal coupled field simulation model was run under each group of electro-thermal combined working conditions. The tiny solid impurity particles migrated and aggregated under the combined action of the electric field and the temperature field. After each simulation run, the characteristic parameters of the tiny solid impurity particles under the current working condition are collected, and the electrical performance parameters of the oil-paper insulation under the corresponding working condition are collected simultaneously.

[0032] Specifically, at least three local overheating temperature gradients are set, such as 80℃, 100℃, and 150℃, which correspond to three typical operating conditions of transformer overheating: mild overheating, moderate overheating, and severe overheating, respectively.

[0033] Set at least three partial discharge energy gradients, such as 0.2J, 0.8J, and 1.5J, to correspond to three typical discharge conditions: weak discharge, medium discharge, and strong discharge, respectively.

[0034] By combining the above temperature gradient and energy gradient with all factors, multiple sets of electro-thermal combined action conditions are constructed.

[0035] In the oil channel region and oil-paper interface region of the electro-thermal coupling field simulation model under each set of operating conditions, three typical small solid impurities in transformer oil are introduced: free carbon particles, fiber aging particles, and metal corrosion particles. Independent particle phase identifiers are assigned to free carbon particles, insulating fiber particles, and metal corrosion particles, covering... The initial distribution of the micro-particle size range adopts a combination of random uniform distribution and enrichment in the high electric field region, which conforms to the natural distribution law of impurities in the actual operation of transformers.

[0036] Simulation models were run under various electro-thermal combined effects. Tiny solid impurity particles migrated and aggregated at the oil-paper interface under the combined action of electric field force (dielectric electrophoresis force) and thermal field force (thermophoresis force). During the simulation, the spatial position, trajectory, and interface adhesion state of each type of particle were recorded using the Lagrange particle tracking method.

[0037] After each set of electro-thermal combined operating condition simulations reaches steady state, five characteristic parameters of tiny solid impurities under the current operating condition are collected simultaneously: particle concentration, median particle size, average sphericity, and particle deposition density at the oil-paper interface (total mass of the corresponding type of impurities attached to the oil-paper interface per unit area). Synchronously collect the electrical performance parameters of oil-paper insulation under the corresponding working conditions: breakdown voltage (obtained by applying a gradually increasing voltage through simulation until breakdown), partial discharge initiation voltage (obtained by monitoring the local field strength to reach the critical voltage of the initiation discharge field strength), and surface flashover voltage (obtained by monitoring the voltage of the interface tangential field strength to reach the flashover critical value).

[0038] The characteristic parameters of the three types of impurities are statistically analyzed independently and are not combined. After the full factorial simulation is completed, a multi-impurity joint sample database is constructed. Each sample contains 15 input characteristic parameters of the three types of impurities, as well as 3 output labels: breakdown voltage, partial discharge initiation voltage, and surface flashover voltage. Each set of operating conditions is subjected to 30 independent simulations (with random seeds that change the initial distribution of particles), and all simulation results are included in the sample database.

[0039] In a preferred embodiment of this invention, step S3 employs a gradient boosting regression tree algorithm to establish the mapping association model, specifically including: Outlier removal and normalization were performed on the acquired simulation data to obtain a standardized simulation sample dataset. The standardized simulation sample dataset is divided into a training set and a validation set according to a preset ratio. The feature parameters of tiny solid impurities are set as the model input features, and the electrical performance parameters of oil paper insulation are set as the model output labels. A gradient boosting regression tree initial model is constructed, wherein the gradient boosting regression tree initial model uses the mean squared error as the loss function and adopts a forward step-by-step addition model to generate a CART regression tree round by round. In each iteration, the current regression tree is fitted based on the prediction residuals of the model after the previous iteration, and the node splitting rules and leaf node output weights are updated. After each round of training, the determination coefficients of the model are calculated using the validation set. When the determination coefficient Training stops when a preset threshold is reached or the number of iterations reaches a set upper limit, thus obtaining a mapping correlation model between the characteristic parameters of minute solid impurities and the electrical performance parameters of oil-paper insulation after training is completed.

[0040] Specifically, all simulation sample data in the sample database are preprocessed using... The criteria remove outliers, and then the valid data undergoes Min-Max normalization to uniformly map all input features and output labels to... Interval.

[0041] The input features are 15 dimensions in total: five feature parameters for each of the three types of impurities (particle concentration, median particle size, average sphericity, particle size distribution width, and particle deposition density at the oil-paper interface) are input in parallel; The output labels have three dimensions: breakdown voltage, partial discharge initiation voltage, and surface flashover voltage.

[0042] The preprocessed simulation sample data was randomly divided into training and validation sets in a 7:3 ratio.

[0043] An initial gradient boosting regression tree model was constructed to fit the variation law of insulation performance under the synergistic effect of multiple impurities. The hyperparameters of the initial gradient boosting regression tree model were set as follows: learning rate of 0.08, maximum depth of regression tree of 8, minimum number of samples in leaf nodes of 6, maximum number of iterations of 200 rounds, and loss function of multi-output mean squared error.

[0044] The training process employs a forward stepwise additive model, generating CART regression trees round by round. In each iteration, the prediction residuals of the previous model for each sample in the training set are calculated. A new regression tree is fitted using these residuals as the target variable, and this tree is added to the current model with the learning rate as weights. After each training round, the determination coefficient R is calculated using the validation set. 2 .

[0045] After multiple iterations, the validation set R 2 Once the value reaches 0.935, exceeding the preset threshold of 0.90, training stops, and the completed mapping-association model is obtained.

[0046] In a preferred embodiment of this invention, in step S4, based on the actual operating conditions of the transformer under test, the mapping correlation model is corrected through adaptive weighted incremental training. The adaptive weighted incremental training specifically includes: Obtain the parameter vector of the transformer under test under actual operating conditions. , where m is the number of operating condition parameters, and the operating condition parameters include at least voltage level, service life and operating oil temperature; For the j-th simulation sample data from the historical simulation experiment, obtain the simulation condition parameter vector corresponding to the simulation sample data. Calculate the weighted feature distance between the simulated operating condition parameter vector and the actual operating condition parameter vector:

[0047] in, The preset weighting coefficient for the k-th operating condition parameter is... The standard deviation of the k-th operating condition parameter across all simulation samples; the weighting coefficient The sensitivity of the electrical performance of the oil-paper insulation to various operating parameters is preset, and the higher the sensitivity, the greater the weight. Based on the weighted feature distance Determine the weight value of the j-th simulation sample data. The weight value and They are inversely proportional, and the sum of the weights of all simulated sample data is 1; Weight values ​​based on each simulation sample data The training set of the mapping correlation model is weighted, and the weighted mean square error is used as the loss function. The incremental gradient boosting regression algorithm is used to continue training on the original mapping correlation model until the error of the validation set no longer decreases, so as to obtain an oil-paper insulation fault assessment model that is adapted to the actual operating conditions of the transformer under test.

[0048] Specifically, the actual operating condition parameter vector of the 220kV transformer is obtained, which includes: voltage level 220kV, operating years 12 years, operating oil temperature 62℃, and average load rate 60%, forming the operating condition characteristic vector.

[0049] For each sample in the sample database, the weighted characteristic distance between the sample and the operating condition of the transformer under test is calculated. Weights are assigned to each sample based on the weighted characteristic distance, with samples closer to each other receiving higher weights, and the sum of all weights is 1. The weights for voltage level, service life, operating oil temperature, and load rate are preset to 0.4, 0.3, 0.2, and 0.1, respectively, based on their impact on insulation performance.

[0050] Using weighted mean square error as the loss function, 20 rounds of incremental training were conducted on the already trained general mapping correlation model to fine-tune the model parameters, making the model more consistent with the actual operating conditions of the transformer under test, and finally obtaining an oil-paper insulation fault assessment model adapted to the 220kV transformer.

[0051] A 500mL operating oil sample collected from the transformer under test was imaged using a microfluidic imaging detection system to detect minute solid impurities. Image processing combined with a support vector machine classification algorithm was used to automatically identify three types of impurities, and characteristic parameters for each type were extracted. Among these, the characteristic parameter for detecting free carbon was: particle concentration. particles / mL, particle size range Particles of this type accounted for 97.8% of the total; the median particle size was... The particle size distribution width (geometric standard deviation) is 1.72; the average sphericity is 0.78. Based on dielectric electrophoretic adsorption theory and colloidal kinetics, the impurity deposition density of free carbon at the oil-paper interface is calculated. The impurity deposition density at the oil-paper interface is quantitatively correlated with the suspended concentration in the oil, the normal electric field strength at the interface, and the operating oil temperature, satisfying the following in equilibrium:

[0052] In the formula, To measure the particle concentration, The normal electric field strength at the oil-paper interface is obtained from simulation. To operate at thermodynamic temperature, K is the gas constant; K is the adsorption equilibrium constant. For electric field influence index, The adsorption activation energy corresponds to the impurity type K. , The values ​​are calibrated through regression analysis using a simulated sample database.

[0053] Based on the measured particle concentration of free carbon and operating conditions in this embodiment, the impurity deposition density of free carbon at the oil-paper interface was calculated to be: .

[0054] The concentrations of the other two impurity types were below the detection limit, and their respective characteristic parameters were all zero.

[0055] In contrast, the same oil sample was tested using an automatic particle counter, and the results showed that the oil sample contained free carbon impurities, and The concentration of free carbon particles was only The particle size was [number] particles / mL, and other minute impurities were not detected, meeting the relevant standard requirements. This comparison shows that traditional detection methods cannot reflect particles smaller than [number] particles. The accumulation of free carbon particles.

[0056] The extracted characteristic parameters of minute solid impurities are input into the oil-paper insulation fault assessment model adapted to the actual operating conditions of the transformer under test. The model calculates and outputs the comprehensive electrical performance assessment value of the oil-paper insulation system of the transformer under test, including the predicted breakdown voltage, the predicted partial discharge initiation voltage, and the predicted surface flashover voltage.

[0057] Based on the transformer insulation operation standards in GB / T 7595 "Quality of Transformer Oil in Operation" and DL / T 596 "Preventive Testing Procedures for Power Equipment", voltage thresholds for the electrical performance of transformer oil-paper insulation are set. These thresholds include at least the breakdown voltage threshold, the partial discharge initiation voltage threshold, and the surface flashover voltage threshold. Each threshold has two risk boundary values: the first threshold (low-risk boundary value) serves as the insulation qualification benchmark, and the second threshold (high-risk boundary value) serves as the insulation warning threshold. The predicted breakdown voltage, partial discharge initiation voltage, and surface flashover voltage are compared with the corresponding two-level risk boundary values ​​to comprehensively determine the insulation failure risk level of the transformer's oil-paper insulation system, and an assessment report is generated.

[0058] Example 2

[0059] Standard oil-paper insulation samples containing free carbon particles of different concentrations and sizes were prepared under laboratory conditions to conduct comparative verification tests on the method of the present invention.

[0060] The same type of mineral insulating oil as in Example 1 was selected as the base oil. The base oil was purified before use: firstly, using… Polytetrafluoroethylene (PTFE) filter membranes were vacuum-filtered to remove existing solid impurities; then dried and degassed for 48 hours at 60°C and a vacuum of less than 50 Pa to reduce the moisture content to below 10 ppm. After purification, a microfluidic imaging system was used to confirm that the particle size in the oil was larger than [missing value]. The particle concentration is less than 50 particles / mL.

[0061] High-density insulating paperboard for transformers with a thickness of 1.5 mm was selected and cut into circular test pieces with a diameter of 60 mm. The test pieces were dried in a vacuum drying oven at 80℃ for 48 hours, and then immersed in purified insulating oil at 60℃ and a vacuum degree of less than 50 Pa for 24 hours.

[0062] Free carbon particles were prepared in pure insulating oil using an arc discharge method: Two copper electrodes were immersed in 500 mL of pure insulating oil, and a 10 kV AC voltage was applied to perform multiple arc discharges. Free carbon particles in the oil were collected after the discharges. The carbon-containing oil sample was then sequentially passed through a 5 mm pore size... 3 1 The nuclear pore membrane was used for graded filtration to obtain particles with a median particle size of approximately 4 μm. 2 1 The three scores have a geometric standard deviation of less than 1.5 for each score.

[0063] Three particle size fractions of free carbon particles were added to the purified base oil at set concentrations to prepare nine groups of carbon-containing oil samples. Another group of pure oil samples without free carbon particles was prepared as a control group.

[0064] A ball-plate electrode structure was used as the standard test model. The high-voltage electrode was a polished copper ball with a diameter of 12.5 mm, and the low-voltage electrode was a polished copper plate with a diameter of 75 mm. An impregnated insulating paperboard sample was placed in the center of the low-voltage plate electrode, and the high-voltage ball electrode was placed vertically above the insulating paperboard. The oil gap distance between the ball electrode and the paperboard surface was controlled to be 3 mm using a precision fine-tuning device. The assembled electrode system was placed in a sealed acrylic test container, and the corresponding oil sample was injected. Three parallel samples were prepared for each parameter group, for a total of 30 standard samples.

[0065] The test was conducted using a power frequency breakdown voltage test system, with the voltage increased from zero at a rate of 2kV / s until insulation breakdown (based on sudden changes in circuit current and voltage drops), and the breakdown voltage value was recorded. Each sample was tested five times, and the average value was taken. The test environment temperature was 25±2℃, and the relative humidity was <45%.

[0066] The partial discharge test system conforming to IEC 60270 was used for testing. The voltage was increased incrementally in 0.5kV increments, with each voltage level held for 30 seconds. The voltage value at the first occurrence of a discharge pulse with an apparent discharge quantity ≥5pC was recorded. Each sample was tested three times, and the average value was taken.

[0067] After the breakdown test, observe the flashover marks on the surface of the insulating paperboard and record the voltage value when obvious carbonization channels appear.

[0068] Based on the actual structural parameters of the standard specimen, a simulation model of the ball-plate electrode electro-thermal coupling field, completely corresponding to the geometry of the standard specimen, was established in COMSOL Multiphysics: the ball electrode diameter is 12.5 mm, the plate electrode diameter is 75 mm, the insulating paperboard thickness is 1.5 mm, and the oil gap distance is 3 mm. The material parameters are consistent with the actual material of the standard specimen, and the boundary conditions are consistent with the actual test conditions.

[0069] Free carbon particles were introduced into the simulation model according to the electric field intensity gradient distribution. Thirty independent simulations were performed on each set of particle parameters (with the initial random distribution changed) to obtain multiple sets of simulation samples. Impurity characteristic parameters and corresponding electrical performance parameters were collected in each simulation.

[0070] After preprocessing, the simulation data was divided into training and validation sets in a 7:3 ratio. A gradient boosting regression tree algorithm was used to establish a mapping-related model, with hyperparameters the same as in Example 1. The validation set R after iteration... 2 Training stops when the preset threshold is reached. The particle characteristic parameters of each group of samples were input into the model, and the predicted values ​​of electrical performance of each group were calculated and compared with the measured values. The average prediction error of breakdown voltage was 1.13%, the average prediction error of partial discharge initiation voltage was 1.29%, and the average prediction error of surface flashover voltage was 1.28%. The maximum error of all predictions was less than 2%, which showed high evaluation accuracy.

[0071] The verification results demonstrate that the method of this invention can accurately capture the nonlinear influence of minute solid impurities on the electrical properties of oil-paper insulation, and the evaluation results are highly accurate, effectively compensating for the limitations of traditional detection and evaluation methods on 5... The following technical blind spots, where minute impurities are not sensitive, provide reliable technical support for early fault warning and condition assessment of transformer oil-paper insulation.

[0072] Obviously, the embodiments described above are only some, not all, of the embodiments of this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without inventive effort are within the scope of protection of this invention.

Claims

1. A method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil, characterized in that, The method includes: S1. Based on the typical structural parameters of transformer oil-paper insulation, the dielectric properties and thermophysical properties of insulating oil and insulating paperboard, establish an electro-thermal coupling field simulation model of the transformer oil-paper insulation system. S2. Introduce tiny solid impurity particles into the electro-thermal coupled field simulation model, set multiple gradients of local overheating temperature and multiple energy levels of partial discharge conditions, and conduct simulation experiments under the combined action of electro-thermal to obtain multiple sets of characteristic parameters of tiny solid impurities and corresponding electrical performance parameters of oil-paper insulation; the characteristic parameters of tiny solid impurities include at least impurity type, particle concentration, particle size distribution and morphological characteristics; the electrical performance parameters include at least breakdown voltage, partial discharge initiation voltage and surface flashover voltage; S3. Based on the acquired multiple sets of simulation data, a mapping correlation model between the characteristic parameters of minute solid impurities and the electrical performance parameters of oil-paper insulation is established using the gradient boosting regression tree algorithm. S4. Obtain the actual operating condition parameters of the transformer under test, and modify the mapping association model based on the actual operating condition parameters to generate an oil-paper insulation fault assessment model that is adapted to the operating conditions of the transformer under test. S5. Extract the characteristic parameters of tiny solid impurity particles in the insulating oil of the transformer under test, input the characteristic parameters into the oil-paper insulation fault assessment model, and calculate the electrical performance assessment value of the oil-paper insulation system of the transformer under test. S6. Based on the electrical performance evaluation value and the preset electrical performance voltage threshold of the oil-paper insulation, determine the insulation failure risk level of the oil-paper insulation system of the transformer under test, and output the oil-paper insulation fault risk assessment result.

2. The method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil according to claim 1, characterized in that: In step S1, the typical structural parameters of the transformer oil-paper insulation include at least the type of oil-paper insulation electrode structure, the thickness of the insulation paperboard, the oil gap distance, and the size of the insulation oil channel. The dielectric properties parameters include at least the relative permittivity, volume resistivity, and dielectric loss factor. The thermophysical parameters include at least thermal conductivity, specific heat capacity, and viscosity coefficient.

3. The method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil according to claim 1, characterized in that: In step S1, the establishment of the electro-thermal coupling field simulation model specifically includes: Construct a geometric structural model of the transformer oil-paper insulation system, and reconstruct the electrode structure, insulation paperboard stacking structure and oil channel distribution pattern based on typical structural parameters; Set the dielectric and thermal properties parameters of the transformer insulating oil and insulating paperboard, and set the correlation function of the dielectric parameters with temperature. Set the electric field boundary conditions and temperature field boundary conditions of the transformer oil-paper insulation system. The electric field boundary conditions include AC / DC composite voltage excitation and grounding constraint. The temperature field boundary conditions include ambient temperature and heat source distribution. The heat source distribution is determined by conductor loss and dielectric loss. The constructed geometric model is meshed, and the electric field control equation and thermal field control equation are solved by coupled iteration using the finite element method to obtain the reference electric field distribution and reference temperature field distribution of the transformer oil-paper insulation system under impurity-free conditions.

4. The method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil according to claim 3, characterized in that: In solving the electric field control equation: based on the solution results of the thermal field control equation, the dielectric parameter in the electric field control equation is corrected by calculating the correlation function of the dielectric parameter changing with temperature; In solving the thermal field control equation: based on the solution results of the electric field control equation, the heat source term of the thermal field control equation is corrected by updating the heat source distribution parameters.

5. The method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil according to claim 1, characterized in that: In step S2, the simulation experiment specifically includes: Set at least three temperature levels of local overheating temperature gradient and at least three energy levels of local discharge energy gradient to construct multiple sets of electro-thermal combined action conditions; In the oil gap region and / or oil-paper interface region of the electro-thermal coupled field simulation model, tiny solid impurity particles are introduced according to a preset distribution method, and initial characteristic parameters of the tiny solid impurity particles are set; the preset distribution method includes random distribution or distribution according to electric field intensity gradient; the initial characteristic parameters include at least impurity type, initial particle concentration, initial particle size distribution, and initial morphology characteristic parameters; the impurity type includes free carbon particles, fiber particles, and metal particles; The electro-thermal coupled field simulation model was run under each group of electro-thermal combined working conditions. The tiny solid impurity particles migrated and aggregated under the combined action of the electric field and the temperature field. After each simulation run, the characteristic parameters of the tiny solid impurity particles under the current working condition are collected, and the electrical performance parameters of the oil-paper insulation under the corresponding working condition are collected simultaneously.

6. The method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil according to claim 1, characterized in that: In step S3, the gradient boosting regression tree algorithm is used to establish the mapping association model, specifically including: Outlier removal and normalization were performed on the acquired simulation data to obtain a standardized simulation sample dataset. The standardized simulation sample dataset is divided into a training set and a validation set according to a preset ratio. The feature parameters of tiny solid impurities are set as the model input features, and the electrical performance parameters of oil paper insulation are set as the model output labels. A gradient boosting regression tree initial model is constructed, wherein the gradient boosting regression tree initial model uses the mean squared error as the loss function and adopts a forward step-by-step addition model to generate a CART regression tree round by round. In each iteration, the current regression tree is fitted based on the prediction residuals of the model after the previous iteration, and the node splitting rules and leaf node output weights are updated. After each round of training, the determination coefficients of the model are calculated using the validation set. When the determination coefficient Training stops when a preset threshold is reached or the number of iterations reaches a set upper limit, thus obtaining a mapping correlation model between the characteristic parameters of minute solid impurities and the electrical performance parameters of oil-paper insulation after training is completed.

7. The method for assessing transformer oil paper insulation faults based on the characteristics of minute impurities in transformer oil according to claim 1, characterized in that: In step S4, based on the actual operating conditions of the transformer under test, the mapping correlation model is corrected through adaptive weighted incremental training. The adaptive weighted incremental training specifically includes: Obtain the actual operating condition parameter vector of the transformer under test , where m is the number of operating condition parameters, and the operating condition parameters include at least voltage level, service life and operating oil temperature; For the j-th simulation sample data from the historical simulation experiment, obtain the simulation condition parameter vector corresponding to the simulation sample data. Calculate the weighted feature distance between the simulated operating condition parameter vector and the actual operating condition parameter vector: in, The preset weighting coefficient for the k-th operating condition parameter is... The standard deviation of the k-th operating condition parameter across all simulation samples; the weighting coefficient The sensitivity of the electrical performance of the oil-paper insulation to various operating parameters is preset, and the higher the sensitivity, the greater the weight. Based on the weighted feature distance Determine the weight value of the j-th simulation sample data. The weight value and They are inversely proportional, and the sum of the weights of all simulated sample data is 1; Weight values ​​based on each simulation sample data The training set of the mapping correlation model is weighted, and the weighted mean square error is used as the loss function. The incremental gradient boosting regression algorithm is used to continue training on the original mapping correlation model until the error of the validation set no longer decreases, so as to obtain an oil-paper insulation fault assessment model that is adapted to the actual operating conditions of the transformer under test.

Images