Electric field analysis and optimization method for transformer withstand voltage and partial discharge test platform

Abstract

The invention discloses an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform, and the method comprises the steps: building a digital model of the test platform, obtaining a preparation technology standard and historical manufacturing measurement data of a target test platform, defining an influence variable which influences the distribution of an electric field, simulating the distribution of the electric field under different preparation differences, and carrying out the optimization of the electric field. Establishing an electric field simulation database; identifying test platform electric field sensitive areas under different preparation differences by adopting a spectral clustering algorithm based on the electric field simulation database, constructing a knowledge graph, constructing a sensitive area identification model based on a graph neural network, and performing model training through the knowledge graph; structural features and preparation features of the to-be-analyzed test platform are obtained, a local sensitive area of the to-be-analyzed test platform is recognized in the sensitive area recognition model, and if the local sensitive area exists, an optimization scheme is formulated for test platform optimization assistance, so that the design reliability and operation safety of the transformer test platform are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of transformer withstand voltage and partial discharge test platform design and optimization, and particularly relates to an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform. BACKGROUND

[0002] In the research and development and engineering application of the transformer withstand voltage and partial discharge test platform, the accurate analysis of the electric field distribution and the optimization of the insulation structure are always the core challenges to ensure the effectiveness of the test and the safety of the equipment. The traditional method usually estimates the electric field strength based on the empirical formula and the simplified model, and relies on the static finite element simulation technology to predict the field strength extreme value under specific working conditions, but such method has significant limitations: firstly, the modeling precision of the electromagnetic-thermal multi-physical field coupling effect of the complex three-dimensional structure in the test platform is insufficient, especially in the high voltage gradient area, the grid discretization error easily leads to distortion of the field strength calculation; secondly, the statistical influence of the uncertainty factors such as manufacturing assembly tolerance, material parameter batch fluctuation and contact interface state change on the electric field distribution is not fully considered, so that there is systematic deviation between the simulation results and the measured data; thirdly, the existing sensitive area identification technology mostly uses fixed threshold segmentation or manual experience calibration, which is difficult to adapt to different test voltage levels, insulation medium aging degrees and other dynamic conditions, resulting in large positioning error of the partial discharge starting point.

[0003] In view of the above problems, an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform is proposed to break through the technical bottlenecks of the traditional method in terms of analysis accuracy, optimization efficiency and engineering applicability, and to improve the design reliability and operation safety of the transformer test platform. SUMMARY

[0004] The present application overcomes the defects of the prior art and provides an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform.

[0005] To achieve the above purpose, the first aspect of the present application provides an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform, comprising:

[0006] A test platform digital model is constructed, the preparation process standard and the historical manufacturing measurement data of the target test platform are obtained, the influence variables affecting the electric field distribution are defined, the electric field distribution under different preparation differences is simulated, and an electric field simulation database is established;

[0007] Based on the electric field simulation database, a spectral clustering algorithm is used to identify the test platform electric field sensitive area under different preparation differences and construct a knowledge graph, a sensitive area identification model is constructed based on a graph neural network, and the model is trained through the knowledge graph;

[0008] Obtaining the structural characteristics and preparation characteristics of the test platform to be analyzed, identifying the local sensitive area of the test platform to be analyzed by using a sensitive area identification model, and if the local sensitive area exists, formulating an optimization scheme to assist in optimizing the test platform.

[0009] In the scheme, the test platform digital model is constructed, specifically including:

[0010] Obtaining the design specification of the target test platform, extracting the three-dimensional design parameters of the target test platform according to the obtained design specification, constructing a geometric model of the target test platform based on the three-dimensional design parameters of the target test platform by using three-dimensional modeling technology, and obtaining and defining the material properties of the target test platform by using a big data network;

[0011] Based on the geometric model of the target test platform, the test platform construction characteristics are extracted, the heat source analysis is combined with the design specification, the heat source type and heat source area are defined, and the test platform heat source analysis information is obtained;

[0012] According to the test platform heat source analysis information, a test platform CFD model is constructed by using CFD software in combination with the geometric model of the target test platform, and the physical parameter assignment and electrical conductivity parameter setting of different regions in the CFD model are performed through the material properties, the two-way interaction relationship between the electric field and the temperature field is established by coupling the Joule heat generation equation and the heat conduction equation;

[0013] The test platform CFD model is meshed, the initial calculation grid is generated based on the triangle subdivision method, the grid quality is defined by maximizing the minimum dihedral angle of the tetrahedral element, and the initial meshing is completed;

[0014] After the initial meshing, the initial grid is calculated based on the Poisson equation to obtain the preliminary field strength distribution, and the electric field gradient amplitude is extracted as the basis for grid optimization through the preliminary field strength distribution;

[0015] If the electric field gradient amplitude is greater than the preset threshold value, the local grid encryption operation is triggered, the target element is recursively subdivided by using the iterative convergence strategy based on the h-adaptive method through the edge bisection and surface reconstruction technology, and a transition layer element is inserted in the adjacent element;

[0016] The electric field distribution is recalculated after each grid encryption is completed, and the test platform digital model is output after the preset stopping criterion is met.

[0017] In the scheme, the preparation process standard and the historical manufacturing measurement data of the target test platform are obtained, the influence variables affecting the electric field distribution are defined, the electric field distribution under different preparation differences is simulated, and an electric field simulation database is established, specifically including:

[0018] Obtain the preparation process standard and historical manufacturing measurement data of the target test platform, define the influence variables affecting the electric field distribution, including electrode assembly position deviation, insulation medium thickness tolerance, batch fluctuation of material dielectric constant, and contact interface resistance change;

[0019] Extract the historical manufacturing measurement features corresponding to each influence variable from the historical manufacturing measurement data, compare with the preparation process standard, analyze the preparation differences in the actual manufacturing process, and generate the difference probability distribution of each influence variable based on the preparation differences;

[0020] Use Latin hyper-sampling method to combine historical manufacturing measurement data and difference probability distribution of each influence variable, randomly combine different influence variables to construct a parameter combination sample set with different categories and different degrees of preparation differences;

[0021] Obtain the test platform digital model, import the constructed parameter combination sample set as model input into the test platform digital model for simulation analysis, analyze the electric field distribution under different parameter combinations based on the finite element solver, and obtain the simulation result data set;

[0022] Extract the global electric field intensity distribution features from the simulation result data set, including field strength peak position, maximum gradient area, field strength value of partial discharge sensitive point and grid cell position, associate the global electric field intensity distribution features with the corresponding input parameter combination to construct the mapping relationship between preparation differences and electric field response, and generate the electric field simulation database.

[0023] In this scheme, the electric field simulation database is used to identify the test platform electric field sensitive area under different preparation differences based on the spectral clustering algorithm, and a knowledge graph is constructed, specifically including:

[0024] Obtain the electric field simulation database, extract the electric field simulation features of each preparation difference combination from the electric field simulation database, map the unit field strength to the risk probability value based on the extracted electric field simulation features, and use it as the risk level of local discharge at different positions;

[0025] Construct a feature space according to the obtained electric field simulation features and risk probability values, calculate the similarity weight between any grid cells in the feature space and construct a similarity matrix using an adaptive Gaussian kernel function, wherein the bandwidth parameter of the Gaussian kernel function is determined by local density estimation;

[0026] Construct a weighted undirected graph using the similarity matrix, wherein the graph nodes represent grid cells and the edge weights represent feature similarity, and calculate the normalized Laplacian matrix through the weighted undirected graph;

[0027] The normalized Laplace matrix calculated is subjected to eigenvalue decomposition to obtain eigenvectors corresponding to the first k smallest eigenvalues, the obtained eigenvectors are mapped to a low-dimensional embedding space, and a spectral clustering algorithm is introduced to identify sensitive areas;

[0028] The elbow rule is used to determine the optimal number of clusters, the silhouette coefficient and the sum of intra-class distances corresponding to different cluster numbers are calculated, a cluster number change curve is generated, and the k value corresponding to the inflection point of the curve is selected as the optimal classification number;

[0029] The low-dimensional eigenvectors are divided by the K-means algorithm, the low-dimensional eigenvectors are clustered by minimizing the sum of Euclidean distances of intra-class samples according to iterative optimization of the center position, and a plurality of class clusters are output after clustering is completed;

[0030] The field strength mean, partial discharge probability peak and gradient distribution skewness of each class cluster are calculated, and are respectively compared with preset thresholds, and the sensitive area is marked based on the judgment result to obtain sensitive area identification information;

[0031] The sensitive area identification information is associated with each preparation difference combination in the electric field simulation database, and a knowledge graph with preparation difference-electric field simulation feature-sensitive area as a meta-path is constructed.

[0032] In the scheme, the sensitive area identification model is constructed based on the graph neural network, and the model is trained through the knowledge graph, and specifically includes:

[0033] The knowledge graph is obtained, and the meta-random walk algorithm is used to perform random walk in the knowledge graph with the preparation difference as the starting entity and the sensitive area as the terminal entity, and a plurality of meta-paths from the starting entity to the terminal entity in the knowledge graph are sampled;

[0034] The electric field simulation features corresponding to each meta-path are extracted, and the corresponding electric field simulation features are used as additional features of the preparation difference node to construct a heterogeneous information network containing the preparation difference node, the sensitive area node and the feature relationship edge;

[0035] The heterogeneous information network is represented by a graph, the similarity values between the meta-paths are calculated according to the graph representation, a preset meta-path similarity threshold is set, and a strong correlation path group is screened, and the neighborhood relationship of the preparation difference node and the sensitive area node contained therein is merged;

[0036] For each target node, the 1-hop neighbor node set thereof across the meta-paths is aggregated, a dynamic weighted adjacency matrix is constructed, the matrix element value is determined by the weighted product of the meta-path similarity and the feature similarity between the nodes, and the weight coefficient is optimized by grid search;

[0037] The sensitive area recognition model is constructed based on a graph neural network, the dynamic weighted adjacency matrix is introduced into the sensitive area recognition model for model training, and a multi-head attention mechanism is introduced to calculate adaptive weights between nodes;

[0038] The output features of each attention head are spliced by using the adaptive weights obtained by calculation, and a gating fusion model is used for integration, and a message passing function is used for node representation update;

[0039] In the last layer of the network, hierarchical mean pooling is performed on all node representations to extract an implicit feature vector of the global graph, which is input into a fully connected layer to be mapped into an electric field analysis result. The result is verified using verification data, and when it meets the preset standard, the trained sensitive area recognition model is output.

[0040] In the scheme, the structure characteristics and preparation characteristics of the test platform to be analyzed are obtained, and the local sensitive area of the test platform to be analyzed is recognized in the sensitive area recognition model. If there is a local sensitive area, an optimization scheme is formulated to assist in optimizing the test platform, which specifically includes:

[0041] The structure characteristics and preparation characteristics of the test platform to be analyzed are obtained and input into a test platform digital model to perform electric field distribution simulation to obtain an electric field distribution simulation result. The electric field distribution simulation result and the structure characteristics and preparation characteristics of the test platform to be analyzed are input into the sensitive area recognition model to recognize the local sensitive area of the test platform to be analyzed, and sensitive area recognition information is obtained.

[0042] A big data network is introduced, and different historical test platform optimization instances are obtained through the big data network. The electric field distribution characteristics, structure characteristics and preparation characteristics of the test platform to be analyzed are calculated for similarity with each historical test platform optimization instance.

[0043] A number of historical test platform optimization instances within a preset similarity range are selected by calculating the obtained similarity values, and the corresponding historical test platform optimization schemes are extracted to form an initial parameter space. A multi-objective grey wolf optimization algorithm is introduced to formulate an optimization scheme for the test platform to be analyzed.

[0044] An initial grey wolf population is generated by random sampling from the initial parameter space, a preset objective function is set, and a constraint condition is set. The objective function value of each individual in the initial grey wolf population is calculated by the objective function. All individuals in the population are divided into different front levels by non-dominated sorting according to the calculated objective function values.

[0045] The congestion distance is obtained by congestion calculation for each front level, the individual with the maximum congestion distance in each level is selected as the leader wolf, the remaining individuals are selected as the follower wolves, and the direction vector of the leader wolf in the decision space is calculated for position updating.

[0046] Output the optimal solution set after repeated iteration optimization until the stop condition is met, and generate an optimal platform optimization scheme for the target test platform to be analyzed according to the optimal solution set, and push it.

[0047] The second aspect of the present application provides a computer readable storage medium, characterized in that the computer readable storage medium comprises a transformer withstand voltage and partial discharge test platform electric field analysis and optimization method program, and the transformer withstand voltage and partial discharge test platform electric field analysis and optimization method program is executed by a processor to realize the steps of the transformer withstand voltage and partial discharge test platform electric field analysis and optimization method according to any one of the above.

[0048] The present application discloses a transformer withstand voltage and partial discharge test platform electric field analysis and optimization method, comprising: constructing a test platform digital model, obtaining the preparation process standard and historical manufacturing measurement data of the target test platform, defining the influence variables affecting the electric field distribution, simulating the electric field distribution under different preparation differences, and establishing an electric field simulation database; identifying the test platform electric field sensitive area under different preparation differences based on the electric field simulation database using a spectral clustering algorithm and constructing a knowledge graph, constructing a sensitive area identification model based on a graph neural network, and training the model through the knowledge graph; obtaining the structure characteristics and preparation characteristics of the test platform to be analyzed, identifying the local sensitive area of the test platform to be analyzed using the sensitive area identification model, and if there is a local sensitive area, formulating an optimization scheme to assist in optimizing the test platform, thereby improving the design reliability and operation safety of the transformer test platform. BRIEF DESCRIPTION OF DRAWINGS

[0049] In order to more clearly illustrate the technical solutions in the embodiments or examples of the present application, the drawings needed in the embodiment or example description will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and those skilled in the art can obtain other drawings according to the drawings shown without creating any creative labor.

[0050] Figure 1 A flow chart of a transformer withstand voltage and partial discharge test platform electric field analysis and optimization method is provided for an embodiment of the present application.

[0051] Figure 2 An optimization scheme analysis flow chart for a transformer withstand voltage and partial discharge test platform is provided for an embodiment of the present application.

[0052] The implementation, functional features and advantages of the present application will be further described with reference to the embodiments and the accompanying drawings. DETAILED DESCRIPTION

[0053] In order to enable a more clear understanding of the above-mentioned objects, features and advantages of the present application, the present application will be further described in detail below with reference to the accompanying drawings and specific embodiments. It should be noted that the embodiments of the present application and the features in the embodiments can be combined with each other without conflict.

[0054] In the following description, a large number of specific details are set forth in order to facilitate a thorough understanding of the present application, but the present application can also be implemented in other manners different from those described herein, and therefore, the scope of protection of the present application is not limited to the specific embodiments disclosed below.

[0055] Figure 1 A flow chart of an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform is provided for an embodiment of the present application;

[0056] As Figure 1 shown, the present application provides a flow chart of an electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform, comprising:

[0057] S102, constructing a test platform digital model, obtaining preparation process standards and historical manufacturing measurement data of a target test platform, defining influence variables affecting electric field distribution, simulating electric field distribution under different preparation differences, and establishing an electric field simulation database;

[0058] S104, identifying electric field sensitive areas of the test platform under different preparation differences based on the electric field simulation database using a spectral clustering algorithm and constructing a knowledge graph, constructing a sensitive area identification model based on a graph neural network, and training the model through the knowledge graph;

[0059] S106, obtaining structure characteristics and preparation characteristics of a test platform to be analyzed, identifying local sensitive areas of the test platform to be analyzed using the sensitive area identification model, and if there are local sensitive areas, formulating an optimization scheme to assist in optimizing the test platform.

[0060] Further, in a preferred embodiment of the present application, the construction of the test platform digital model specifically comprises:

[0061] Obtaining a design specification of the target test platform, extracting three-dimensional design parameters of the target test platform according to the obtained design specification, constructing a geometric model of the target test platform based on the three-dimensional design parameters of the target test platform using three-dimensional modeling technology, and obtaining and defining material properties of the target test platform using a big data network;

[0062] Extracting test platform construction features based on the geometric model of the target test platform, combining with heat source analysis of the design specification, defining heat source types and heat source areas, and obtaining test platform heat source analysis information;

[0063] According to the heat source analysis information of the test platform, a CFD model of the test platform is constructed by using CFD software in combination with a geometric model of the target test platform, and physical parameter assignment and electrical conductivity parameter setting are performed on different regions in the CFD model through material properties, a two-way interaction relationship between an electric field and a temperature field is established by coupling a Joule heat generation equation and a heat conduction equation;

[0064] The CFD model of the test platform is meshed, an initial calculation grid is generated based on a triangle partitioning method, grid quality is defined by maximizing the minimum dihedral angle of a tetrahedral unit, and initial meshing is completed;

[0065] After the initial meshing, a preliminary field strength calculation is performed on the initial grid based on a Poisson equation solution to obtain a preliminary field strength distribution, and an electric field gradient amplitude is extracted from the preliminary field strength distribution as a basis for grid optimization;

[0066] If the electric field gradient amplitude is greater than a preset threshold, a local grid densification operation is triggered, an iterative convergence strategy is adopted based on an h-adaptive method to recursively subdivide target units through edge bisection and surface reconstruction techniques, and transition layer units are inserted in adjacent units;

[0067] The electric field distribution is recalculated after each grid densification is completed, and the test platform digital model is output after the preset stopping criterion is met.

[0068] It should be noted that first, three-dimensional geometric parameters are extracted from the design specification of the target test platform, including electrode structure size, insulating medium layout, cooling channel topology, and other key structural features. These parameters are converted into an accurate geometric model through a three-dimensional modeling software (such as SolidWorks or CATIA), and the dielectric constant, thermal conductivity, electrical conductivity, and other properties of the materials of each component are matched through a material database (such as CESSelector or Granta MI) to ensure the consistency of the model physical parameters with the actual test platform. For non-standard materials or composite insulating media, the nonlinear electrical conductivity model needs to be fitted based on historical test data (such as partial discharge tolerance curves and thermal aging test reports) in the big data platform to define the dynamic material properties under field strength-temperature coupling. Subsequently, based on the geometric model, the heat source distribution characteristics of the test platform are extracted. According to the current load, voltage level, and loss calculation formula in the design specification, the type of heat source (such as winding eddy current loss and insulating medium polarization loss) and its spatial distribution mode are determined. The rationality of heat source region division is verified by using thermal imaging historical data, the heat flux density of high loss areas (such as electrode contact surfaces and insulator surfaces) is calibrated, and a heat source analysis information table containing heat source position, power density, and boundary conditions is generated to provide input conditions for subsequent multi-physical field coupling.

[0069] Further, based on the heat source analysis results, an electromagnetic-thermal coupling model of the test platform is constructed using a computational fluid dynamics (CFD) software (such as ANSYS Fluent or COMSOL Multiphysics). In the spatial domain of the geometric model, material properties are respectively assigned to different functional areas. By coupling the Joule heat generation equation (describing the heat generated by the current) and the transient heat conduction equation (describing the heat transfer process), a two-way mechanism of temperature field evolution under electric field excitation is established. Meanwhile, the thermal expansion effect is introduced to correct the feedback influence of geometric deformation on electric field distribution. In the mesh division stage, an unstructured tetrahedral mesh generation strategy is adopted, and the initial calculation mesh is constructed by the Delaunay triangulation algorithm. To ensure the stability of the calculation, the minimum dihedral angle threshold of the tetrahedral element is defined, and the grid optimization algorithm (such as Laplacian smoothing) is used to eliminate the abnormal elements. The Jacobian matrix determinant value is used to quantify the grid quality. After completing the initial mesh division, the static electric field is preliminarily solved based on the Poisson equation, and the global electric field gradient amplitude distribution data is extracted. The gradient threshold is set as the trigger condition for adaptive grid refinement: when the gradient amplitude of a certain element exceeds the threshold, the h-adaptive method is used to recursively bisect the element, and the pyramid transition layer is inserted between adjacent elements through surface reconstruction technology to avoid numerical oscillation caused by sudden change of grid size. After each grid refinement, the electric field distribution is re-solved until the element size of all high gradient regions reaches the preset precision. Finally, the digital twin model of the test platform that meets the requirements of multi-physical field coupling calculation is output, providing a high-fidelity simulation basis for subsequent partial discharge risk assessment and optimization design.

[0070] Further, in a preferred embodiment of the present application, the preparation process standard and historical manufacturing measurement data of the target test platform are obtained, the influence variables affecting the electric field distribution are defined, the electric field distribution under different preparation differences is simulated, and an electric field simulation database is established, specifically including:

[0071] The preparation process standard and historical manufacturing measurement data of the target test platform are obtained, and the influence variables affecting the electric field distribution are defined, including electrode assembly position deviation, insulation medium thickness tolerance, batch fluctuation of material dielectric constant, and contact interface resistance change;

[0072] From the historical manufacturing measurement data, the historical manufacturing measurement characteristics corresponding to each influence variable are extracted, compared with the preparation process standard, the preparation differences in the actual manufacturing process are analyzed, and the difference probability distribution of each influence variable is generated based on the preparation differences;

[0073] The Latin hyper-sampling method is used to combine the historical manufacturing measurement data and the difference probability distribution of each influence variable, and a plurality of parameter combination sample sets with different categories and different degrees of preparation differences are constructed by randomly combining different influence variables;

[0074] The simulation result data set is obtained by analyzing the electric field distribution under different parameter combinations based on a finite element solver.

[0075] Global electric field intensity distribution characteristics are extracted from the simulation result data set, including field strength peak position, maximum gradient region, field strength value of partial discharge sensitive point, and grid cell position. The mapping relationship between preparation difference and electric field response is constructed by correlating the global electric field intensity distribution characteristics with the corresponding input parameter combination, and an electric field simulation database is generated.

[0076] It should be noted that first, the preparation process standard file of the target test platform is obtained (including assembly accuracy requirements, material acceptance specifications, etc.) and the historical manufacturing measurement database. Key influence variables are identified through process failure mode analysis, including electrode assembly position deviation, insulating medium thickness tolerance in the geometric level, dielectric constant batch fluctuation and random change of contact interface resistance in the material performance level. Based on the comparison between the preparation process standard and the actual measurement data, the preparation difference in the actual manufacturing process is analyzed, and the difference probability distribution of each influence variable is generated based on the preparation difference, reflecting the multi-variable interaction characteristics in the real manufacturing. The parameter combination sample set is constructed by using the improved Latin hypercube sampling method. On the basis of traditional LHS, the maximum and minimum distance optimization criterion is introduced, and the space distribution of sample points is adjusted iteratively by simulated annealing algorithm, so that the sample points in high-dimensional parameter space uniformly cover the extreme working condition combinations of each variable. Each sample represents a specific combination of preparation differences, while meeting the probability constraints of marginal distribution and the correlation constraints between variables, ensuring that the sampling results not only conform to the statistical law but also have engineering representativeness.

[0077] Further, the parameter sample set is input into the test platform digital twin model for batch simulation. Through the finite element solver (such as ANSYS Maxwell or COMSOL), the geometric parameters of the model (such as electrode coordinate translation, insulating layer thickness scaling) are dynamically modified, the dielectric constant and contact resistance value in the material property library are updated, and the boundary conditions are reset. After each simulation is completed, the moving cube algorithm is used to extract the peak coordinates of the three-dimensional field strength distribution and its gradient vector, and the region growing method is used to segment the high field strength cluster and mark the partial discharge sensitive points. At the same time, the topological information (cell ID, vertex coordinates, adjacency relationship) of each sensitive point in the grid cell is recorded, and the global electric field feature vector containing spatial position, field strength amplitude and gradient direction is constructed. Finally, the input parameter combination and the output electric field features are associated and mapped to form a multi-dimensional electric field simulation database, providing data-driven decision basis for subsequent optimization design.

[0078] Further, in a preferred embodiment of the present application, the electric field simulation database based on the spectral clustering algorithm identifies the electric field sensitive area of the test platform under different preparation differences, and constructs a knowledge graph, specifically including:

[0079] An electric field simulation database is obtained, electric field simulation features of each preparation difference combination are extracted from the electric field simulation database, and the unit field strength is mapped to a risk probability value based on the extracted electric field simulation features, as a risk level of local discharge at different positions;

[0080] A feature space is constructed according to the obtained electric field simulation features and risk probability values, and an adaptive Gaussian kernel function is used to calculate the similarity weight between any grid cells in the feature space and construct a similarity matrix, wherein the bandwidth parameter of the Gaussian kernel function is determined by local density estimation;

[0081] A weighted undirected graph is constructed using the similarity matrix, wherein the graph nodes represent the grid cells, and the edge weight represents the feature similarity degree. A normalized Laplacian matrix is calculated through the weighted undirected graph;

[0082] The calculated normalized Laplacian matrix is subjected to feature decomposition to obtain the feature vectors corresponding to the first k smallest eigenvalues, the obtained feature vectors are mapped to a low-dimensional embedding space, and a spectral clustering algorithm is introduced for sensitive area identification;

[0083] The elbow rule is used to determine the optimal number of clusters, the silhouette coefficient and the sum of intra-class distances corresponding to different cluster numbers are calculated, a cluster number change curve is generated, and the k value corresponding to the inflection point of the curve is selected as the optimal number of clusters;

[0084] The low-dimensional feature vectors are divided by the K-means algorithm, the low-dimensional feature vectors are clustered by minimizing the sum of Euclidean distances of intra-class samples according to iterative optimization of the centroid position, and a plurality of class clusters are output after clustering is completed;

[0085] The field strength mean, partial discharge probability peak and gradient distribution skewness of each class cluster are calculated, and are respectively compared with preset threshold values, the sensitive area is marked based on the judgment result, and sensitive area identification information is obtained;

[0086] The sensitive area identification information is associated with each preparation difference combination in the electric field simulation database, and a knowledge graph with preparation difference-electric field simulation feature-sensitive area as a meta-path is constructed.

[0087] It should be noted that firstly, the electric field feature data corresponding to different preparation difference combinations is extracted based on an electric field simulation database, including the field strength value, gradient distribution and partial discharge sensitive point coordinates of each grid unit. The field strength is mapped to a partial discharge risk probability value through a Weibull distribution model. The specific method is as follows: the shape parameter and scale parameter of the Weibull model are calibrated based on historical partial discharge experimental data, the cumulative failure probability corresponding to the field strength of each unit is calculated, the risk level of insulation breakdown occurring at different positions is quantified, and a global risk probability heat map is generated. Subsequently, a multi-dimensional feature space is constructed by fusing the field strength, gradient and risk probability, and the similarity weight between any two units in the feature space is calculated using an adaptive Gaussian kernel function. The bandwidth parameter of the kernel function is dynamically adjusted through local density estimation: a smaller bandwidth is used in the feature dense area (such as a high field strength cluster) to improve the resolution, and a larger bandwidth is used in the sparse area (such as a uniform electric field area) to avoid overfitting. A weighted undirected graph is constructed based on the similarity weight, and the nodes in the graph correspond to the grid units, and the edge weight represents the feature similarity between the units. By performing eigenvalue decomposition on the normalized Laplacian matrix, the feature vectors corresponding to the first k smallest eigenvalues are extracted, the original high-dimensional features are projected into a low-dimensional embedding space, noise interference is eliminated and the data manifold structure is preserved, and a more separable feature representation is provided for spectral clustering.

[0088] Further, in the clustering process, the elbow rule is used to determine the optimal number of clusters: the preset cluster number range is traversed, the silhouette coefficient (measuring the separation degree between classes and the compactness within classes) and the sum of intra-class distances corresponding to each candidate number are calculated, the index variation curve is drawn, and the k value corresponding to the curvature inflection point is selected. Subsequently, the K-means algorithm is applied to iteratively divide the low-dimensional feature vectors, and the centroid position is optimized by measuring the Euclidean distance, until the intra-class sample distance converges. After clustering, the field strength mean value, risk probability peak value and gradient distribution skewness of each class cluster are calculated, the high-risk clusters are selected by setting a threshold condition, and the sensitive areas are marked and the spatial coordinates and associated grid unit topology information are recorded. Finally, the sensitive area recognition result is associated with the preparation difference parameters in the electric field simulation database, and a knowledge graph in the form of triplets is constructed. In the graph, the preparation difference entity is taken as the starting point, the electric field simulation feature entity is taken as the intermediate node, and it is connected to the sensitive area entity, forming a meta-path of "preparation difference → electric field feature → sensitive area", which provides an interpretable decision basis for tolerance control and optimization of the test platform.

[0089] Further, in a preferred embodiment of the present application, the sensitive area recognition model is constructed based on a graph neural network, and the model is trained through the knowledge graph, specifically including:

[0090] Obtaining a knowledge graph, using a meta random walk algorithm to prepare a difference as a starting entity and a sensitive area as a terminal entity to perform random walk in the knowledge graph, and sampling a number of meta-paths from the starting entity to the terminal entity in the knowledge graph;

[0091] Extracting the electric field simulation features corresponding to each meta-path, and using the corresponding electric field simulation features as additional features of the preparation difference node to construct a heterogeneous information network containing the preparation difference node, the sensitive area node and the feature relationship edge;

[0092] Converting the heterogeneous information network into a graph representation, calculating the similarity values between each meta-path according to the graph representation, pre-setting a meta-path similarity threshold and screening a strong correlation path group, and merging the neighborhood relationships of the preparation difference nodes and the sensitive area nodes contained therein;

[0093] For each target node, aggregate its 1-hop neighbor node set across meta-paths, construct a dynamic weighted adjacency matrix, and determine the matrix element value by the weighted product of the meta-path similarity and the feature similarity between nodes, and the weight coefficient is optimized by grid search;

[0094] Based on the graph neural network, a sensitive area recognition model is constructed, the dynamic weighted adjacency matrix is introduced into the sensitive area recognition model for model training, and a multi-head attention mechanism is introduced to calculate the adaptive weight between nodes;

[0095] The output features of each attention head are spliced by the calculated adaptive weight and integrated using a gated fusion model, and a message passing function is used for node representation update;

[0096] In the last layer of network, all node representations are subjected to hierarchical mean pooling, the implicit feature vector of the global graph is extracted, input into a fully connected layer to map into an electric field analysis result, and the result is verified using verification data, and when it meets the preset standard, the trained sensitive area recognition model is output.

[0097] It should be noted that based on the entity relationship network in the knowledge graph, the meta random walk algorithm with meta path constraint is used for path sampling, taking the preparation difference entity as the starting point and the sensitive area entity as the end point. By defining the transition probability in the walking strategy, the path needs to be explored along the semantic relationship of "preparation difference -> electric field characteristics -> sensitive area" in multiple hops, and a meta path set conforming to the physical causal logic is generated. Each meta path carries the electric field simulation characteristics (such as the average field strength of the nodes in the path, the gradient covariance) it passes through, which are used as attribute extensions of the preparation difference nodes. A heterogeneous information network containing preparation difference nodes (with attributes), sensitive area nodes and feature relationship edges is constructed, where the weight of the feature relationship edge is determined by the statistical significance (path occurrence frequency) of the meta path. After mapping the heterogeneous network to a graph structure representation, the semantic similarity between meta paths is calculated. Using a path embedding-based similarity measurement method, each meta path is encoded into a low-dimensional vector, and the correlation strength between paths is quantified by cosine similarity. Strongly correlated path groups with similarity exceeding the threshold are screened. The neighborhood relationships of the preparation difference nodes and the sensitive area nodes in these path groups are merged to form a joint adjacency topology across meta paths. For each target node, the 1-hop neighbor node set connected by different meta paths is aggregated to construct a dynamic weighted adjacency matrix, and the matrix element value is determined by the weighted fusion result of the meta path similarity and the feature attribute similarity between nodes. The weight coefficient is optimized on the validation set through grid search to maximize the physical consistency of the neighborhood relationship.

[0098] Further, based on the constructed dynamic adjacency matrix and node features, a graph attention network of the heterogeneous graph is designed for model training. A multi-head attention mechanism is introduced in each network layer, and each attention head independently calculates the adaptive weight between nodes: the node features are mapped to query vectors and key vectors through a trainable parameter matrix, the attention score is calculated and normalized by ReLU activation function and Softmax, and the attention weight reflecting the correlation strength between nodes is generated. The output features of each attention head are spliced, and a gated fusion module is used to integrate multi-view features, suppress noise interference and retain key signals. The weighted features of neighbor nodes are aggregated through a message passing function, and the node representation is updated combined with its own features, and the node embedding vector with high-order semantic coding is obtained after iterative optimization. In the last layer of the network, a hierarchical mean pooling strategy is used to extract the implicit features of the global graph: first, the embedding vector of each sensitive region node is locally mean-pooled to aggregate the features of its associated difference nodes; then, the global node set is globally mean-pooled to generate a compressed feature vector input into the fully connected layer. The fully connected layer maps the feature vector to the risk level (such as low risk, medium risk, and high risk) and spatial position probability distribution of the sensitive region through nonlinear transformation. In the verification stage, a time sliding window is used to divide the training set and test set, and the intersection over union and KL divergence between the predicted sensitive region and the actual measurement result are calculated to evaluate the model performance, and finally a deployable sensitive region identification model is output.

[0099] Figure 2 An optimization scheme analysis flowchart for a transformer withstand voltage and partial discharge test platform is provided for an embodiment of the application;

[0100] As Figure 2 shown, the application provides an optimization scheme analysis flowchart for a transformer withstand voltage and partial discharge test platform, which includes:

[0101] S202, the structure features and preparation features of the test platform to be analyzed are obtained and input into the test platform digital model to perform electric field distribution simulation to obtain an electric field distribution simulation result, and the electric field distribution simulation result and the structure features and preparation features of the test platform to be analyzed are input into a sensitive region identification model to identify the local sensitive region of the test platform to be analyzed, and sensitive region identification information is obtained;

[0102] S204, a big data network is introduced, different historical test platform optimization instances are obtained through the big data network, and the electric field distribution features, structure features and preparation features of the test platform to be analyzed are calculated for similarity with each historical test platform optimization instance;

[0103] S206, select several historical test platform optimization instances in a preset similarity range through the calculated similarity value, extract the corresponding historical test platform optimization scheme to form an initial parameter space, and introduce a multi-objective grey wolf optimization algorithm to formulate a test platform optimization scheme to be analyzed;

[0104] S208, generate an initial grey wolf population by randomly sampling the initial parameter space, preset a target function and set a constraint condition, calculate the target function value of each individual in the initial grey wolf population through the target function, and divide all individuals in the population into different front levels through non-dominated sorting according to the calculated target function value;

[0105] S210, calculate the congestion degree distance for each front level, select the individual with the maximum congestion degree in each level as the leader wolf, and calculate the direction vector of the leader wolf in the decision space to update the position;

[0106] S212, output the optimal solution set after repeated iteration optimization until the stop condition is met, generate the optimal platform optimization scheme of the target test platform to be analyzed according to the optimal solution set, and push it.

[0107] It should be noted that first, based on the three-dimensional structure parameters (such as electrode geometric size, insulating medium layout) and preparation characteristics (such as material dielectric constant, assembly tolerance range) of the test platform to be analyzed, the electric field distribution is calculated through the finite element simulation model, and the simulation results including field strength peak value, gradient distribution and local discharge sensitive point coordinates are generated. The structural characteristics, preparation characteristics and electric field distribution data are input into the pre-trained sensitive area recognition model, the nonlinear correlation between multi-dimensional characteristics is analyzed by using the graph attention network, and high-precision sensitive area recognition information (including risk level, spatial coordinates and associated preparation difference factors) is output. Subsequently, a big data network is introduced, and different historical test platform optimization instances, electric field optimization schemes and their performance indicators (such as field strength reduction, cost increment, structure change) under different design parameter combinations are obtained through the big data network. Calculate the comprehensive similarity value of the test platform to be analyzed and the historical instances in three dimensions of structural characteristics (such as electrode curvature similarity), preparation characteristics (such as material parameter deviation) and electric field distribution (such as field strength contour difference). Select the historical instances with similarity values in the top 10%, extract the key parameters in the optimization scheme as the initial parameter space, and construct a search domain covering multi-objective optimization directions.

[0108] Further, in this parameter space, a multi-objective grey wolf optimization algorithm is applied to generate the optimization scheme. In the initialization stage, a Latin hypercube sampling is used to generate the grey wolf population, and each individual represents a set of optimization parameter combinations. The objective functions are set as the minimization of the field intensity peak value, the maximization of the controllability of material cost, and the minimization of the structural change amount, while the process feasibility constraints (such as the electrode displacement amount ≤ the upper limit of the assembly tolerance) are applied. The objective function values of each individual are calculated through finite element simulation, and the fast and non-dominated sorting algorithm is used to divide the population into different front levels, and the first front level is the Pareto optimal solution set. In order to maintain the diversity of the population, the crowding distance (based on the neighborhood density of the objective space) of each individual in the front is calculated, and the individual with the largest crowding distance in each layer is selected as the leader wolf, and the rest are the following wolves. In the iterative optimization stage, the leader wolf updates the position according to the gradient direction of the objective function: by calculating the direction vector of the leader wolf and the current individual in the parameter space, combined with the adaptive step size adjustment strategy (the step size decays exponentially with the iteration number), the following wolves are driven to move towards the Pareto front. After each position update, the out-of-bound parameters are reflected or truncated to ensure that the process constraints are met. When the preset stopping condition (such as the maximum iteration number 200 times or the optimal solution is not improved for 20 consecutive iterations) is reached, the Pareto optimal solution set is output, the final optimization scheme is generated and pushed, and the optimization process data is fed back to the historical instance library to form a closed-loop knowledge accumulation and scheme evolution mechanism.

[0109] The second aspect of the present application provides a computer readable storage medium, characterized in that the computer readable storage medium comprises a transformer withstand voltage and partial discharge test platform electric field analysis and optimization method program, and the transformer withstand voltage and partial discharge test platform electric field analysis and optimization method program is executed by a processor to realize the steps of the transformer withstand voltage and partial discharge test platform electric field analysis and optimization method according to any one of the above.

[0110] In several embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are only schematic. For example, the division of the units is only a logical function division. In actual implementation, there can be another division manner, such as: multiple units or components can be combined, or can be integrated into another system, or some features can be ignored or not executed. In addition, the coupling or direct coupling or communication connection between the displayed or discussed components can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

[0111] The units described as separate components above can or can not be physically separate, and the components shown as units can or can not be physical units; they can be located in one place or distributed on multiple network units; and part or all of the units can be selected according to actual needs to achieve the purposes of the embodiments.

[0112] In addition, each functional unit in each embodiment of the present application can be integrated into one processing unit, or each unit can be a separate unit, or two or more units can be integrated into one unit; the integrated unit can be implemented in the form of hardware or in the form of hardware plus software functional units.

[0113] Those skilled in the art can understand that all or part of the steps of the above-mentioned method embodiments can be completed by program instruction related hardware, and the aforementioned program can be stored in a computer readable storage medium, and the program executes the steps including the above-mentioned method embodiments when executed; and the aforementioned storage medium includes mobile storage devices, read-only memories (ROMs), random access memories (RAMs), magnetic discs or optical discs, and various storage medium that can store program codes.

[0114] Alternatively, the integrated units of the present application, if implemented in the form of software functional modules and sold or used as independent products, can also be stored in a computer readable storage medium. Based on this understanding, the technical solutions of the embodiments of the present application can be embodied in the form of software products, which are stored in a storage medium and include a number of instructions for causing a computer device (which can be a personal computer, a server, or a network device, etc.) to execute all or part of the methods described in the embodiments of the present application. The aforementioned storage medium includes mobile storage devices, ROMs, RAMs, magnetic discs or optical discs, and various storage medium that can store program codes.

[0115] The above is only a specific implementation of the present application, but the protection scope of the present application is not limited thereto, and any person skilled in the art can easily think of changes or replacements within the technical range disclosed by the present application, which should be covered within the protection scope of the present application. Therefore, the protection scope of the present application should be subject to the protection scope of the claims.

Claims

1. An electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform, characterized in that: include: Build a digital model of the test platform, obtain the preparation process standards and historical manufacturing measurement data of the target test platform, define the influencing variables that affect the electric field distribution, simulate the electric field distribution under different preparation differences, and establish an electric field simulation database; Based on the electric field simulation database, a spectral clustering algorithm is used to identify the electric field sensitive areas of the test platform under different preparation differences and construct a knowledge graph. A sensitive area recognition model is constructed based on a graph neural network, and the model is trained using the knowledge graph. The structural characteristics and preparation characteristics of the test platform to be analyzed are obtained, and the local sensitive areas of the test platform to be analyzed are identified using the sensitive area identification model. If local sensitive areas exist, an optimization plan is formulated to assist in optimizing the test platform.

2. The electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform according to claim 1, characterized in that: The construction of the digital model of the test platform specifically includes: Obtaining the design specifications of the target test platform, extracting the three-dimensional design parameters of the target test platform based on the obtained design specifications, using three-dimensional modeling technology to construct a geometric model of the target test platform based on the three-dimensional design parameters of the target test platform, and using a big data network to obtain and define the material properties of the target test platform; Extract the structural features of the target test platform based on its geometric model, conduct heat source analysis based on the design specifications, define the heat source type and heat source area, and obtain the heat source analysis information of the test platform; Based on the heat source analysis information of the test platform, a CFD model of the test platform is constructed using CFD software combined with the geometric model of the target test platform. Physical parameters and conductivity parameters are assigned to different areas in the CFD model through material properties. By coupling the Joule heat generation equation with the heat conduction equation, a two-way interaction relationship between the electric field and the temperature field is established. Meshing the constructed test platform CFD model, generating the initial computational mesh based on the triangulation method, and completing the initial meshing by defining the mesh quality by maximizing the minimum dihedral angle of the tetrahedral elements; After the initial grid is divided, a preliminary field strength calculation is performed on the initial grid based on the solution of the Poisson equation to obtain a preliminary field strength distribution, and the electric field gradient amplitude is extracted from the preliminary field strength distribution as a basis for grid optimization; If the electric field gradient amplitude is greater than a preset threshold, a local mesh refinement operation is triggered. Based on the h-adaptive method, an iterative convergence strategy is adopted to recursively subdivide the target unit through edge bisection and surface reconstruction technology, and a transition layer unit is inserted between adjacent units. After each grid encryption is completed, the electric field distribution is recalculated until it meets the preset stopping criteria and then the digital model of the test platform is output.

3. The electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform according to claim 1 is characterized in that: The method of obtaining the preparation process standards and historical manufacturing measurement data of the target test platform, defining the influencing variables affecting the electric field distribution, simulating the electric field distribution under different preparation differences, and establishing an electric field simulation database specifically includes: Obtain the preparation process standards and historical manufacturing measurement data of the target test platform, and define the influencing variables that affect the electric field distribution, including electrode assembly position deviation, insulation medium thickness tolerance, batch fluctuation of material dielectric constant, and contact interface resistance variation; Extracting historical manufacturing measurement features corresponding to each influencing variable from the historical manufacturing measurement data, comparing them with the manufacturing process standards, analyzing manufacturing differences in the actual manufacturing process, and generating a difference probability distribution of each influencing variable based on the manufacturing differences; The Latin supersampling method is used to combine historical manufacturing measurement data and the probability distribution of the differences of each influencing variable. Different influencing variables are randomly combined to construct several parameter combination sample sets with different categories and degrees of manufacturing differences. Obtain a digital model of the test platform, import the constructed parameter combination sample set as model input into the digital model of the test platform for simulation analysis, analyze the electric field distribution under different parameter combinations based on the finite element solver, and obtain a simulation result data set; The global electric field intensity distribution characteristics are extracted from the simulation result data set, including the field intensity peak position, the maximum gradient area, the field intensity value and the grid unit position of the local discharge sensitive point. The global electric field intensity distribution characteristics are associated with the corresponding input parameter combination to construct a mapping relationship between the preparation difference and the electric field response, and generate an electric field simulation database.

4. The electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform according to claim 1, characterized in that: The method uses a spectral clustering algorithm based on the electric field simulation database to identify the electric field sensitive areas of the test platform under different preparation differences and construct a knowledge graph, specifically including: Obtaining an electric field simulation database, extracting electric field simulation features of each preparation difference combination from the electric field simulation database, and mapping unit field strength into risk probability values ​​based on the extracted electric field simulation features as risk levels of partial discharge occurring at different locations; A feature space is constructed based on the acquired electric field simulation features and risk probability values. An adaptive Gaussian kernel function is used to calculate the similarity weights between any grid cells in the feature space and construct a similarity matrix. The bandwidth parameter of the Gaussian kernel function is determined by local density estimation. A weighted undirected graph is constructed using the similarity matrix, wherein the graph nodes represent grid cells and the edge weights represent the degree of feature similarity, and a normalized Laplace matrix is ​​calculated using the weighted undirected graph; Perform eigendecomposition on the calculated normalized Laplace matrix to obtain the eigenvectors corresponding to the first k smallest eigenvalues. The obtained eigenvectors are mapped to a low-dimensional embedding space, and a spectral clustering algorithm is introduced to identify sensitive areas. The elbow rule is used to determine the optimal number of clusters. The silhouette coefficient and the sum of squared distances within the cluster corresponding to different numbers of clusters are calculated to generate a cluster number change curve. The k value corresponding to the inflection point of the curve is selected as the optimal number of classifications. The low-dimensional feature vectors are divided by the K-means algorithm, and the low-dimensional feature vectors are clustered by iteratively optimizing the centroid position to minimize the sum of the Euclidean distances of samples within the class. After clustering is completed, several clusters are output; Calculate the mean field strength, partial discharge probability peak, and gradient distribution skewness of each cluster, compare them with the preset thresholds, mark sensitive areas based on the judgment results, and obtain sensitive area identification information; The sensitive area identification information is associated with each preparation difference combination in the electric field simulation database to construct a knowledge graph with preparation difference-electric field simulation feature-sensitive area as the meta-path.

5. The electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform according to claim 1, characterized in that: The sensitive area recognition model is constructed based on the graph neural network, and the model is trained through the knowledge graph, specifically including: Obtain a knowledge graph, use a meta random walk algorithm to perform a random walk in the knowledge graph with the prepared difference as the starting entity and the sensitive area as the ending entity, and sample several meta paths from the starting entity to the ending entity in the knowledge graph; The electric field simulation features corresponding to each element path are extracted and used as additional features of the preparation difference node to construct a heterogeneous information network including preparation difference nodes, sensitive area nodes and feature relationship edges; The heterogeneous information network is represented by a graph, and similarity values ​​between meta-paths are calculated based on the graph representation. A meta-path similarity threshold is preset and a strongly associated path group is screened, and the neighborhood relationships between the prepared difference nodes and the sensitive area nodes contained therein are merged; For each target node, the set of 1-hop neighbor nodes across its meta-paths is aggregated to construct a dynamic weighted adjacency matrix. The matrix element values ​​are determined by the weighted product of the meta-path similarity and feature similarity between nodes, and the weight coefficients are optimized through grid search. A sensitive area recognition model is constructed based on a graph neural network. The dynamic weighted adjacency matrix is ​​imported into the sensitive area recognition model for model training. A multi-head attention mechanism is introduced to calculate the adaptive weights between nodes. The output features of each attention head are spliced ​​through the calculated adaptive weights and integrated using a gated fusion model, and the node representation is updated using a message passing function; In the last layer of the network, all node representations are hierarchically mean pooled to extract the implicit feature vector of the global graph. The input is mapped into the fully connected layer as the electric field analysis result. The result is verified using verification data. When it meets the preset standards, the trained sensitive area recognition model is output.

6. The electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform according to claim 1, characterized in that: The structural characteristics and preparation characteristics of the test platform to be analyzed are obtained, and the local sensitive areas of the test platform to be analyzed are identified using the sensitive area identification model. If the local sensitive areas exist, an optimization plan is formulated to assist in optimizing the test platform, specifically including: The structural characteristics and preparation characteristics of the test platform to be analyzed are obtained, and the inputs are made into the digital model of the test platform to perform electric field distribution simulation to obtain electric field distribution simulation results. The electric field distribution simulation results and the structural characteristics and preparation characteristics of the test platform to be analyzed are input into the sensitive area identification model to identify the local sensitive area of ​​the test platform to be analyzed, and obtain sensitive area identification information; A big data network is introduced to obtain optimization examples of different historical test platforms. The electric field distribution characteristics, structural characteristics, and preparation characteristics of the test platform to be analyzed are similar to the optimization examples of each historical test platform. The similarity values ​​obtained by calculation are used to select several historical test platform optimization instances within the preset similarity range, extract the corresponding historical test platform optimization schemes to form the initial parameter space, and introduce the multi-objective grey wolf optimization algorithm to formulate the optimization scheme for the test platform to be analyzed; An initial gray wolf population is generated by random sampling in the initial parameter space, an objective function is preset and constraints are set, an objective function value of each individual in the initial gray wolf population is calculated using the objective function, and all individuals in the population are divided into different frontier levels by performing non-dominated sorting based on the calculated objective function values; Congestion calculation is performed on each frontier layer to obtain the crowding distance. The individual with the largest crowding degree in each layer is selected as the leader wolf, and the remaining individuals are used as follower wolves. The direction vector of the leader wolf in the decision space is calculated to update its position. The optimal solution set is output after repeated iterative optimization until the stopping condition is met. The optimal platform optimization solution of the target test platform to be analyzed is generated based on the optimal solution set and pushed.

7. A computer-readable storage medium, characterized in that The computer-readable storage medium includes an electric field analysis and optimization method program for a transformer withstand voltage and partial discharge test platform. When the electric field analysis and optimization method program applicable to a transformer withstand voltage and partial discharge test platform is executed by a processor, the steps of the electric field analysis and optimization method for a transformer withstand voltage and partial discharge test platform as described in any one of claims 1 to 6 are implemented.

Images