CN119150625A

CN119150625A - GIS electromagnetic field distribution rapid high-precision simulation method, simulator, storage medium and system

Info

Publication number: CN119150625A
Application number: CN202411626746.3A
Authority: CN
Inventors: 李奇越; 钱斌斌; 章海斌; 李帷韬; 郭振宇; 孙伟; 张学友; 许渊; 马欢; 冯秋洋; 刘鑫; 罗欢
Original assignee: Hefei University of Technology; Super High Voltage Branch of State Grid Anhui Electric Power Co Ltd
Current assignee: Hefei University of Technology; Super High Voltage Branch of State Grid Anhui Electric Power Co Ltd
Priority date: 2024-11-14
Filing date: 2024-11-14
Publication date: 2024-12-17
Anticipated expiration: 2044-11-14
Also published as: CN119150625B

Abstract

The invention discloses a GIS electromagnetic field distribution rapid high-precision simulation method, a simulator, a storage medium and a system, wherein the method comprises the steps of adopting a time domain finite difference algorithm or a finite element algorithm to carry out numerical solution on a three-dimensional simulation model of GIS equipment to obtain a GIS full-wave electromagnetic coarse grid simulation result, wherein the coarse grid simulation result comprises coarse grid structure data and coarse grid field intensity data; the invention can realize the GIS electromagnetic field distribution simulation with high speed and high precision.

Description

GIS electromagnetic field distribution rapid high-precision simulation method, simulator, storage medium and system

Technical Field

The invention relates to the technical field of power equipment, in particular to a GIS electromagnetic field distribution rapid high-precision simulation method, a simulator, a storage medium and a system.

Background

Gas insulated switchgear (Gas Insulated Switchgear, GIS) plays a vital role in modern power systems. Such devices utilize inert gases as insulating mediums to enable safe and reliable transmission and distribution of high voltage power in a relatively small space. The high performance and reliability of the GIS are important to ensure the stable operation of the power system, improve the transmission efficiency of the power grid and ensure the safety of power supply. Along with the rapid development of a power system and the improvement of the requirements on the safety performance of a power grid, the electromagnetic simulation of the GIS becomes one of the key points of research, and the electromagnetic simulation method can help engineers and researchers to understand the electromagnetic field distribution, electromagnetic interference and the influence of the electromagnetic field distribution and the electromagnetic interference on the GIS performance in depth.

The simulation research of the GIS generally utilizes a Finite element simulation method or a Finite time Domain difference method (FDTD) and the like, wherein the Finite element simulation method is used for simulating and optimizing a switching equipment product based on Finite element analysis, and the simulation and calculation of the three-dimensional electric field intensity of the novel vacuum circuit breaker are carried out by utilizing Finite element analysis software, which is proposed in literature, ouyang Zhenguo and a master paper. The finite time domain difference method is that a set of FDTD (finite time division method) method suitable for transient electric field calculation of GIS equipment is proposed in ' GIS equipment transient electric field simulation calculation based on a finite difference method of a time domain ' of a Shuoshi paper, yang Rui, university of science and technology ', and the literature is used for researching key problems such as grid subdivision algorithm and the like in GIS equipment by selecting a proper grid subdivision algorithm and absorption boundary conditions and dividing the GIS into two parts of coaxial bus pipelines and GIS complex equipment according to structural characteristics during modeling; because GIS electromagnetic field simulation focuses on the accurate simulation of equipment model and electromagnetic field distribution interaction, the complexity is high, the time cost and the simulation accuracy are required to be considered at the same time, but the literature mainly researches on improvement of the accuracy degree of a simulation algorithm, neglects the time cost required in the actual simulation process, focuses on research on a theoretical direction and lacks practical applicability, and mainly aims to analyze the electric field distribution simulation of GIS equipment under the action of transient voltage, and the simulation usually focuses on the response of the equipment under the action of voltage impact or current impact in short time, so that the generalization capability is weak.

In addition, the traditional numerical method has the problems that the coupling treatment between coarse and fine grids is complex, the boundary condition is difficult to establish and the like when the electromagnetic field non-uniform grid simulation is carried out, meanwhile, the non-uniform grid division simulation has high calculation resource requirement, long calculation time and low simulation speed when the high-grid division quantity and the density are simulated, the detection time can be seriously influenced when the high-frequency GIS partial discharge detection problem is processed, and the simulation precision is low although the time cost can be reduced when the coarse grid simulation is adopted, the partial discharge detection quality can be influenced, so that the quick high-precision grid electromagnetic field distribution simulation technology is urgently needed.

Disclosure of Invention

The invention aims to solve the technical problem of realizing quick and high-precision GIS electromagnetic field distribution simulation.

The invention solves the technical problems by the following technical means:

the invention provides a GIS electromagnetic field distribution rapid high-precision simulation method, which is characterized by comprising the following steps:

Carrying out numerical solution on a three-dimensional simulation model of GIS equipment by adopting a time domain finite difference algorithm or a finite element algorithm to obtain a GIS full-wave electromagnetic coarse grid simulation result, wherein the coarse grid simulation result comprises coarse grid structure data and coarse grid field intensity data;

And adopting an enhancement model to enhance the coarse grid simulation result to obtain enhanced fine grid data.

Further, the process of carrying out numerical solution on the three-dimensional simulation model of the GIS equipment by adopting the time domain finite difference algorithm comprises the following steps:

Dispersing a three-dimensional simulation model of GIS equipment into space grids, and determining the corresponding relation between the serial number of each grid node and space coordinate data;

And converting Maxwell's equations for describing the GIS electromagnetic field into a fourth-order matrix, solving the fourth-order matrix by adopting a four-step HIE-FDTD algorithm, and calculating field intensity data of each grid node.

Further, the converting maxwell's equations for describing the GIS electromagnetic field into a fourth-order matrix, and solving the fourth-order matrix by using a four-step HIE-FDTD algorithm, calculating the field intensity data of each grid node includes:

the maxwell's equations for describing GIS electromagnetic fields are converted into a sixth order matrix form as:

in the formula, Is a vector composed of components of an electric field and a magnetic field in a rectangular coordinate system,Is a six-order matrix;

the conversion of the sixth-order matrix form into the fourth-order rectangular form is as follows:

in the formula, Is that,

Is that;

And solving a fourth-order matrix form by adopting a four-step HIE-FDTD algorithm, and calculating field intensity data of each grid node.

Further, the sixth order matrixThe formula of (2) is:

in the formula, For the dielectric constant of the material to be a dielectric constant,Is magnetic permeability.

Further, the solving the fourth-order matrix form by adopting a four-step HIE-FDTD algorithm, and calculating the field intensity data of each grid node includes:

the four-step HIE-FDTD algorithm is decomposed into 、、、Calculating three pairs of angles implicit about grid node field intensity data by adopting a semi-implicit differential format for each step;

and solving the tri-diagonal implicit expression by adopting a catch-up method to obtain grid node field intensity data.

Further, the enhancement model adopts a differential deep learning network model, the differential deep learning network model comprises an enhancement network and a structural similarity network which are sequentially connected, and the enhancement network comprises a self-adjusting module and a differential convolution module;

calculating the coarse grid structure data and the coarse grid field intensity data by using the self-adjusting module and the differential convolution module respectively to obtain fine grid reinforced grid structure characteristics and fine grid reinforced field intensity characteristics;

Matching the fine grid enhanced grid structural characteristics with the fine grid enhanced field intensity characteristics by using the structural similarity network, and calculating similarity characteristics;

based on the similarity features, enhanced fine mesh data is calculated.

Further, the self-adjusting module comprises a first convolution layer and a second convolution layer which are sequentially connected, and the output characteristics of the first convolution layer and the output characteristics of the second convolution layer are output to an activation function layer after a first addition operation;

The coarse mesh structure data is used as the input of the first convolution layer, the mesh size supplementary information of the coarse mesh structure data is used as the input of the second convolution layer, the deviation vector of the coarse mesh structure data is used as the input of the first addition operation, and the coarse mesh structure data and the mesh size supplementary information of the coarse mesh structure data are output to the first addition operation through residual connection.

Further, the differential convolution module comprises a differential convolution layer, a size integration layer, a self-attention mechanism layer and a third convolution layer which are sequentially connected, wherein an activation function is connected behind the third convolution layer;

The coarse grid field intensity data is used as input of the differential convolution layer, and the differential convolution layer is used for calculating field intensity characteristics of different sizes of the coarse grid field intensity data by adopting a differential algorithm;

And the dimension integration layer is used for summing the field intensity characteristics of different dimensions calculated by the differential convolution layer to obtain reformed field intensity characteristics.

Further, the convolution kernel of the differential convolution layer adopts any one of five-point differential convolution kernel, weighted differential convolution kernel, multi-scale differential convolution kernel, directional differential convolution kernel, nine-point differential convolution kernel and mixed mode differential convolution.

Further, the structural similarity network comprises a first branch network, a second adding operation and a first multi-layer perceptron, wherein the outputs of the first branch network and the second branch network are connected with the second adding operation, the output of the second adding operation is connected to the first multi-layer perceptron, and the multi-layer perceptron is connected with an activating function.

Further, the first branch network comprises a convolutional neural network layer CNN and a batch regularization operation which are sequentially connected, and an activation function is connected to the regularization operation;

the second branch network comprises a second multi-layer perceptron, which is followed by an activation function.

Further, the matching the fine mesh enhanced grid structure feature and the fine mesh enhanced field strength feature by using the structural similarity network, and calculating the similarity feature includes:

Extracting structural feature information of the fine mesh enhanced grid structural feature by using the first branch network, wherein the formula is as follows:

in the formula, Is the firstThe fine mesh enhances the characteristic information of the mesh structural characteristics,Reinforcing grid structural features for the fine grid,For the convolution operation of the stack,For a batch regularization operation,Is an activation function;

Extracting field intensity characteristic information of the fine grid enhanced field intensity characteristic by using the second branch network, wherein the formula is as follows:

in the formula, Is the firstThe fine grid enhances the characteristic information of the field strength characteristics,The field strength characteristics are enhanced for the fine grid,An operation performed for a second multi-layer perceptron;

Calculating the combination characteristic of the field intensity and the structure by utilizing the characteristic information of the fine grid enhanced grid structure characteristic and the modulation weight corresponding to the characteristic information of the fine grid enhanced field intensity characteristic, wherein the formula is as follows:

in the formula, For the field strength and structure combination characteristics,AndThe characteristic information of the fine-grid enhanced grid structure characteristic and the modulation weight corresponding to the characteristic information of the fine-grid enhanced field intensity characteristic are respectively represented;

based on the field strength and structure combination characteristics, calculating the similarity characteristics, wherein the formula is as follows:

in the formula, In order to be able to use the similarity feature,An operation performed for the first multi-layer perceptron.

Further, the computing enhanced fine mesh data based on the similarity features includes:

Calculating normalized coarse grid structure data, normalized coarse grid field strength data, normalized fine grid enhancement grid structure characteristics and normalized fine grid enhancement field strength characteristics based on the coarse grid structure data, the coarse grid field strength data, the fine grid enhancement grid structure characteristics and the fine grid enhancement field strength characteristics, respectively;

Calculating grid structure consistency based on the normalized coarse grid structure data and the normalized fine grid enhanced grid structure features;

Calculating grid field intensity consistency based on the normalized coarse grid field intensity data and the normalized fine grid enhancement field intensity characteristics;

Calculating grid field strength loss compensation based on the grid structure consistency and the grid field strength consistency;

calculating fine-grid-enhanced structural data based on the fine-grid-enhanced grid structural features, the grid field strength loss compensation, and the similarity features;

And calculating fine grid enhancement field intensity data based on the fine grid enhancement field intensity characteristics, the grid field intensity loss compensation and the similarity characteristics.

Further, the enhancement model adopts a generated countermeasure network model, the generated countermeasure network model comprises a generator and a discriminator, the generator is used for extracting characteristics of the coarse grid simulation result, and enhancing the circulation times of the extracted characteristics to obtain enhanced fine grid data;

The discriminator comprises a structural similarity module, a field strength consistency module and an accuracy discriminating module, wherein:

the structure similarity module is used for calculating first confidence values of the structure information contained in the enhanced fine grid data and the real fine grid data obtained through simulation;

The field intensity consistency module is used for calculating second confidence coefficient values of field intensity information contained in the enhanced fine grid data and the real fine grid data obtained through simulation;

the accuracy judging module is used for drawing the distance between the enhanced fine grid data and the real fine grid data obtained through simulation.

Further, the structure similarity module comprises a first structure branch network, a second structure branch network and a first regression network, wherein the enhancement structure information contained in the enhancement fine grid data and the real fine structure information contained in the real fine grid data are respectively used as the input of the first structure branch network and the second structure branch network, and the output of the first structure branch network and the output of the second structure branch network are subjected to matrix multiplication operation to output a structure feature matrix;

And the structural feature matrix is used as an input of the first regression network to obtain a first confidence value between the enhanced structural information and the real fine structural information.

Further, the field strength consistency module comprises a first enhanced branch network, a second enhanced branch network and a second regression network, wherein enhanced field strength information contained in the enhanced fine grid data and real fine field strength information contained in the real fine grid data are respectively used as inputs of the first enhanced branch network and the second enhanced branch network, and outputs of the first enhanced branch network and the second enhanced branch network are subjected to matrix multiplication operation to output field strength characteristic matrixes;

And the field intensity characteristic matrix is used as input of the second regression network to obtain a second confidence value between the enhanced field intensity information and the real fine field intensity information.

Further, the generator comprises a plurality of cascaded feature extraction and expansion networks, and a multi-layer perceptron is connected behind each feature extraction and expansion network;

The feature extraction and expansion network comprises a feature extraction module and a feature expansion module which are sequentially connected, wherein the output of the feature extraction module in the upper-level feature extraction and expansion network is connected with the output of the feature extraction module in the lower-level feature extraction and expansion network;

the feature extraction module comprises a first MLPs layer, a KNN layer, a second MLPs layer, a maximum pooling layer maxpooling and a data compression layer which are sequentially connected, the coarse grid simulation result is used as the input of the first MLPs layer, the output of the first MLPs layer and the output of the second MLPs layer are spliced and then output to the maximum pooling layer maxpooling, the coarse grid simulation result and the output of the maximum pooling layer maxpooling are spliced and then used as the input of the data compression layer, and the output of the data compression layer is connected with the feature expansion module;

the feature expansion module is used for adding a vector which is coded into 1 or-1 to each feature output by the feature extraction module to generate position disturbance.

Further, when the features output by the feature extraction module in the previous-stage feature extraction and expansion network are copied to the features output by the feature extraction module in the next-stage feature extraction and expansion network, interpolation technology is adopted to match the features of different levels.

Further, when the finite element algorithm is adopted to carry out numerical solution on the three-dimensional simulation model of the GIS equipment, the method comprises the following steps:

and discretizing a simulation model of the GIS equipment into a plurality of tetrahedron units by adopting a finite element method, and simulating to obtain a simulation result of the GIS full-wave electromagnetic coarse grid, wherein each tetrahedron unit comprises a plurality of nodes.

Further, the enhancement processing is performed on the coarse mesh simulation result by adopting an enhancement model to obtain enhanced fine mesh data, and the method further comprises the following steps:

Abstracting simulation model of GIS equipment into network topology structure And any two adjacent free tetrahedrons in the network topology are treated as one sheet unit, wherein,A set of nodes is represented and,Representing an edge set, wherein connecting lines among grid vertexes are used as edges;

Taking the coarse grid simulation result as an initial reference feature, and adopting GRAPHSAGE algorithm to perform once aggregation treatment on neighbor nodes of each node in each slice unit to generate an initial reference feature of a newly added node;

and carrying out secondary aggregation processing on neighbor nodes of the newly added node by adopting GRAPHSAGE algorithm to generate the feature vector representation of the newly added node.

Further, the step of using the coarse grid simulation result as an initial reference feature, and performing a primary aggregation treatment on neighbor nodes of each node in each slice unit by adopting GRAPHSAGE algorithm to generate an initial reference feature of a newly added node, includes:

Taking the coarse grid simulation result of each node as an initial reference characteristic of each node;

traversing each node in each of the patch units All adjacent nodes of each node are taken as sampling points, and the node is obtainedNeighbor set of (c)And aggregate nodes using an aggregation functionInitial reference features corresponding to all nodes in the neighbor set to obtain one-time aggregation neighbor features;

Concatenating the initial reference features and the aggregated neighbor featuresObtaining a first joint feature, performing matrix multiplication operation on a first weight matrix and the first joint feature, and obtaining node features after one-time traversal through nonlinear transformation;

And aggregating the initial reference features of the common nodes of the two free tetrahedrons in each sheet unit by using an aggregation function to obtain the initial reference features of the newly added nodes.

Further, the adopting GRAPHSAGE algorithm to perform secondary aggregation processing on the neighbor nodes of the newly added node, generating the feature vector representation of the newly added node includes:

node characteristics after one-time traversal As a new reference feature of the corresponding node;

with each node in each of the patch units The neighbor sampling points serving as newly added nodes are used for aggregating new reference features of all nodes of each slice unit by using an aggregation function to obtain secondary aggregation neighbor features;

And connecting the initial reference feature of the newly added node with the secondary aggregation neighbor feature to obtain a second joint feature, performing matrix multiplication operation on the second weight matrix and the second joint feature, and obtaining the feature vector representation of the newly added node through nonlinear transformation.

Further, any two adjacent pairs of free tetrahedral coincident faces generate a new node.

In addition, the invention also provides a GIS electromagnetic field distribution rapid high-precision simulator which comprises a memory and a processor, wherein the processor runs a program corresponding to executable program codes stored in the memory by reading the executable program codes, so as to realize the GIS electromagnetic field distribution rapid high-precision simulation method.

In addition, the invention also provides a computer readable storage medium, on which a computer program is stored, which when being executed by a processor, realizes the GIS electromagnetic field distribution rapid high-precision simulation method.

In addition, the invention also provides a multi-structure adaptive GIS field-electric fusion real-time state sensing and early warning system, which comprises the signal synchronous acquisition and measurement embedded device and the industrial personal computer, wherein the industrial personal computer is provided with the simulator, the simulator is connected with the signal synchronous acquisition and measurement embedded device, the signal synchronous acquisition and measurement embedded device is used for acquiring time domain full waveform data of GIS partial discharge, the simulator is used for obtaining enhanced fine grid data based on the time domain full waveform data, and the industrial personal computer is used for searching the actual discharge position of the GIS partial discharge source based on the time domain full waveform data and the enhanced fine grid data.

The invention has the advantages that:

(1) According to the invention, a three-dimensional simulation model of the GIS equipment is subjected to numerical solution by utilizing a time domain finite difference algorithm or a finite element algorithm, a coarse grid simulation result can be obtained rapidly, and data enhancement is performed on a data basis by using the coarse grid simulation result, so that fine grid data with enhanced GIS electromagnetic field distribution, namely high-precision grid data, is obtained, and therefore, the high-precision GIS electromagnetic field distribution data can be obtained rapidly.

(2) The invention can obviously improve the precision and stability of the numerical solution by introducing a higher-order time integration scheme and a space integration scheme, the traditional FDTD method can be influenced by numerical dissipation and numerical dispersion, and the invention can reduce the problems by a more accurate integration method, and is particularly effective in processing long-time simulation or high-frequency electromagnetic wave propagation. For electromagnetic field simulation of complex structures, such as medium, non-uniform medium or non-uniform structure, the four-step hybrid explicit-implicit time domain finite difference (hybrid implicit explicit finite-DIFFERENCE TIME-domain, HIE-FDTD) method is adopted in the embodiment to better handle the complexity. The precise integration and numerical methods can effectively simulate electromagnetic field propagation and interaction between different media.

(3) Because the GIS electromagnetic field simulation is more focused on the accurate simulation of the detail interaction between equipment and the electromagnetic field, the differential deep learning network model is designed and constructed, and the convolution combined with a differential algorithm is adopted in the differential deep learning network model to process the coarse grid field intensity data, so that the electromagnetic field distribution of the GIS can be accurately simulated, particularly under the condition of processing complex geometric structures and boundary conditions, the simulation process is optimized by utilizing the deep learning network, the simulation precision is improved, the simulation process is accelerated, and the design and evaluation period is effectively shortened, thereby the precision and the efficiency of the GIS electromagnetic simulation are considered.

(4) The real coarse grid data and the real fine grid data obtained through simulation are utilized to train a generated countermeasure network, the real coarse grid data obtained through real-time simulation can be enhanced through a trained generator to generate GIS electromagnetic field fine data, and as GIS equipment has a certain physical structure, data points of electromagnetic field simulation cannot exceed a certain range.

(5) The method comprises the steps of carrying out enhancement processing on coarse grid data of GIS equipment obtained by adopting GRAPHSAGE algorithm to simulate the finite element method, obtaining new node data which is updated and iterated by neighbor nodes between every two adjacent free tetrahedrons in the coarse grid data after GRAPHSAGE algorithm, so that more new node data are generated on the basis of original grid node data, and enhancing and refining of grid node data generated by simulation calculation are completed.

(6) Under the condition of ensuring precision, the time required by simulation of the multi-structure GIS equipment tends to be increased along with the rising of the fineness of the grids, the data solution of the relatively coarse grids is obtained through the simulation method, and the obtained data is enhanced through the designed grid data enhancement method, so that the data result similar to the simulation of the relatively fine grids is achieved, the time cost is reduced under the condition of not losing the precision as much as possible, and compared with the traditional simulation method, the time and precision contradiction exists when the simulation of the multi-structure GIS equipment is similar to the relatively complex equipment model.

Additional aspects and advantages of the invention will be set forth in part in the description which follows and, in part, will be obvious from the description, or may be learned by practice of the invention.

Drawings

FIG. 1 is a schematic flow chart of a GIS electromagnetic field distribution rapid high-precision simulation method according to a first embodiment of the present invention;

FIG. 2 is a schematic flow chart of enhancement processing of coarse grid simulation results by using a differential deep learning network model in a second embodiment of the invention;

FIG. 3 is a schematic block diagram of a differential network-based coarse grid data enhancement of a GIS electromagnetic field in a second embodiment of the present invention;

FIG. 4 is a schematic diagram of an enhancement network in a differential deep learning network model according to a second embodiment of the present invention;

FIG. 5 is a schematic diagram of the design principle of the differential convolution module in the second embodiment of the present invention;

FIG. 6 is a schematic diagram of a structure similarity network in a differential deep learning network model according to a second embodiment of the present invention;

FIG. 7 is a schematic block diagram of a PU-GAN based GIS electromagnetic field simulation data enhancement in a third embodiment of the present invention;

FIG. 8 is a schematic diagram of a generator in a third embodiment of the invention;

FIG. 9 is a schematic diagram of a feature extraction module according to a third embodiment of the invention;

FIG. 10 is a schematic diagram of a similar module in a third embodiment of the invention;

FIG. 11 is a schematic diagram of a field strength uniformity module in accordance with a third embodiment of the present invention;

FIG. 12 is a schematic view of the structure of a self-focusing unit according to a third embodiment of the present invention;

FIG. 13 is a schematic block diagram of GIS simulation data enhancement based on GRAPHSAGE algorithm in a fourth embodiment of the present invention;

FIG. 14 is a flowchart of enhancing coarse grid data of a GIS electromagnetic field according to a fourth embodiment of the present invention;

FIG. 15 is a schematic diagram of a fourth embodiment of the present invention using GRAPHSAGE algorithm to implement two-layer aggregation;

fig. 16 is a schematic structural diagram of a GIS electromagnetic field distribution fast high-precision simulator according to a fifth embodiment of the present invention.

Detailed Description

For the purpose of making the objects, technical solutions and advantages of the embodiments of the present invention more apparent, the technical solutions in the embodiments of the present invention will be clearly and completely described in the following in conjunction with the embodiments of the present invention, and it is apparent that the described embodiments are some embodiments of the present invention, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the invention without making any inventive effort, are intended to be within the scope of the invention.

Example 1

As shown in fig. 1, a first embodiment of the present invention provides a method for fast and high-precision simulation of GIS electromagnetic field distribution, where the method includes:

S101, carrying out numerical solution on a three-dimensional simulation model of GIS equipment by adopting a time domain finite difference algorithm or a finite element algorithm to obtain a GIS full-wave electromagnetic coarse grid simulation result, wherein the coarse grid simulation result comprises coarse grid structure data and coarse grid field intensity data;

s102, enhancement processing is carried out on the coarse grid simulation result by adopting an enhancement model, and enhanced fine grid data are obtained.

According to the embodiment, firstly, a three-dimensional simulation model of the GIS equipment is subjected to numerical solution by using a time domain finite difference algorithm or a finite element algorithm, a coarse grid simulation result can be obtained quickly, data enhancement is carried out on a data base by using the coarse grid simulation result, and fine grid data with enhanced GIS electromagnetic field distribution, namely high-precision grid data, is obtained, so that the high-precision GIS electromagnetic field distribution data can be obtained quickly.

As a further preferable technical solution, in the step S101, a process of performing numerical solution on the three-dimensional simulation model of the GIS device by using a time domain finite difference algorithm includes the following steps:

s111, discretizing a three-dimensional simulation model of GIS equipment into space grids, and determining the corresponding relation between the serial number of each grid node and space coordinate data;

specifically, the three-dimensional simulation model of the GIS equipment is discretized into rectangular space grids, each grid node is numbered by using the Ye grid idea, and the number of each grid node and the corresponding relation of the space coordinate data of each grid node are determined.

S112, converting Maxwell' S equations for describing the GIS electromagnetic fields into fourth-order matrixes, solving the fourth-order matrixes by adopting a four-step HIE-FDTD algorithm, and calculating field intensity data of each grid node.

As a further preferable technical scheme, the step S112 is to convert Maxwell' S equations for describing the GIS electromagnetic field into a fourth-order matrix, solve the fourth-order matrix by adopting a four-step HIE-FDTD algorithm, and calculate the field intensity data of each grid node, and specifically comprises the following steps:

s1121, converting Maxwell' S equations for describing GIS electromagnetic fields into a six-order matrix form:

in the formula, Is a vector composed of components of an electric field and a magnetic field in a rectangular coordinate system,Wherein:

Wherein, Is a six-order matrix, and has the following expression form:

S1122, converting the sixth-order matrix form into a fourth-order rectangular form is as follows:

in the formula, AndThe expression forms of (a) are respectively as follows:

s1123, solving a fourth-order matrix form by adopting a four-step HIE-FDTD algorithm, and calculating field intensity data of each grid node.

The embodiment can remarkably improve the precision and stability of the numerical solution by introducing a higher-order time integration scheme and a space integration scheme. While conventional FDTD methods may suffer from numerical dissipation and numerical dispersion, the present embodiment may reduce these problems by a more accurate integration method, particularly when dealing with long-term simulations or high-frequency electromagnetic wave propagation. For electromagnetic field simulations of complex structures, such as those of dielectric, non-uniform dielectric or non-uniform structures, the present embodiment can better address these complexities. The precise integration and numerical methods can effectively simulate electromagnetic field propagation and interaction between different media.

Specifically, the step S1123 is to solve a fourth-order matrix form by using a four-step HIE-FDTD algorithm, and calculate field intensity data of each grid node, which specifically includes:

the four-step HIE-FDTD algorithm is decomposed into 、、、Four steps, wherein:

(1):

(2):

(3):

(4) Step by step:

For the following Step by step, adopting semi-implicit differential format to introduce auxiliary vectorThe stability condition of the algorithm is kept unchanged, and the method can be obtained:

Wherein the method comprises the steps of AndTo cancel the mutual coupling equation by substitutionIs a tri-diagonal implicit equation for (a), the following is shown:

in the formula, As a point loss parameter of the medium,,,For the time step size of the time step,Representing the electric field components of the electric field in the x, y, z directions,The magnetic field components of the magnetic field in the x, y, z directions, the superscript denoting the time step, e.gThe Ex of the time step n is indicated,The electric field Ex representing the 1/4 position between time steps n and n +1,Is a sixth order identity matrix.

For the followingWill assist the vectorSubstitution and expansion of the expression of (2) to obtain:

For the following Will assist the vectorSubstitution and expansion of the expression of (2) to obtain:

the components of the electric field and the magnetic field in the x, y and z directions in each step can be solved according to the above equation.

It should be noted that, in this embodiment, a simulation calculation result may be obtained through calculation, and the result shows that the spatial coordinates corresponding to each grid node and the distribution data of the electromagnetic field at the point are shown, because the thickness degree of the simulation grid division is proportional to the calculation precision and inversely proportional to the calculation time length, in order to solve the contradiction between precision and time, the grid data real-time enhancement simulation method needs to be designed to achieve grid enhancement on the coarse grid simulation electromagnetic field distribution result, so as to achieve the result of finer grid simulation.

As a further preferable technical solution, in step S101, a process of performing numerical solution on the three-dimensional simulation model of the GIS device by using a finite element algorithm specifically includes:

It should be noted that, according to the structural parameters of the GIS device, the embodiment may utilize simulation software such as COMSOL Multiphysics to build a simulation model of three-dimensional data, and discrete the three-dimensional data model into a plurality of tetrahedral units by using a finite element method, and perform simulation calculation to obtain coarse grid simulation data to be processed. Deriving the result of the simulation calculation to obtain each nodeIs (are) simulated feature dataAs a feature vector of the node, in whichRepresenting the three-dimensional spatial coordinates of the node,Representing the electric field strength of the node.

Example two

On the basis of the disclosure of the first embodiment, the present embodiment specifically adopts a differential deep learning network model to enhance the coarse grid simulation result, so as to obtain enhanced fine grid data, and the specific implementation flow is as shown in fig. 2 to 3:

S201, inputting the coarse grid simulation result into a differential deep learning network model, wherein the differential deep learning network model comprises an enhancement network and a structural similarity network which are sequentially connected, and the enhancement network comprises a self-adjustment module and a differential convolution module;

S202, calculating the coarse grid structure data and the coarse grid field intensity data by using the self-adjusting module and the differential convolution module respectively to obtain fine grid reinforced grid structure characteristics and fine grid reinforced field intensity characteristics;

S203, matching the fine grid enhanced grid structural characteristics with the fine grid enhanced field intensity characteristics by utilizing the structural similarity network, and calculating similarity characteristics;

S204, calculating enhanced fine grid data based on the similarity characteristics.

The differential deep learning network model is designed and built, the convolution combined with a differential algorithm is adopted in the differential deep learning network model to process coarse grid field intensity data, electromagnetic field distribution of a GIS can be accurately simulated, particularly under the condition of processing complex geometric structures and boundary conditions, the simulation process is optimized by utilizing the deep learning network, the simulation precision is improved, the simulation process is accelerated, the design and evaluation period is effectively shortened, and therefore the precision and the efficiency of GIS electromagnetic simulation are considered.

As a further preferable technical solution, as shown in fig. 4, the self-adjusting module includes a first convolutional layer Conv1 and a second convolutional layer Conv2 that are sequentially connected, where outputs of the first convolutional layer Conv1 and the second convolutional layer Conv2 are output to an activation function layer after being connected through a first addition operation;

the coarse mesh structure data is used as the input of the first convolution layer Conv1, the mesh size supplementary information of the coarse mesh structure data is used as the input of the second convolution layer Conv2, the deviation vector of the coarse mesh structure data is used as the input of the first addition operation, and the coarse mesh structure data and the mesh size supplementary information of the coarse mesh structure data are connected through residual errors Output to the first addition operation.

It should be noted that, in this embodiment, the fine enhancement of the coarse grid structure data is achieved by designing the self-adjusting module, and the fine enhancement of the coarse grid structure data of the electromagnetic field of the GIS device is achieved by combining the convolution operation, the residual connection and the sigmoid activation function, so that the method not only improves the detail expressive force of the grid data, but also can adaptively adjust the enhancement strategy according to the original grid data through the self-adjusting mechanism, and ensures the accuracy and the adaptability of the enhanced grid structure.

As a further preferable technical scheme, the function of the self-adjusting module is to finely enhance the grid structure, and the embodiment is thatCoarse grid data of electromagnetic field of strip GIS equipmentSelf-adjusting module in input enhancement module for obtaining fine-grid enhanced grid structure characteristicsThe formula is:

in the formula, The grid structure features are enhanced for fine grids,A trellis-enhanced convolution operation is represented,A fine-tuning convolution operation is shown,For the coarse mesh structure data,Supplementary information for the mesh size of the coarse mesh structure data,For the bias vector of the coarse mesh structure data,For the residual link,To activate the function.

As a further preferable technical solution, as shown in fig. 4, the differential convolution module includes differential convolution layers connected in sequenceSize integration layerSelf-attention mechanism layerAnd a third convolution layer Conv3, wherein the third convolution layer Conv3 is connected withActivating a function;

the coarse grid field intensity data is used as the differential convolution layer Is input to the differential convolution layerThe method comprises the steps of calculating field intensity characteristics of different sizes of the coarse grid field intensity data by adopting a differential algorithm;

The size integration layer The method comprises the steps of summing field intensity characteristics of different sizes obtained through calculation of the differential convolution layer to obtain reformed field intensity characteristics;

the field intensity characteristics after reforming pass through a self-focusing mechanism layer And outputting the fine-grid enhanced field intensity characteristic after the third convolution layer Conv 3.

It should be noted that, in this embodiment, the differential convolution module is designed to accurately extract field intensity data, and the differential convolution module can accurately extract electromagnetic field intensity characteristics through specially designed differential convolution operation, so as to effectively simulate physical distribution of electromagnetic field, and in combination with a self-attention mechanism, the module can highlight important field intensity characteristics, so that the extraction accuracy of field intensity data and the attention concentration of a model can be improved. And the multi-scale characteristics of electromagnetic field data can be captured by carrying out characteristic extraction and summation through convolution kernels with different sizes, which is particularly important for understanding and simulating complex changes of electromagnetic fields.

As a further preferable technical scheme, the convolution kernel of the differential convolution layer adopts any one of five-point differential convolution kernel, weighted differential convolution kernel, multi-scale differential convolution kernel, directional differential convolution kernel, nine-point differential convolution kernel and mixed mode differential convolution.

It should be noted that, a person skilled in the art may select other differential algorithms to combine with the deep learning network according to practical application requirements, and the embodiment is not limited specifically.

Further, taking five-point difference as an example, as shown in FIG. 5, the difference convolution layerThe design is as follows:

basic convolution kernel is designed according to five-point differential algorithm :

By dimension transformation kernel considering three-dimensional GIS electromagnetic field oriented simulation data enhancementObtaining a three-dimensional differential convolution kernel:

Wherein, the element corresponding multiplication is indicated by the letter.

Using three-dimensional differential convolution kernelsExtracting the field intensity gradient information of the coarse grid (extracting the gradient change of the electromagnetic field of the coarse grid and guiding the generation of fine grid data):

wherein ∗ denotes the convolution operation, For applications ofGradient information data obtained later;

enhancement of coarse grid field strength data with upsampling to obtain upsampled fine grid field strength data U (to obtain refined coarse grid data):

Wherein, The up-sampling convolution kernel is obtained along with network training;

preliminary fine grid field intensity data combining coarse grid field intensity gradient information G reinforcement

。

The present embodiment directly utilizes the form of a differential algorithm to define the convolution kernel, ensuring that the physical principles can be mapped accurately into the convolution operation. This is achieved by taking the core calculation formula of the difference algorithm and converting it into the form of a convolution kernel, thus ensuring the correct application of the continuous physical law on the discrete data. And the present embodiment allows for a dimensional transformation in designing the differential convolution kernel, allowing the electromagnetic field to be analyzed at different resolutions. By adjusting the size and shape of the convolution kernel, different physical resolutions can be simulated, capturing multi-scale features.

The two-dimensional convolution kernel is extended to a three-dimensional space through certain conversion, namely, a traditional differential method is applied to a key step of three-dimensional GIS electromagnetic field simulation, and the physical effect of each direction in the three-dimensional space is considered by the dimension conversion. The problem of rotational invariance can be solved by designing a symmetrical three-dimensional convolution kernel, one symmetrical kernel can generate consistent response in all directions, and the isotropy problem in electromagnetic field simulation can be solved.

It should be noted that, the special differential convolution operation of the present embodiment is a method based on solving a partial differential equation of a numerical value, such as a five-point differential method, and this method is commonly used to approximate the solution of the partial differential equation in the numerical analysis. In deep learning, conventional convolution layers are typically used to extract features of the input data, and special differential convolution kernels may be designed to better accommodate simulation of specific physical phenomena, such as electromagnetic field simulation. The design of such convolution kernels is typically based on discrete approximations of the second derivative of the function space in a differential algorithm, thereby enabling the network to learn the data while also capturing the physical characteristics of the data.

Compared with the traditional convolution layer, the differential convolution can ensure that the data accords with a specific physical rule while maintaining high-quality data, so that the data generated by the network is more physically credible. And the combination mode ensures that the model is better in performance when dealing with the problems needing accurate physical modeling, such as electromagnetic field simulation, climate change simulation and the like.

Further, the differential convolution module provided in this embodiment has the function of acquiring accurate fine-grid field intensity data, and uses the differential convolution module to pair the firstCoarse grid field intensity data of strip GIS equipment electromagnetic fieldProcessing to obtain fine-grid enhanced field intensity characteristicsThe formula is:

in the formula, The field strength characteristics generated for the differential convolution,In the case of a differential convolution operation,For the coarse grid field strength data,As a characteristic of the field strength after reforming,Indicating that convolution kernels of different sizes are applied,The field strength characteristics are enhanced for the fine grid,In order for the self-attention mechanism to operate,In order for the convolution operation to be performed,To activate the function.

Furthermore, in the embodiment, through the combination of the self-adjusting module and the differential convolution module, the calculation efficiency and the precision of simulation are optimized, the self-adjusting module reduces unnecessary calculation through an intelligent enhancement strategy, the differential convolution module ensures high-precision extraction of field intensity data, and the self-adjusting module and the differential convolution module act together to realize efficient and accurate electromagnetic field simulation. In general, the design and implementation of the self-adjusting module and the differential convolution module greatly improve the accuracy, efficiency and applicability of GIS electromagnetic field simulation through intelligent enhancement, accurate feature extraction and efficient calculation methods.

As a further preferable solution, as shown in fig. 6, the structural similarity network includes a first branch network, a second adding operation, and a first multi-layer perceptron, where outputs of the first branch network and the second branch network are both connected to the second adding operation, and an output of the second adding operation is connected to the first multi-layer perceptronThe multi-layer perceptronFollowed by an activation function。

Specifically, the first branch network comprises a convolutional neural network layer CNN and a batch regularization operation which are connected in sequenceThe regularization operationFollowed by an activation function;

The second branch network comprises a second multi-layer perceptronThe second multi-layer perceptronFollowed by an activation function。

Further, the function of the structural similarity network is to match the enhanced grid data with the field strength data for the firstFine grid enhanced grid structure characteristics of coarse grid number of electromagnetic field of strip GIS equipmentEnhanced field strength features with fine gridsCorresponding features of the field strength and structure:

in the formula, Is the firstExtracting information of the structural characteristics of the fine grid reinforced grids corresponding to the number of the coarse grids of the electromagnetic field of the GIS equipment; Is the first The fine grid enhanced field intensity characteristic extraction information corresponding to the number of the coarse grids of the electromagnetic field of the strip GIS equipment; And Respectively representing fine-grid enhanced grid structure feature extraction informationInformation extracted from fine-grid enhanced field strength characteristicsIs used for modulating the weight of the (a); The method is characterized in that the information comprehensive characteristics are extracted from the fine-grid reinforced grid structure characteristics and the fine-grid reinforced field intensity characteristics; Is a similarity feature; Is a batch regularization operation, which is performed, Is a convolution operation and is performed by,Is a multi-layer perceptron of a structural similarity module,Is the combination characteristic of field intensity and structure.

It should be noted that, the function of the structural similarity network designed in this embodiment is that (1) the physical structure is preserved, that is, the structural similarity module keeps the structural integrity and continuity while ensuring that the physical accuracy of the enhanced data is preserved by focusing on the structural attribute of the data. This is particularly important in the field of simulation, since the physical behaviour of an electromagnetic field is strongly dependent on its spatial structural characteristics. (2) The module is favorable for generating data with higher quality, and particularly in the recovery and super-resolution tasks, the module can ensure that the detail of the generated data has higher similarity with the original coarse grid data in structure. (3) The enhanced feature characterization is that the structural similarity module provides an effective mechanism for enhancing and remarkably characterizing key features in electromagnetic field data, particularly in field intensity and grid structure aspects by fusing the features extracted by CNN and MLP.

As a further preferable technical scheme, the step S204 is to calculate GIS fine mesh enhanced data based on the similarity feature, and specifically includes the following steps:

s241, calculating normalized coarse grid structure data, normalized coarse grid field intensity data, normalized fine grid enhancement grid structure characteristics and normalized fine grid enhancement field intensity characteristics based on the coarse grid structure data, the coarse grid field intensity data, the fine grid enhancement grid structure characteristics and the fine grid enhancement field intensity characteristics respectively;

It should be noted that, in this embodiment, the setting field intensity consistency module is specifically used to process the coarse grid structure data, the coarse grid field intensity data, the fine grid enhancement grid structure feature, the fine grid enhancement field intensity feature and the similarity feature, so as to converge the difference between the fine grid enhancement data and the coarse grid data.

The following calculation was used to obtain the firstElectromagnetic field fine grid enhanced grid structure characteristic of strip GIS equipmentAnd fine-grid enhanced field strength featuresAnd the firstCoarse grid data of electromagnetic field of strip GIS equipmentAnd coarse grid field strength dataFine-grid enhanced data after integrated compensation of (a):

In the formula,、、AndThe method comprises the steps of respectively obtaining structured characteristics of a structured electromagnetic field fine grid enhanced grid, structured fine grid enhanced field intensity characteristics, structured electromagnetic field coarse grid data and structured coarse grid field intensity data; Is matrix 2 norm.

S242, calculating the consistency of the grid structure based on the regularized coarse grid structure data and the regularized fine grid enhanced grid structure characteristics;

specifically, the calculation formula of the grid structure consistency is:

in the formula, Is the consistency of the grid structure.

S243, calculating grid field intensity consistency based on the normalized coarse grid field intensity data and the normalized fine grid enhancement field intensity characteristics;

specifically, the calculation formula of the grid field intensity consistency is as follows:

in the formula, For grid field strength uniformity.

S244, calculating grid field intensity loss compensation based on the grid structure consistency and the grid field intensity consistency;

Specifically, the calculation formula of grid field intensity loss compensation is:

in the formula, Compensating for grid field strength loss.

S245, calculating fine grid reinforced structure data based on the fine grid reinforced grid structure characteristics, the grid field intensity loss compensation and the similarity characteristics;

specifically, the calculation formula of the fine mesh enhanced structure data is:

in the formula, Structural data is enhanced for the fine mesh.

S246, calculating fine grid enhancement field intensity data based on the fine grid enhancement field intensity characteristics, the grid field intensity loss compensation and the similarity characteristics;

specifically, the calculation formula of the fine mesh enhanced field intensity data is as follows:

in the formula, Enhancing field strength data for the fine grid; And (3) with Is a scale factor.

As a further preferable technical solution, the differential deep learning network model used in the present embodiment is a model that is trained in advance to be used for calculating the similarity between the grid structure and the grid field intensity, and before the step S201 is to input the coarse grid simulation result into the differential deep learning network model, the method further includes the following steps:

The method comprises the steps of adopting a GIS electromagnetic field coarse grid historical simulation data set, training the differential deep learning network model by utilizing the historical simulation data set until the total loss function value of the model is minimum, wherein the total loss function comprises a constraint loss function based on Maxwell equations and a field strength and structure consistency loss function, and the formula is as follows:

in the formula, As a function of the total loss,For the field strength and structural consistency loss function,For maxwell constraint loss function,And (3) withIs a weight parameter.

It should be noted that, the embodiment performs multi-dimensional loss function design, and comprehensively considers a plurality of important characteristics of simulation data by combining structural similarity loss, field strength consistency loss and physical constraint loss, so that the design of the multi-dimensional loss function is beneficial to ensuring simulation accuracy and simultaneously can regulate and control weights among different losses so as to adapt to specific simulation requirements and optimization targets.

Further, the field strength and structure consistency loss function is formulated as:

in the formula, Representing the gradient obtained by applying the difference method,In order to achieve a loss of structural similarity,In order to achieve a loss of consistency of the field strength,And (3) withAs a scale factor, the number of the elements is,The grid structure features are enhanced for fine grids,In the form of coarse-mesh structure data,The field strength characteristics are enhanced for a fine grid,For the coarse grid field strength data,Is the vector 2 norm.

It should be noted that, through the structural similarity loss and the field intensity consistency loss, the embodiment ensures that the enhanced electromagnetic field data is kept highly consistent with the original data in structure and field intensity, and the consistency not only promotes the convergence of the simulation data, but also ensures the physical authenticity of the simulation result and improves the reliability of the simulation.

And by introducing a scaling factor, a flexible loss function adjustment mechanism is provided. The mechanism not only can adjust the importance of different parts in the loss function according to specific tasks, but also can improve the generalization capability and adaptability of the model when processing different types of electromagnetic field data.

Further, the formula of the constraint loss function based on maxwell's equations is expressed as:

in the formula, For the magnetic field component,For the loss of the gaussian law,For the Gao Sici law loss,For faraday's law of electromagnetic induction losses,For the ampere law loss,The proportionality constant is used to determine the ratio of the components,、、、As a scale factor, the number of the elements is,The rotation is indicated by the rotation of the rotor,Is the vector 2 norm; is the magnetic permeability of the vacuum and, Is the permittivity of the vacuum and,Is fine-meshed enhanced field strength data.

It should be noted that, the physical constraint loss function ensures that the enhanced electromagnetic field data strictly follows the basic law of electromagnetism through the key components of maxwell's equation set, namely the Gauss law, gao Sici law, faraday electromagnetic induction law and ampere law, and the constraint ensures that the electromagnetic field data output by the model is accurate in numerical value and strict in physical law, thereby increasing the scientificity and practicability of the model.

When the differential method is combined with the generation countermeasure network to form the differential enhancement generation countermeasure network, the differential convolution kernel is directly integrated into the generator architecture, so that the differential enhancement of the spatial feature is introduced. The challenge here is mainly reflected in the loss of the model, i.e. when using a generated challenge network, the total loss function of the model should include not only the physical constraint loss function and the field strength and structure consistency loss function, but also the challenge loss part.

In a word, the loss function designed in the embodiment combines the advantages of the traditional physical model and deep learning, and is beneficial to optimizing the stability and convergence speed of the algorithm and improving the training efficiency through accurate mathematical expression and physical constraint. The total loss function not only enhances the processing capacity of the model to electromagnetic field simulation data, but also ensures the accuracy and physical reality of the simulation result, thereby providing an effective optimization strategy for electromagnetic field simulation by combining the electromagnetic field theory and the deep learning technology.

Example III

On the basis of the disclosure of the first embodiment, the present embodiment specifically adopts generating an countermeasure network model to enhance the coarse grid simulation result, so as to obtain enhanced fine grid data, and the specific implementation schematic block diagram 7 shows:

The enhancement model adopts a generated countermeasure network model, the generated countermeasure network model comprises a generator and a discriminator, the generator is used for extracting characteristics of the coarse grid simulation result and enhancing the circulation times of the extracted characteristics to obtain enhanced fine grid data;

It should be noted that, in this embodiment, the real coarse mesh data and the real fine mesh data are input into the generation countermeasure network in advance to train until the overall optimization loss function is minimum, and a trained generator is obtained as a data enhancement model for enhancing real coarse mesh data generated in real time, where the generator network is responsible for learning potential distribution characteristics of the real sample and synthesizing a new sample, and the role of the discriminator network is to discriminate the real sample from the generated new sample. The generator and the discriminator achieve the countermeasure effect through alternate training, the respective generating capacity and discrimination capacity are continuously improved, and finally the generator network and the discriminator network countermeasure achieves Nash equilibrium.

The function of the design discriminator in the embodiment is to discriminate whether the input data meets the requirement, the discriminator consists of a structural similarity module, a field intensity consistency module and an accuracy discriminating module, and the grid data is divided into structural data and field intensity data for comparison analysis respectively, so that the discrimination accuracy can be effectively improved.

Because the GIS equipment has a certain physical structure, the data points of electromagnetic field simulation cannot exceed a certain range, the coarse grid data point multiple enhancement is carried out through up-sampling in the PU-GAN (Projection Upsampling GENERATIVE ADVERSARIAL Network) Network designed by the embodiment of the invention, and the generator adopts the cyclic multiple enhancement to carry out electromagnetic field enhancement among the coarse grid data points, so that the physical structure category of the GIS is not exceeded, and the accuracy of GIS electromagnetic field simulation data enhancement is improved.

As a further preferable technical scheme, as shown in fig. 8, the generator includes a plurality of cascaded feature extraction and expansion networks, and a multi-layer perceptron is connected behind each feature extraction and expansion network;

The feature extraction and expansion network comprises a feature extraction module Featuer Extraction and a feature expansion module which are sequentially connected, wherein the output of the feature extraction module Featuer Extraction in the upper-stage feature extraction and expansion network is connected with the output of the feature extraction module Featuer Extraction in the lower-stage feature extraction and expansion network;

As shown in fig. 9, the feature extraction module Featuer Extraction includes a first MLPs layer, a KNN layer, a second MLPs layer, a maximum pooling layer maxpooling, and a data compression layer compression that are sequentially connected, where the coarse mesh simulation result is used as an input of the first MLPs layer, an output of the first MLPs layer and an output of the second MLPs layer are spliced and then output to the maximum pooling layer maxpooling, and the coarse mesh simulation result is spliced with an output of the maximum pooling layer maxpooling and then used as an input of the data compression layer compression, and an output of the data compression layer compression is connected with the feature expansion module;

It should be noted that the function of the generator designed by the embodiment is that the GIS enhances the coarse grid data to generate the needed GIS fine grid data, the designed generator is designed by adopting the cyclic multiple enhancement, the coarse grid enhancement multiple can be adjusted automatically according to the requirement, and the multiple enhancement can better keep the characteristics of the coarse grid data.

Specifically, the working process of the feature extraction module in this embodiment is that three-dimensional space coordinates of each sampling point are input firstAnd field strength valueThe N multiplied by 4 dimensional information is extracted into a feature C' with a fixed scale through a first MLPs layer, then the features are grouped into N multiplied by K groups by utilizing a KNN layer, and the group features are further optimized by utilizing a second MLPs layer to obtain the features of the G channel. The second MLPs layer output features are combined with the first MLPs layer output features to obtain new G channel features (N x K x (2g+c')). The packet information is finally concentrated by maxpooling to yield the final nxx (2g+c') with the nx4-dimensional characteristics of the originally output sample points added. The feature is taken as a multi-scale feature aggregation of the modules, and is fed into a subsequent module to be multiplexed, so that the reconstruction precision is improved, the parameter efficiency is improved, the model size is reduced, and the NxC feature is obtained under the compression action of the compression layer of the last layer of data. After copying the resulting nxc features, a vector encoded as 1 or-1 is added to each feature to generate a position disturbance, and then the 2 nx (c+1) features are regressed to 2 nx 4 data using the following multi-layer perceptron MLP.

It should be noted that, a feature of 2 ^r n×c can be obtained through multiple cycle feature extraction and expansion, and finally a multi-layer perceptron MLP is returned to four-dimensional data。

It should be noted that the number of the substrates,The value of (c) depends on the accuracy difference between the input simulation coarse mesh and fine mesh, and the present embodiment is not particularly limited.

Specifically, for data accuracy for different modules, good interpolation techniques are required to match features at different levels if the last feature needs to be passed on to the next layer. In the embodiment, when the characteristics output by the characteristic extraction module in the characteristic extraction and expansion network of the previous stage are copied to the characteristics output by the characteristic extraction module in the characteristic extraction and expansion network of the next stage, the interpolation technology is adopted to match the characteristics of different levels.

As a further preferable technical solution, this embodiment uses a bilateral interpolation technique, and uses features around the target point to interpolate the pointThe interpolation of the characteristic value output by the previous-stage characteristic extraction module is as follows:

Wherein, Is the firstThe coordinates and field strength of the individual points,Is the firstThe coordinates and field strength of the individual points,Is the firstThe characteristic value corresponding to the point is set,Is the firstThe characteristic value corresponding to the point is set,Is the set of data points that need to be interpolated, where the joint weighting function is expressed as follows:

Wherein the width parameter And (3) withIs the average distance of the point to the nearest neighbor,Representing the modulus of the vector.

The embodiment enhances coarse grid data, and can effectively avoid overlapping generated grid data and source grid data by using bilateral interpolation, thereby avoiding waste.

As a further preferable technical solution, as shown in fig. 10, the structure similarity module includes a first structure branch network, a second structure branch network, and a first regression network, enhanced structure information included in the fine mesh generated data and real fine structure information included in the real fine mesh data are respectively used as inputs of the first structure branch network and the second structure branch network, and outputs of the first structure branch network and the second structure branch network output a structure feature matrix after matrix multiplication operation;

Specifically, as shown in fig. 10, the first structural branch network and the second structural branch network each include a multi-layer perceptron and a maximum pooling layer that are sequentially connected, and output structural feature matrices of the first structural branch network and the second structural branch network after matrix multiplication operation are:

in the formula, In order to be a structural feature matrix,Generating structural features of the data for the fine grid,As a structural feature of the real fine mesh data,、Respectively are provided with、Is used for the modulation weights of the (c),Indicating the operation of the multi-layer perceptron,Representing the maximum pooling operation and,Generating data for fine meshThe structural information of the individual points is used,First to true fine mesh dataStructural information of the individual points.

Specifically, as shown in fig. 10, the first regression network includes a first self-attention unit, a multi-layer perceptron and a fully-connected layer connected in sequence, where the first self-attention unit is configured to generate a structural attention weight based on the structural feature matrix, and the structural attention weight outputs a first confidence value after passing through the multi-layer perceptron and the fully-connected layer.

As a further preferable technical solution, as shown in fig. 11, the field strength consistency module includes a first enhanced branch network, a second enhanced branch network, and a second regression network, where enhanced field strength information included in the fine grid generated data and real fine field strength information included in the real fine grid data are respectively used as inputs of the first enhanced branch network and the second enhanced branch network, and outputs of the first enhanced branch network and the second enhanced branch network output field strength feature matrices after matrix multiplication operation;

Specifically, the first enhanced branch network and the second enhanced branch network both comprise a multi-layer perceptron and an activation function which are sequentially connected, and the output field intensity feature matrix of the first enhanced branch network and the output field intensity feature matrix of the second enhanced branch network after matrix multiplication operation is that:

in the formula, For the field intensity characteristic matrix,Generating field strength characteristics of the data for the fine grid,As a field strength characteristic of the real fine mesh data,'、Respectively are provided with、Is used for the modulation weights of the (c),Generating data for fine meshInformation of the field strength of the individual points,First to true fine mesh dataField strength information for each point.

Specifically, the second regression network comprises a second self-attention unit, a multi-layer perceptron and a full-connection layer which are sequentially connected, wherein the second self-attention unit is used for generating field intensity attention weights based on the field intensity feature matrix, and the field intensity attention weights output a second confidence value after passing through the multi-layer perceptron and the full-connection layer.

As a further preferred solution, as shown in fig. 12, the self-attention unit adopted in this embodiment is used for enhancement feature integration, specifically, the input features are converted into G, H and F by three independent MLPs, and then the attention weights are generated by G and H:

M

Wherein, Representing the softmax function.

And then obtaining a weighted feature M, and finally generating an output feature, wherein the output feature is the sum of the input feature and the weighted feature.

As a further preferred technical solution, in this embodiment, the outputs of the structural similarity module and the field strength consistency module are both connected to an accuracy discrimination module, where the accuracy discrimination module is configured to control generation of grid structure and field strength data and distinguish the accuracy of the generated result, and the accuracy discrimination module may be defined as a region-level full convolution two-classifier, so as to achieve a distance between the fine grid simulation data generated by pulling up and the fine grid simulation data distribution by a numerical method, and at the same time, ensure that the training process is stable by using least square contrast loss.

The contrast loss between fine mesh generation data and real fine mesh data is defined as:

in the formula, And (3) withFine grid generated data output from the generator by the discriminant D respectivelyAnd real fine grid dataConfidence value of the prediction.

As a further preferable technical solution, the formula of the overall optimization loss function is expressed as:

in the formula, The loss function is optimized for the population,Generating a first confidence value for the fine mesh data and the structural information contained in the real fine mesh data,Generating a second confidence value for the fine mesh data and field strength information contained by the real fine mesh data,Generating data for said fine mesh and countermeasures against losses between said real fine mesh data,AndIs a weight parameter.

The overall optimized loss function set in this embodiment comprehensively considers the structure and field intensity confidence of generating the fine grid and the real fine grid, and has the advantages of 1) promoting the generator to generate realistic data, namely, the generator is forced to generate samples which are as close to real data distribution as possible through optimizing the loss function. 2) The accuracy of the arbiter is improved by optimizing the loss function to improve its ability to distinguish between real data and generated data. This includes enabling the arbiter to more accurately identify the data generated by the generator, as well as to better distinguish between the generated data and the real data. 3) The balance between the generator and the arbiter is realized, the optimization loss function ensures that a balance state exists between the generator and the arbiter, and the unstable training caused by one of the excessive optimizations is avoided. This balancing helps the generator generate more realistic data while ensuring that the arbiter can effectively distinguish between real data and generated data.

Example IV

On the basis of the disclosure of the first embodiment, the present embodiment performs enhancement processing on a coarse grid simulation result obtained by numerically solving a three-dimensional simulation model of a GIS device by using a finite element method, as shown in fig. 13 to 14, and includes the following steps:

s301, abstracting a simulation model of GIS equipment into a network topology structure And any two adjacent free tetrahedrons in the network topology are treated as one sheet unit, wherein,A set of nodes is represented and,Representing an edge set, wherein connecting lines among grid vertexes are used as edges;

Specifically, in order to adopt a neighbor set of a fixed scale, this embodiment regards any adjacent two free tetrahedrons M and N as one slice unit, the free tetrahedrons M and N respectively include a node A, B, C, D and a node B, C, D, E. Wherein three nodes of two adjacent free tetrahedrons are duplicated, node B, C, D.

S302, taking the coarse grid simulation result as an initial reference feature, and adopting GRAPHSAGE algorithm to perform once aggregation processing on neighbor nodes of each node in each slice unit to generate an initial reference feature of a newly added node;

it should be noted that, when the first aggregation is performed by adopting the GRAPHSAGE algorithm in this embodiment, information of adjacent nodes of all nodes in each slice unit is aggregated, so that a new node is generated by any two adjacent faces where tetrahedrons overlap.

And S303, performing secondary aggregation processing on neighbor nodes of the newly added node by adopting GRAPHSAGE algorithm, and generating the feature vector representation of the newly added node.

It should be noted that, for the newly added node generated by the overlapping surface of any two adjacent free tetrahedrons, each node in the sheet unit formed by the two free tetrahedrons can be used as the neighbor node of the newly added node, the neighbor node of the newly added node is subjected to secondary aggregation processing by adopting GRAPHSAGE algorithm to generate the feature vector representation of the newly added node, and the three-dimensional space coordinate data and the electric field intensity of the newly added node are obtained, so that more new node data are generated on the basis of the original grid node data, and the enhancement refinement of the grid node data generated by the simulation calculation is completed.

In detail, in the GRAPHSAGE algorithm, a random sampling of neighboring nodes is typically employed to update the representation of the target node. In GIS simulation, however, the selection of neighboring nodes needs to take into account more factors, such as physical distances between nodes, similarity of nodes, and the like. Therefore, in this embodiment, two free tetrahedral units are regarded as one slice unit, so that neighbors of newly added nodes are gathered in one slice unit, so that similarity and relevance between nodes are higher, meanwhile, in GIS simulation data, various types of node characteristics are considered, and the spatial coordinates and electric field intensity of the nodes are used as node characteristics to better conform to GIS simulation data. In addition, the embodiment adjusts the depth of the model of GRAPHSAGE algorithm, namely the aggregation layer number, aiming at the complexity and the scale of GIS simulation data, and improves the expression capability of the algorithm on the data so as to obtain better performance.

As a further preferable technical solution, as shown in fig. 15, the step S302 includes using simulation feature data of nodes as initial reference features, performing a aggregation process on neighboring nodes of each node in each slice unit by using GRAPHSAGE algorithm, and generating reference features of newly added nodes, including the following steps:

s321, taking simulation characteristic data of each node as initial reference characteristics of each node;

S322, traversing each node in each slice unit All adjacent nodes of each node are taken as sampling points, and the node is obtainedNeighbor set of (c)And aggregate nodes using an aggregation functionInitial reference features corresponding to all nodes in the neighbor set to obtain one-time aggregation neighbor features;

Specifically, each node in the network topology() Is characterized in that the feature vector, namely the simulation feature data, is taken as the reference feature corresponding to each nodeFor each tile, traversing each node in the tile in a first aggregation() For a nodeFirst, acquire its neighbor setNodeAs its initial reference feature。

Specifically, nodes are aggregated by an aggregator using an aggregation functionInitial reference features corresponding to all nodes in the neighbor set to obtain one-time aggregation neighbor featuresThe method comprises the following steps:

in the formula, Representing an aggregation function.

S323, connecting the initial reference feature and the aggregated neighbor featureObtaining a first joint feature, performing matrix multiplication operation on a first weight matrix and the first joint feature, and obtaining node features after one-time traversal through nonlinear transformation;

Specifically, the present embodiment traverses each node onceAs a new feature vector representation of the node and as a reference feature for each node during a second traversal, wherein:

in the formula, Is a function of the splice-in,Is a first matrix of weights that are to be used,Is thatA function.

Further, the aggregation function in this embodiment uses an average aggregation function MEANI.e.。

S324, the initial reference features of the public nodes of the two free tetrahedrons in each sheet unit are aggregated by using an aggregation function, and the initial reference features of the newly added nodes are obtained.

Specifically, since the newly added node generated by the common plane of two adjacent free tetrahedrons is free of initial reference features, the embodiment obtains the initial reference features of the newly added node P by aggregating the initial reference features of the common node of two adjacent free tetrahedrons using an aggregation functionThe formula is:

in the formula, An initial reference feature that is a common node of two adjacent free tetrahedrons,

Further, the aggregation function in this embodiment adopts an average aggregation function, and there are。

As a further preferable technical scheme, the step S303 is to perform secondary aggregation processing on neighbor nodes of the newly added node by adopting GRAPHSAGE algorithm to generate a feature vector representation of the newly added node, and comprises the following steps:

s331, node characteristics after one-time traversal As a new reference feature of the corresponding node;

S332, each node in each slice unit The neighbor sampling points serving as newly added nodes are used for aggregating new reference features of all nodes of each slice unit by using an aggregation function to obtain secondary aggregation neighbor features;

Specifically, the present embodiment will be described with respect to all nodes in a slice unitNeighbor set of newly added node generated as the sheet unitAggregating new reference features of all nodes within a neighbor set using an aggregation function by a second aggregator, namely() Obtaining the aggregated neighbor featuresThe formula is:

in the formula, As a function of the aggregation function,Is a new reference feature for each node in the tile.

S333, connecting the initial reference feature of the newly added node with the secondary aggregation neighbor feature to obtain a second joint feature, performing matrix multiplication operation on the second weight matrix and the second joint feature, and obtaining the feature vector representation of the newly added node through nonlinear transformation.

Specifically, the feature vector of the newly added node obtained in this embodiment is expressed as:

in the formula, Is a function of the splice-in,Is a second matrix of weights that are to be used,Is thatA function.

Further, since the neighbors of the nodes have no natural order, the aggregator in this embodimentUsing an average aggregation function, there isTherefore, a new node characteristic vector, namely a new node three-dimensional space coordinate and electric field intensity, is generated between any slice units, so that the enhancement of GIS finite element method simulation grid node data is realized.

As a further preferred technical solution, the algorithm GRAPHSAGE adopted in this embodiment has the following complexity:

Wherein, The number of aggregators, the number of weight matrices and the number of layers of the network. The embodiment usesPersonal aggregatorFor aggregating node neighbor information, andIndividual weight matrixFor propagating information between different layers. To make the effect of sampling aggregation better, takeTwo polymerizations were performed.

As a further preferred solution, the sampling of the nodes in this embodiment is fixed-length,Represent the firstThe number of sampling neighbors of a layer is defined in advance by the number of sampling neighborsThe repeated sampling or negative sampling method with the substitution ensures that the complexity becomes stable, and when the number of the defined sampling neighbors is larger than the number of the actual neighbors, the repeated sampling is adopted to make the number of the actual sampling neighbors complementWhen the number of the defined sampling neighbors is smaller than the number of the actual neighbors, adopting negative sampling to reduce the number of the actual sampling neighbors to。

Example five

As shown in fig. 16, a fifth embodiment of the present invention proposes a GIS electromagnetic field distribution fast high-precision simulator, which is connected to a processor, a memory and a network interface through a system bus, wherein the memory may include a nonvolatile storage medium and an internal memory. The non-volatile storage medium may store an operating system and a computer program. The computer program includes program instructions that, when executed, cause the processor to perform the method for fast and high-precision simulation of GIS electromagnetic field distribution in any of the above-described embodiments one, two, three, and four.

Wherein the processor is configured to provide computing and control capabilities to support the operation of the entire computer device.

The internal memory provides an environment for the execution of a computer program in a non-volatile storage medium that, when executed by a processor, causes the processor to perform any one of a number of methods for determining a quantum radar correction geometry factor.

The network interface is used for network communication such as transmitting assigned tasks and the like. It will be appreciated by those skilled in the art that the structure shown in FIG. 16 is merely a block diagram of some of the structures associated with the present inventive arrangements and is not limiting of the computer device to which the present inventive arrangements may be applied, and that a particular computer device may include more or fewer components than shown, or may combine some of the components, or have a different arrangement of components.

It should be appreciated that the Processor may be a central processing unit (Central Processing Unit, CPU), it may also be other general purpose processors, digital signal processors (DIGITAL SIGNAL Processor, DSP), application SPECIFIC INTEGRATED Circuit (ASIC), field-Programmable gate array (Field-Programmable GATE ARRAY, FPGA) or other Programmable logic device, discrete gate or transistor logic device, discrete hardware components, or the like. Wherein the general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.

Example six

The sixth embodiment of the invention provides a multi-structure adaptive GIS field-electric fusion real-time state sensing and early warning system, which comprises a signal synchronous acquisition and measurement embedded device and an industrial personal computer, wherein the industrial personal computer is provided with the simulator in the fifth embodiment, the simulator is connected with the signal synchronous acquisition and measurement embedded device, the signal synchronous acquisition and measurement embedded device is used for acquiring time domain full waveform data of GIS partial discharge, the simulator is used for obtaining enhanced fine grid data based on the time domain full waveform data, and the industrial personal computer is used for searching an actual discharge position of a GIS partial discharge source based on the time domain full waveform data and the enhanced fine grid data.

It should be noted that, other embodiments of the emulator or the implementation method of the present invention may refer to the above method embodiments, and will not be described herein.

Example seven

The sixth embodiment of the present invention further proposes a computer readable storage medium, on which a computer program is stored, which when executed by a processor, implements the rapid high-precision simulation method for GIS electromagnetic field distribution according to any of the embodiments above.

It should be noted that the logic and/or steps represented in the flowcharts or otherwise described herein, for example, may be considered as a ordered listing of executable instructions for implementing logical functions, and may be embodied in any computer-readable medium for use by or in connection with an instruction execution system, apparatus, or device, such as a computer-based system, processor-containing system, or other system that can fetch the instructions from the instruction execution system, apparatus, or device and execute the instructions. For the purposes of this description, a "computer-readable medium" can be any means that can contain, store, communicate, propagate, or transport the program for use by or in connection with the instruction execution system, apparatus, or device. More specific examples (a non-exhaustive list) of the computer-readable medium would include an electrical connection (an electronic device) having one or more wires, a portable computer diskette (a magnetic device), a Random Access Memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), an optical fiber device, and a portable compact disc read-only memory (CDROM). In addition, the computer readable medium may even be paper or other suitable medium on which the program is printed, as the program may be electronically captured, via, for instance, optical scanning of the paper or other medium, then compiled, interpreted or otherwise processed in a suitable manner, if necessary, and then stored in a computer memory.

It is to be understood that portions of the present invention may be implemented in hardware, software, firmware, or a combination thereof. In the above-described embodiments, the various steps or methods may be implemented in software or firmware stored in a memory and executed by a suitable instruction execution system. For example, if implemented in hardware, as in another embodiment, may be implemented using any one or combination of techniques known in the art, discrete logic circuits with logic gates for implementing logic functions on data signals, application specific integrated circuits with appropriate combinational logic gates, programmable Gate Arrays (PGAs), field Programmable Gate Arrays (FPGAs), and the like.

In the description of the present specification, a description referring to terms "one embodiment," "some embodiments," "examples," "specific examples," or "some examples," etc., means that a particular feature, structure, material, or characteristic described in connection with the embodiment or example is included in at least one embodiment or example of the present invention. In this specification, schematic representations of the above terms do not necessarily refer to the same embodiments or examples. Furthermore, the particular features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Furthermore, the terms "first," "second," and the like, are used for descriptive purposes only and are not to be construed as indicating or implying a relative importance or implicitly indicating the number of technical features indicated. Thus, a feature defining "a first" or "a second" may explicitly or implicitly include at least one such feature. In the description of the present invention, the meaning of "plurality" means at least two, for example, two, three, etc., unless specifically defined otherwise.

While embodiments of the present invention have been shown and described above, it will be understood that the above embodiments are illustrative and not to be construed as limiting the invention, and that variations, modifications, alternatives and variations may be made to the above embodiments by one of ordinary skill in the art within the scope of the invention.

Claims

1. A fast and high-precision simulation method for GIS electromagnetic field distribution, characterized in that the method comprises:

A finite-difference time-domain algorithm or a finite element algorithm is used to numerically solve a three-dimensional simulation model of the GIS equipment to obtain a GIS full-wave electromagnetic coarse-grid simulation result, wherein the coarse-grid simulation result includes coarse-grid structure data and coarse-grid field strength data;

The coarse grid simulation result is enhanced by using an enhanced model to obtain enhanced fine grid data.

2. The method for rapid and high-precision simulation of electromagnetic field distribution of GIS according to claim 1, characterized in that the process of numerically solving the three-dimensional simulation model of GIS equipment using finite-difference time-domain algorithm comprises:

Discretize the three-dimensional simulation model of GIS equipment into spatial grids, and determine the corresponding relationship between the number and spatial coordinate data of each grid node;

The Maxwell equations used to describe the electromagnetic field of GIS are converted into a fourth-order matrix, and the four-step HIE-FDTD algorithm is used to solve the fourth-order matrix and calculate the field strength data of each grid node.

3. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 2 is characterized in that the Maxwell equations used to describe the GIS electromagnetic field are converted into a fourth-order matrix, and the fourth-order matrix is solved using a four-step HIE-FDTD algorithm to calculate the field strength data of each grid node, including:

The Maxwell equations used to describe the electromagnetic field of GIS are converted into a sixth-order matrix form:

In the formula, is the vector consisting of the components of the electric and magnetic fields in the rectangular coordinate system, is a sixth-order matrix;

Convert the sixth-order matrix form to a fourth-order rectangular form:

In the formula, and is a sixth-order matrix;

The four-step HIE-FDTD algorithm is used to solve the fourth-order matrix form and calculate the field intensity data of each grid node.

4. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 3 is characterized in that the four-step HIE-FDTD algorithm is used to solve the fourth-order matrix form and calculate the field strength data of each grid node, including:

The four-step HIE-FDTD algorithm is decomposed into , , , Four sub-steps, for each sub-step a semi-implicit difference scheme is used to calculate the three diagonal implicits of the field strength data about the grid nodes;

The pursuit method is used to solve the three diagonal implicit equations and obtain the grid node field strength data.

5. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 1, characterized in that the enhanced model adopts a differential deep learning network model, the differential deep learning network model includes an enhanced network and a structural similarity network connected in sequence, and the enhanced network includes a self-adjusting module and a differential convolution module;

The self-adjusting module and the differential convolution module are used to calculate the coarse grid structure data and the coarse grid field strength data respectively to obtain fine grid enhanced grid structure characteristics and fine grid enhanced field strength characteristics;

Using the structural similarity network, the fine grid enhanced grid structure feature and the fine grid enhanced field strength feature are matched to calculate a similarity feature;

Based on the similarity features, enhanced fine grid data is calculated.

6. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 5, characterized in that the self-adjusting module comprises a first convolutional layer and a second convolutional layer connected in sequence, and the output features of the first convolutional layer and the output features of the second convolutional layer are output to the activation function layer after a first addition operation;

The coarse grid structure data is used as the input of the first convolutional layer, the grid size supplementary information of the coarse grid structure data is used as the input of the second convolutional layer, the bias vector of the coarse grid structure data is used as the input of the first addition operation, and the coarse grid structure data and the grid size supplementary information of the coarse grid structure data are output to the first addition operation via residual connection.

7. The fast and high-precision simulation method for GIS electromagnetic field distribution according to claim 5, characterized in that the differential convolution module comprises a differential convolution layer, a size integration layer, a self-attention mechanism layer and a third convolution layer connected in sequence, and an activation function is connected after the third convolution layer;

The coarse grid field intensity data is used as the input of the differential convolution layer, and the differential convolution layer is used to calculate the field intensity features of different sizes of the coarse grid field intensity data by using a differential algorithm;

The size integration layer is used to sum the field intensity features of different sizes calculated by the differential convolution layer to obtain the reorganized field intensity features.

8. The fast and high-precision simulation method for GIS electromagnetic field distribution as described in claim 7 is characterized in that the convolution kernel of the differential convolution layer adopts any one of a five-point differential convolution kernel, a weighted differential convolution kernel, a multi-scale differential convolution kernel, a directional differential convolution kernel, a nine-point differential convolution kernel, and a mixed mode differential convolution.

9. The fast and high-precision simulation method for GIS electromagnetic field distribution as described in claim 5 is characterized in that the structural similarity network includes a first branch network, a second branch network, a second addition operation and a first multi-layer perceptron, the outputs of the first branch network and the second branch network are both connected to the second addition operation, the output of the second addition operation is connected to the first multi-layer perceptron, and the multi-layer perceptron is followed by an activation function.

10. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 9, characterized in that the first branch network comprises a convolutional neural network layer CNN and a batch normalization operation connected in sequence, and the regularization operation is followed by an activation function;

The second branch network includes a second multilayer perceptron, and the second multilayer perceptron is followed by an activation function.

11. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 10, characterized in that the use of the structural similarity network to match the fine grid enhanced grid structure characteristics and the fine grid enhanced field strength characteristics to calculate similarity characteristics comprises:

The structural feature information of the fine grid enhanced grid structure feature is extracted by using the first branch network, and the formula is expressed as:

In the formula, For the The fine grid enhances the characteristic information of the grid structure characteristics, Enhance the grid structure characteristics for the fine grid, is a stacked convolution operation, is the batch normalization operation, is the activation function;

The second branch network is used to extract the field intensity feature information of the fine grid enhanced field intensity feature, and the formula is expressed as:

In the formula, For the The characteristic information of the fine grid enhanced field strength characteristics, Enhance the field strength characteristics for the fine grid, Operations performed by the second multilayer perceptron;

The field strength and structure combination feature is calculated by using the modulation weights corresponding to the feature information of the fine grid enhanced grid structure feature and the feature information of the fine grid enhanced field strength feature. The formula is expressed as:

In the formula, is the field strength and structure combination feature, and The modulation weights corresponding to the feature information of the fine grid enhanced grid structure feature and the feature information of the fine grid enhanced field strength feature are respectively represented;

Based on the field strength and structure combination characteristics, the similarity characteristics are calculated, and the formula is expressed as:

In the formula, For the similarity feature, Operations performed by the first multilayer perceptron.

12. The GIS electromagnetic field distribution fast and high-precision simulation method according to claim 5, characterized in that the calculation of enhanced fine grid data based on the similarity feature comprises:

Calculating regularized coarse grid structure data, regularized coarse grid field intensity data, regularized fine grid enhanced grid structure characteristics, and regularized fine grid enhanced field intensity characteristics based on the coarse grid structure data, the coarse grid field intensity data, the fine grid enhanced grid structure characteristics, and the fine grid enhanced field intensity characteristics, respectively;

Calculating grid structure consistency based on the regularized coarse grid structure data and the regularized fine grid enhanced grid structure features;

Calculating grid field strength consistency based on the regularized coarse grid field strength data and the regularized fine grid enhanced field strength characteristics;

Calculating fine grid enhancement structure data based on the fine grid enhancement grid structure characteristics, the grid field strength loss compensation and the similarity characteristics;

Based on the fine grid enhanced field strength characteristics, the grid field strength loss compensation and the similarity characteristics, fine grid enhanced field strength data is calculated.

13. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 1, characterized in that the enhancement model adopts a generative adversarial network model, the generative adversarial network model includes a generator and a discriminator, the generator is used to extract features from the coarse grid simulation results, and perform cyclic multiple enhancement on the extracted features to obtain enhanced fine grid data;

The discriminator includes a structural similarity module, a field strength consistency module and an accuracy discrimination module, wherein:

The structural similarity module is used to calculate a first confidence value of structural information contained in the enhanced fine grid data and the real fine grid data obtained by simulation;

The field strength consistency module is used to calculate a second confidence value of the field strength information contained in the enhanced fine grid data and the real fine grid data obtained by simulation;

The accuracy determination module is used to shorten the distance between the enhanced fine grid data and the real fine grid data distribution obtained by simulation.

14. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 13, characterized in that the structural similarity module comprises a first structural branch network, a second structural branch network and a first regression network, the enhanced structural information contained in the enhanced fine grid data and the real fine structural information contained in the real fine grid data are respectively used as inputs of the first structural branch network and the second structural branch network, and the outputs of the first structural branch network and the second structural branch network are output as structural feature matrices after matrix multiplication operations;

The structural feature matrix is used as an input of the first regression network to obtain a first confidence value between the enhanced structural information and the real fine structural information.

15. The method for rapid and high-precision simulation of GIS electromagnetic field distribution according to claim 13, characterized in that the field strength consistency module comprises a first enhanced branch network, a second enhanced branch network and a second regression network, the enhanced field strength information contained in the enhanced fine grid data and the real fine field strength information contained in the real fine grid data are respectively used as inputs of the first enhanced branch network and the second enhanced branch network, and the outputs of the first enhanced branch network and the second enhanced branch network are output as field strength characteristic matrices after matrix multiplication operations;

The field strength characteristic matrix is used as the input of the second regression network to obtain a second confidence value between the enhanced field strength information and the real fine field strength information.

16. The GIS electromagnetic field distribution fast and high-precision simulation method according to claim 13, characterized in that the generator comprises a plurality of cascaded feature extraction and expansion networks, each of which is connected to a multi-layer perceptron;

The feature extraction and expansion network includes a feature extraction module and a feature expansion module connected in sequence, and the output of the feature extraction module in the feature extraction and expansion network of the previous level is connected to the output of the feature extraction module in the feature extraction and expansion network of the next level;

The feature extraction module includes a first MLPs layer, a KNN layer, a second MLPs layer, a maximum pooling layer maxpooling and a data compression layer connected in sequence, the coarse grid simulation result is used as the input of the first MLPs layer, the output of the first MLPs layer and the output of the second MLPs layer are spliced and output to the maximum pooling layer maxpooling, the coarse grid simulation result and the output of the maximum pooling layer maxpooling are spliced and used as the input of the data compression layer, and the output of the data compression layer is connected to the feature expansion module;

The feature expansion module is used to add a vector coded as 1 or -1 to each feature output by the feature extraction module to generate position disturbance.

17. The fast and high-precision simulation method for GIS electromagnetic field distribution as described in claim 16 is characterized in that when the features output by the feature extraction module in the feature extraction and expansion network of the previous level are copied to the output of the feature extraction module in the feature extraction and expansion network of the next level, interpolation technology is used to match features at different levels.

18. The method for rapid and high-precision simulation of electromagnetic field distribution of GIS according to claim 1, characterized in that when the finite element algorithm is used to numerically solve the three-dimensional simulation model of the GIS equipment, it includes:

The simulation model of the GIS equipment is discretized into a plurality of tetrahedral units by using the finite element method, and the GIS full-wave electromagnetic coarse grid simulation result is obtained by simulation, wherein each of the tetrahedral units includes a plurality of nodes.

19. The method for rapid and high-precision simulation of electromagnetic field distribution of GIS according to claim 18, characterized in that the method further comprises:

Abstract the simulation model of GIS equipment into a network topology , and any two adjacent free tetrahedrons in the network topology are regarded as a sheet unit, where Represents a collection of nodes. Represents an edge set, where the lines between mesh vertices are considered edges;

Using the coarse grid simulation results as the initial benchmark features, the GraphSAGE algorithm is used to perform an aggregation process on the neighboring nodes of each node in each slice unit to generate the initial benchmark features of the newly added nodes;

The GraphSAGE algorithm is used to perform secondary aggregation on the neighbor nodes of the newly added nodes to generate the feature vector representation of the new nodes.

20. The GIS electromagnetic field distribution fast and high-precision simulation method according to claim 19, characterized in that the coarse grid simulation result is used as the initial reference feature, and the neighbor nodes of each node in each slice unit are aggregated by the GraphSAGE algorithm to generate the initial reference feature of the newly added node, including:

The coarse grid simulation results of each node are used as the initial benchmark features of each node;

Traverse each node in each of the slice units , taking all the adjacent nodes of each node as sampling points, obtain the node The neighbor set , and use the aggregation function to aggregate nodes Corresponding to the initial baseline features of all nodes in the neighbor set, an aggregated neighbor feature is obtained ;

Connect the initial reference features and aggregated neighbor features Get the first joint feature, perform matrix multiplication on the first weight matrix and the first joint feature, and then use a nonlinear transformation to get the node feature after one traversal ;

The initial datum features of the common nodes of the two free tetrahedrons in each sheet element are aggregated using an aggregation function to obtain the initial datum features of the newly added nodes.

21. The GIS electromagnetic field distribution fast and high-precision simulation method according to claim 19, characterized in that the use of the GraphSAGE algorithm to perform secondary aggregation processing on the neighbor nodes of the newly added node to generate a feature vector representation of the newly added node comprises:

The node features after a traversal As a new benchmark feature for the corresponding node;

Each node in each of the slice units As the neighbor sampling point of the newly added node, and using the aggregation function to aggregate the new benchmark features of all nodes in each slice unit, obtain the secondary aggregated neighbor feature ;

The initial reference feature of the newly added node and the secondary aggregated neighbor feature are connected to obtain a second joint feature, and a second weight matrix is multiplied by the second joint feature to obtain a feature vector representation of the newly added node through a nonlinear transformation.

22. The method for rapid and high-precision simulation of electromagnetic field distribution of GIS as claimed in claim 21, characterized in that a new node is generated corresponding to the overlapping faces of any two adjacent free tetrahedrons.

23. A fast and high-precision simulator for electromagnetic field distribution of GIS, characterized in that the simulator comprises a memory and a processor; wherein the processor runs a program corresponding to the executable program code by reading the executable program code stored in the memory, so as to implement the method as described in any one of claims 1 to 22.

24. A computer-readable storage medium having a computer program stored thereon, wherein when the computer program is executed by a processor, the method according to any one of claims 1 to 22 is implemented.

25. A multi-structure adaptive GIS field-electric fusion real-time status perception and early warning system, characterized in that the system includes a signal synchronization acquisition and measurement embedded device and an industrial computer, the industrial computer is provided with a simulator as claimed in claim 23, the simulator is connected to the signal synchronization acquisition and measurement embedded device, the signal synchronization acquisition and measurement embedded device is used to collect time domain full waveform data of GIS local discharge, the simulator is used to obtain enhanced fine grid data based on the time domain full waveform data, and the industrial computer is used to search for the actual discharge position of the GIS local discharge source based on the time domain full waveform data and the enhanced fine grid data.