GIS field-electricity fusion real-time status perception and early warning system and method

Abstract

The present invention discloses a GIS field-electricity fusion real-time state perception and early warning system and method, comprising a signal synchronous acquisition and measurement embedded device and an industrial computer, wherein a simulator, a position searcher and an early warning device are deployed in the industrial computer; when the embedded device monitors the electromagnetic field signal inside the GIS according to the strong correlation feature and contains a local discharge signal, a full time domain waveform diagram containing a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal is obtained; the simulator roughly locates the local discharge source based on the full time domain waveform diagram of the GIS local discharge, performs simulation enhancement calculation with the rough positioning result as the initial injection point, and obtains simulation enhancement data of the local discharge source; the position searcher iteratively searches the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement result to obtain the actual discharge position of the local discharge source; the early warning device issues an early warning based on the full time domain waveform and the early warning information containing the actual discharge position of the local discharge source.

Description

Technical Field

[0001] The present invention relates to the technical field of electric power equipment, and in particular to a GIS field-electricity fusion real-time state perception and early warning system and method. Background Art

[0002] The high performance and reliability of high-voltage gas insulated switchgear (GIS) are crucial to ensuring the stable operation of the power system, improving the transmission efficiency of the power grid, and ensuring the security of power supply.

[0003] With the widespread application of GIS equipment, insulation aging, loose components, and other equipment failures caused by them have caused frequent equipment failures, which have seriously threatened the safe operation of substations at all levels. Among them, partial discharge is an important representation of the defect state of GIS equipment. The research on its positioning and measurement technology is an important way to effectively perceive the state of GIS equipment and analyze and warn. Generally, it is through the collection and discharge monitoring of high-frequency electromagnetic wave signals (Ultra High Frequency, UHF) inside GIS. When it is determined that the partial discharge signal is monitored, the partial discharge source is located, thereby realizing state perception and long-term analysis and warning, and ensuring the safe and stable operation of GIS equipment.

[0004] When using UHF signals for discharge detection, the multiple reflections of UHF signals in different propagation paths in the complex topological structure of GIS will affect the positioning accuracy. The multipath effect will cause different time delays for the signal to reach the UHF sensor, thereby introducing positioning errors and making it impossible for the positioning system to accurately identify the location of the discharge source. The high-precision positioning models used in related technologies (such as fuzzy neural networks, deep learning models, etc.) have complex structures, and the training and tuning processes require a lot of computing resources, making them difficult to apply in real time.

[0005] To this end, a method for fast iterative search for the optimal matching position by fusing UHF signal measurement values ​​with simulation models is proposed, but the technical difficulties are:

[0006] (1) Most of the existing partial discharge signal acquisition devices use PRPD (Phase Resolved Partial Discharge) discharge spectra and other forms to transmit compressed partial discharge signals to the host computer, and cannot restore the detected high-frequency, complete full-time waveform partial discharge point time domain signals to the host computer, and cannot meet the needs of accurate positioning of partial discharge signals inside GIS; in addition, the existing UHF signal acquisition requires high-sampling rate data acquisition equipment. Since the sampling rate is proportional to the positioning accuracy, improving the sampling accuracy is bound to face problems such as insufficient chip computing resources. The high sampling rate brings a large amount of data, which increases the burden of processing and transmission. Transmission delays and data synchronization problems will affect real-time performance; therefore, the relevant technology lacks high-speed and high-performance real-time acquisition / matching / detection methods for partial discharge signals.

[0007] (2) When performing electromagnetic field simulation on non-uniform grids, traditional numerical methods have to face problems such as complex coupling processing between coarse and fine grids and difficulty in establishing boundary conditions. At the same time, non-uniform grid division simulation requires high computing resources. When simulating with a high number of grid divisions and sparse density, the calculation time is long and the simulation speed is slow. When dealing with high-frequency GIS partial discharge detection problems, it will seriously affect the detection time. If coarse grid simulation is used, although the time cost can be reduced, the simulation accuracy is low, which will affect the quality of partial discharge detection. Therefore, there is a lack of fast and high-precision electromagnetic field distribution simulation technology suitable for multi-structure GIS equipment in related technologies. Electromagnetic field distribution simulation is difficult to provide a theoretical basis for GIS high-performance real-time perception.

[0008] (3) Since real-time measured UHF signals usually need to be preprocessed and cleaned to reduce the impact of noise and extract effective signals as much as possible, and the quality and clarity of the input data required for fast electromagnetic field simulation are usually higher than the capabilities of actual measured data, it is difficult to match the simulation results with the measured data, and fast electromagnetic field simulation cannot be organically combined with noisy measurement data. Summary of the invention

[0009] The technical problem to be solved by the present invention is how to achieve high-performance real-time status perception and early warning based on GIS field-electricity fusion.

[0010] The present invention solves the above technical problems by the following technical means:

[0011] The present invention proposes a GIS field-electricity fusion real-time state perception and early warning system, the system includes a signal synchronous acquisition and measurement embedded device and an industrial computer, the industrial computer is deployed with a simulator, a location searcher and an early warning device;

[0012] The signal synchronous acquisition and measurement embedded device is used to monitor the electromagnetic field signal inside the GIS according to the pre-screened strong correlation characteristics, and obtain a full time domain waveform diagram including a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal when the signal contains a local discharge signal;

[0013] The simulator is used to roughly locate the local discharge source based on the full time domain waveform of the GIS local discharge, and use the rough positioning result as the initial injection point to perform simulation enhancement calculation to obtain simulation enhancement data of the local discharge source;

[0014] The position searcher is used to iteratively search the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement result to obtain the actual discharge position of the partial discharge source;

[0015] The early warning device is used for issuing an early warning based on the full waveform in the time domain and the early warning information including the actual discharge position of the partial discharge source.

[0016] Furthermore, the signal synchronous acquisition and measurement embedded device comprises:

[0017] A signal acquisition position determination module is used to determine the installation position of each sensor in the signal acquisition module;

[0018] Signal acquisition module, used to collect electromagnetic field signals inside GIS;

[0019] A signal processing front end, used for performing signal conditioning on the electromagnetic field signal transmitted by the signal acquisition module, and transmitting the conditioned signal to an analog-to-digital converter;

[0020] An analog-to-digital converter, used to convert the conditioned signal from an analog signal to a digital signal and transmit it to a signal measurement module;

[0021] The signal measurement module is used to monitor the digital signal based on the strong correlation features obtained in advance, and when it is determined that a local discharge signal is monitored, a full time domain waveform diagram including a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal is obtained, wherein the strong correlation features include time domain strong correlation features and frequency domain strong correlation features.

[0022] Furthermore, the signal synchronous acquisition and measurement embedded device also includes a feature screening module, which is used to use the time-frequency domain features used to detect local discharge signals as nodes of the graph neural network to perform correlation strength screening, so as to obtain the time domain strong correlation features and the frequency domain strong correlation features, wherein the lines between the nodes represent the relationship between the features represented by the nodes.

[0023] Furthermore, the feature screening module includes:

[0024] Feature map construction unit for complete feature maps in graph neural networks middle, is a node set, is the edge set, let Respectively represent nodes The eigenvectors of the edges, the eigenvectors of the nodes and the state vectors of its surrounding nodes and nodes The feature vectors of the surrounding nodes;

[0025] The aggregation update unit is used to aggregate and update the information of input nodes and edges according to the local transfer function of updating the node state, and output the node label. The formula is expressed as:

[0026]

[0027]

[0028] In the formula, is the local transition function for updating the node state, is the local output function, is the state vector learned by the graph neural network iteration, is the node label;

[0029] The iteration unit is used to calculate the correlation between a node and its adjacent nodes in the continuous iteration of the graph neural network, and obtain the strong correlation features in the time domain and the strong correlation features in the frequency domain.

[0030] Furthermore, the iteration unit is specifically used for:

[0031] set up and are the vectors constructed by superimposing the state vectors of all nodes, all output labels, the feature vectors of all nodes and edges, and the feature vectors of all nodes. The formula can be written in a more compact form as follows:

[0032]

[0033]

[0034] in, express No. Iterations, Represented by the eigenvector and the The state vector of the iteration is obtained by the global transfer function The state vector of all nodes in the iteration, and are the global transfer function and the global output function, respectively, and are the local transfer functions of all nodes. and local output function Stacking;

[0035] During the iteration process, nodes with similar states, nodes with complementary states, and the influence of each node on the overall graph neural network are determined. One of the nodes with similar states is selected, nodes with complementary states are merged into one node, and the node with the greatest influence on the overall graph neural network is used as a strong correlation feature.

[0036] Furthermore, the signal measurement module includes:

[0037] A cache unit, used for performing cache processing on the digital signal;

[0038] A first monitoring unit is used to extract the time domain strong correlation feature from the digital signal, determine that a partial discharge signal is monitored based on the time domain strong correlation feature, and output a first time domain waveform diagram during discharge;

[0039] A second monitoring unit is used to extract the frequency domain strong correlation feature from the digital signal, determine that a partial discharge signal is monitored based on the frequency domain strong correlation feature, and output a frequency domain waveform diagram during discharge;

[0040] A waveform synthesis unit is used to obtain a full time domain waveform diagram of a partial discharge signal based on the first time domain waveform diagram and the frequency domain waveform diagram.

[0041] Furthermore, the second monitoring unit includes:

[0042] A Fourier transform subunit, used for performing at least two fast Fourier transforms on the electromagnetic field signal to obtain frequency domain data of at least two frequency bands;

[0043] The normalization subunit is used to extract the frequency domain strong correlation features from the frequency domain data of each frequency band, and after analyzing and matching the frequency domain strong correlation features, when it is determined that the frequency domain data of each frequency band exceeds the second discharge threshold, the frequency domain waveform corresponding to each frequency band is normalized to obtain a frequency domain waveform diagram during discharge.

[0044] Furthermore, the waveform synthesis unit comprises:

[0045] A waveform conversion subunit, used for converting the frequency domain waveform diagram into a second time domain waveform diagram during discharge;

[0046] The waveform synthesis subunit is used to synthesize the first time domain waveform diagram and the second time domain waveform diagram to obtain a full time domain waveform diagram of the partial discharge signal.

[0047] Furthermore, the simulator comprises:

[0048] A simulation module, used for using the coarse positioning result of the local discharge source obtained based on the full time domain waveform diagram as the initial injection point of the three-dimensional simulation model of the GIS equipment to simulate the local discharge phenomenon of the GIS equipment, and performing simulation calculation to obtain a coarse grid simulation result;

[0049] The enhancement module is used to enhance the coarse grid simulation result by using an enhanced model to obtain simulation enhanced data.

[0050] Furthermore, the simulation module includes:

[0051] The spatial discretization unit is used to discretize the three-dimensional simulation model of the GIS equipment into a spatial grid and determine the corresponding relationship between the number and spatial coordinate data of each grid node;

[0052] The differential simulation unit is used to inject the coarse positioning results as the initial injection point into the corresponding spatial grid to simulate the partial discharge phenomenon of the GIS equipment, convert the Maxwell equations used to describe the electromagnetic field of the GIS into a fourth-order matrix, and use the four-step HIE-FDTD algorithm to solve the fourth-order matrix and calculate the field strength data of each grid node.

[0053] Furthermore, the differential simulation unit comprises:

[0054] The matrix reconstruction subunit is used to convert the Maxwell equations used to describe the electromagnetic field of GIS into a sixth-order matrix form:

[0055]

[0056] 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;

[0057] The matrix conversion subunit is used to convert the sixth-order matrix form into a fourth-order rectangular form:

[0058]

[0059] In the formula, for ,

[0060] for ;

[0061] The solving subunit is used to solve the fourth-order matrix form using the four-step HIE-FDTD algorithm and calculate the field strength data of each grid node.

[0062] Furthermore, the solving subunit is used to perform the following steps:

[0063] 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;

[0064] The pursuit method is used to solve the three diagonal implicit equations and obtain the grid node field strength data.

[0065] Furthermore, the enhanced model adopts a differential deep learning network model, and the coarse grid simulation result includes coarse grid structure data and coarse grid field strength data; 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;

[0066] 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;

[0067] 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;

[0068] Based on the similarity features, the simulation enhancement data is calculated.

[0069] Furthermore, the self-adjusting module includes 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;

[0070] 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.

[0071] Furthermore, the differential convolution module includes a differential convolution layer, a size integration layer, a self-attention mechanism layer and a third convolution layer connected in sequence, and the third convolution layer is connected to an activation function;

[0072] 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;

[0073] 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.

[0074] Furthermore, 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.

[0075] Furthermore, the structural similarity network includes a first branch network, a second branch network, a second addition operation and a first multilayer 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 multilayer perceptron, and the multilayer perceptron is followed by an activation function.

[0076] Furthermore, the first branch network includes a convolutional neural network layer CNN and a batch normalization operation connected in sequence, and the regularization operation is followed by an activation function;

[0077] The second branch network includes a second multilayer perceptron, and the second multilayer perceptron is followed by an activation function.

[0078] Furthermore, the using the structural similarity network to match the fine grid enhanced grid structure feature and the fine grid enhanced field strength feature to calculate the similarity feature includes:

[0079] 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:

[0080]

[0081] 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;

[0082] 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:

[0083]

[0084] 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;

[0085] 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:

[0086]

[0087] 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;

[0088] Based on the field strength and structure combination characteristics, the similarity characteristics are calculated, and the formula is expressed as:

[0089]

[0090] In the formula, For the similarity feature, Operations performed by the first multilayer perceptron.

[0091] Further, the calculating the simulation enhancement data based on the similarity feature includes:

[0092] 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;

[0093] Calculating grid structure consistency based on the regularized coarse grid structure data and the regularized fine grid enhanced grid structure features;

[0094] Calculating grid field strength consistency based on the regularized coarse grid field strength data and the regularized fine grid enhanced field strength characteristics;

[0095] Calculating grid field strength loss compensation based on the grid structure consistency and the grid field strength consistency;

[0096] 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;

[0097] The simulation enhancement data is calculated based on the fine grid enhanced field strength characteristics, the grid field strength loss compensation and the similarity characteristics.

[0098] Further, the simulation enhancement data includes an enhanced discharge position and an enhanced discharge intensity, and the position searcher includes:

[0099] The state transition model building module uses the Markov decision model to reconstruct the state transition model corresponding to the process of the ant colony algorithm exploring the actual discharge position of the local discharge source with the coarse positioning result of the local discharge source as the center, according to the enhanced discharge position and enhanced discharge intensity at each moment;

[0100] An heuristic space parameterization module generates a heuristic metric using a heuristic learner based on a graph neural network, and converts the state transition model into a location exploration model that is affected by the heuristic metric and requires graph traversal;

[0101] The iterative search module is used to iteratively search the position exploration model using a neural-guided perturbation interleaved local search algorithm to obtain an actual discharge position of a local discharge source.

[0102] Furthermore, the state transition model construction module is used to:

[0103] According to the simulated discharge position and simulated discharge intensity at each moment, the state space corresponding to the actual discharge position of the local discharge source is constructed in the simulator :

[0104]

[0105]

[0106] Where: Indicates that in the simulator The local discharge source node is explored at The corresponding status, Indicates The simulated discharge position at each moment, Indicates The simulated discharge intensity at each moment, It represents the initial discharge intensity of the local discharge source obtained based on the full time domain waveform of GIS local discharge. and initial discharge position As the initial state, represents the exploration cycle, ;

[0107] Constructing the action space corresponding to the actual discharge location of the PD source in the simulator :

[0108]

[0109] Where: Indicates The action of the moment Greedy strategy is the standard choice of action ;

[0110] The state transition model corresponding to the process of the reconstructed ant colony algorithm exploring the actual discharge position of the local discharge source with the initial discharge position of the local discharge source as the center is:

[0111]

[0112] Where: Indicates in action The discharge node To the discharge node The probability of transfer, Indicates that in the simulator The local discharge source node is explored at The corresponding state, Represented by the node Transfer to Node The corresponding pheromone concentration, Represented by the node Transfer to Node The corresponding heuristic function is, Represented by the node Transfer to the included Nodes in The corresponding pheromone concentration, Represented by the node Transfer to the included Nodes in The corresponding heuristic function is, and Represent the pheromone concentration and the weight of the heuristic function, Represents the set of non-uniform grid local discharge source nodes that are allowed to be selected at the next moment.

[0113] Furthermore, by the node To Node The reward function for the transfer is , Indicates The measured discharge intensity is obtained by processing the full waveform in the time domain collected by the sensor at each moment. Indicates The partial discharge intensity obtained by simulation at each moment

[0114] Furthermore, the heuristic space parameterization module is used to transform the graph neural network The layer is The node and Edge features connecting nodes Mapping to heuristic metrics ;

[0115] The state transition model is transformed into a state transition model that is influenced by the heuristic metric and requires The location exploration model of step graph traversal is:

[0116]

[0117] Where: represents the location exploration model, Indicates that in the simulator The local discharge source node is explored at The corresponding status, Indicates that in the simulator The local discharge source node is explored at The corresponding status, represents the exploration cycle, .

[0118] Furthermore, the location searcher further comprises a training module, which is used to:

[0119] The gradient strategy is used to train the heuristic learner based on the graph neural network. The objective function used in the training process is for:

[0120]

[0121] Where: represents the objective function corresponding to the actual discharge location of the PD source using the neural-guided perturbation interleaved local search algorithm, Indicates balance and Parameters, In the heuristic metric The expected value of the actual discharge position of the local discharge source under the influence of represents the state space corresponding to the exploration of the actual discharge position of the PD source, represents the objective function;

[0122] Among them, the objective function The gradient of , the formula is:

[0123]

[0124] Where: represents the average target value for directly exploring the actual discharge location of the PD source, represents the average target value of the actual discharge location of the PD source explored using the local search algorithm with neural-guided perturbation interleaving, In the heuristic metric Explore the gradient of the actual discharge location of the partial discharge source under the influence;

[0125] When the maximum number of iterations is reached Or the maximum number of iterations has not been reached But the objective function The training ends when Indicates the minimum threshold corresponding to the objective function.

[0126] Furthermore, the iterative search module is used to:

[0127] A local search unit, used to iteratively search the position exploration model using a local search algorithm to obtain a local optimal solution;

[0128] A perturbation unit is used to perform neural-guided perturbation on the local optimal solution obtained by the current iterative search, and obtain the optimal exploration scheme of the actual discharge position of the local discharge source of the current iterative search through the local search of neural-guided perturbation interleaving;

[0129] A pheromone concentration updating unit, used for updating the pheromone concentrations between the local discharge source nodes after all the local discharge source nodes are selected;

[0130] The exploration unit is used to determine the optimal exploration scheme of the actual discharge position of the partial discharge source when the iterative search reaches the iterative convergence condition, and realize the matching of the actual discharge position of the partial discharge source based on the optimal exploration scheme.

[0131] Furthermore, the process formula of the disturbance unit exploring the optimal exploration scheme of the actual discharge position of the current iteration partial discharge source is expressed as:

[0132]

[0133] Where: represents the optimal exploration scheme of the actual discharge position of the local discharge source in the current iteration obtained by performing neural-guided perturbation on the local optimal solution, is the local optimal solution; represents the number of perturbation moves, Represents a heuristic metric.

[0134] Furthermore, a fault classification module is deployed in the early warning device, and the fault classification module includes:

[0135] A deep feature extraction network is used to obtain deep features corresponding to the UHF signal, wherein the deep feature extraction network includes a feature extraction network and an output network, the feature extraction network is formed by stacking a plurality of feature extraction layers, and each of the feature extraction layers includes a channel attention module and a spatial attention module connected in sequence;

[0136] A reinforcement learning unit, used to construct a data set using the deep features, and train a GIS partial discharge feature matching Markov model using a deep reinforcement learning algorithm to obtain an optimal GIS partial discharge feature matching result;

[0137] The early warning unit generates early warning information based on the optimal GIS local discharge feature matching result and the actual discharge position of the local discharge source to perform fault early warning.

[0138] In addition, the present invention also proposes a GIS field-electricity fusion real-time state perception and early warning method, which is used to realize GIS fault early warning by using the GIS field-electricity fusion real-time state perception and early warning system as described above. The method includes:

[0139] When the electromagnetic field signal inside the GIS is monitored according to the strong correlation characteristics obtained in advance, including the partial discharge signal, a full time domain waveform diagram including the time domain waveform diagram and the frequency domain waveform diagram of the partial discharge signal is obtained;

[0140] Based on the full time domain waveform of GIS partial discharge, the partial discharge source is roughly located, and the simulation enhancement calculation is performed using the rough positioning result as the initial injection point to obtain simulation enhancement data of the partial discharge source;

[0141] Iteratively search the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement result to obtain the actual discharge position of the partial discharge source;

[0142] An early warning is provided based on the full waveform in the time domain and the early warning information including the actual discharge position of the partial discharge source.

[0143] The advantages of the present invention are:

[0144] (1) The present invention detects the digital signal converted from the electromagnetic field signal based on the strong correlation characteristics obtained by pre-screening as an indicator, and when it is determined that the local discharge signal is monitored, it realizes the acquisition of a high-precision full-time domain waveform of the GIS local discharge; then, the local discharge source is roughly located based on the full-time domain waveform of the GIS local discharge, and the rough positioning result is injected into the simulator to simulate the local discharge phenomenon of the GIS equipment and perform simulation enhancement calculation of the simulation enhancement data of the local discharge source, thereby realizing fast and high-precision GIS electromagnetic field simulation; finally, iterative search is performed based on the full-time domain waveform and simulation enhancement data collected in real time to obtain the actual discharge position of the local discharge source; and early warning is realized based on the actual discharge position and other information. Therefore, the present invention realizes high-performance real-time status perception and early warning based on GIS field-electricity fusion.

[0145] (2) The present invention substitutes the time-frequency domain features of multiple electromagnetic field signal detection into the graph neural network, and uses each time-frequency domain feature as a node of the graph neural network. The lines between the nodes represent the association between the features represented by the nodes. The graph neural network is iterated to perform association strength screening to obtain the features with the highest correlation, i.e., the strongly correlated features. The local discharge signal is then detected based on the screened strongly correlated features. Although the amount of calculation is reduced, the accuracy of the local discharge signal detection is not affected because the correlation between the indicators is taken into account. Therefore, the present invention improves the accuracy of the local discharge signal detection while greatly reducing the amount of calculation.

[0146] (3) Since GIS electromagnetic field simulation focuses more on the accurate simulation of the detailed interaction between equipment and electromagnetic fields, the present invention designs and constructs a differential deep learning network model, which uses convolution combined with a differential algorithm to process coarse grid field strength data. This model can accurately simulate the electromagnetic field distribution of GIS, especially when dealing with complex geometric structures and boundary conditions. In addition, the deep learning network is used to optimize the simulation process, improve the simulation accuracy, and accelerate the simulation process, effectively shortening the design and evaluation cycle, thereby taking into account both the accuracy and efficiency of GIS electromagnetic simulation.

[0147] (4) The present invention takes the preliminarily calculated local discharge source as the center, and proposes a heuristic fast iterative search algorithm for the optimal match of the measured values ​​in its neighborhood. The fast iterative search problem of the optimal matching position of the discharge source is input into the graph neural network to obtain its heuristic metric as a substitute for the expert design heuristic, and based on this, the ant colony algorithm is iterated. Under the framework of the ant colony algorithm, local search of the constructed solution will obtain a better solution. The deep ant colony algorithm uses the graph neural network to generate the heuristic metric, reducing the need for expert knowledge; and combined with the probabilistic local search, the local search method of neural-guided perturbation interleaving will achieve better performance to ensure efficient solution and realize the fast iterative search of the optimal matching position of the discharge source.

[0148] Additional aspects and advantages of the present invention will be given in part in the following description and in part will be obvious from the following description, or will be learned through practice of the present invention. BRIEF DESCRIPTION OF THE DRAWINGS

[0149] Figure 1 It is a structural schematic diagram of a GIS field-electricity fusion real-time status perception and early warning system proposed in the first embodiment of the present invention;

[0150] Figure 2 It is a principle block diagram of the embedded device for synchronous signal acquisition and measurement in the first embodiment of the present invention;

[0151] Figure 3 It is a schematic diagram of the calculation process of the global waveform diagram of the partial discharge signal in the first embodiment of the present invention;

[0152] Figure 4 is a complete feature graph of the graph neural network in the first embodiment of the present invention;

[0153] Figure 5 It is a schematic diagram of the iterative process of the neural network in the first embodiment of the present invention;

[0154] Figure 6 It is a block diagram of the principle of implementing GIS electromagnetic field distribution simulation enhancement by finite difference time domain in the second embodiment of the present invention;

[0155] Figure 7 It is a principle block diagram of the enhanced simulation of GIS electromagnetic field distribution using time-domain finite element in the third embodiment of the present invention;

[0156] Figure 8 It is a principle block diagram of GIS electromagnetic field coarse grid data enhancement based on differential network in the fourth embodiment of the present invention;

[0157] Fig. 9 It is a schematic diagram of the structure of the enhanced network in the differential deep learning network model in the fourth embodiment of the present invention;

[0158] Fig.10 Schematic diagram of the design principle of the differential convolution module in the fourth embodiment of the present invention;

[0159] Fig.11 It is a schematic diagram of the structure of the structural similarity network in the differential deep learning network model in the fourth embodiment of the present invention;

[0160] Fig.12 It is a principle block diagram of GIS electromagnetic field simulation data enhancement based on PU-GAN in Embodiment 5 of the present invention;

[0161] Fig.13 is a schematic diagram of the structure of the generator in the fifth embodiment of the present invention;

[0162] Fig.14 is a schematic diagram of the structure of a feature extraction module in Embodiment 5 of the present invention;

[0163] Fig.15 is a schematic diagram of a structure of a module with a similar structure in Embodiment 5 of the present invention;

[0164] Fig.16 This is a schematic diagram of the structure of the field strong consistency module in the fifth embodiment of the present invention;

[0165] Fig.17 is a schematic diagram of the structure of the self-attention unit in the fifth embodiment of the present invention;

[0166] Fig.18 It is a principle block diagram of GIS partial discharge optimal position matching based on deep ant colony algorithm in embodiment 6 of the present invention;

[0167] Fig.19 It is a schematic diagram of the optimal location search process of GIS partial discharge based on the deep ant colony algorithm in the sixth embodiment of the present invention;

[0168] Fig. 20 It is a schematic diagram of a GIS partial discharge feature extraction process based on deep reinforcement learning in Embodiment 7 of the present invention;

[0169] Fig.21 It is a flow chart of a GIS field-electricity fusion real-time status perception and early warning method proposed in Embodiment 8 of the present invention. DETAILED DESCRIPTION

[0170] In order to make the purpose, technical solution and advantages of the embodiments of the present invention clearer, the technical solution in the embodiments of the present invention will be clearly and completely described below in combination with the embodiments of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0171] Embodiment 1

[0172] like Figure 1 As shown, the first embodiment of the present invention proposes a GIS field-electricity fusion real-time state perception and early warning system, the system includes a signal synchronous acquisition and measurement embedded device 100 and an industrial computer 200, the industrial computer 200 is deployed with a simulator 201, a location searcher 202 and an early warning device 203, the output of the signal synchronous acquisition and measurement embedded device 100 is respectively connected to the simulator 201 and the location searcher 202, the output of the simulator 201 is connected to the location searcher 202, and the output of the location searcher 202 is connected to the early warning device 203;

[0173] The signal synchronous acquisition and measurement embedded device 100 is used to monitor the electromagnetic field signal inside the GIS according to the pre-screened strong correlation characteristics, and obtain a full time domain waveform diagram including a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal when the signal contains a local discharge signal;

[0174] The simulator 201 is used to roughly locate the local discharge source based on the full time domain waveform of the GIS local discharge, and use the rough positioning result as the initial injection point to perform simulation enhancement calculation to obtain simulation enhancement data of the local discharge source;

[0175] The position searcher 202 is used to iteratively search the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement result to obtain the actual discharge position of the partial discharge source;

[0176] The alarm device 203 is used for issuing an alarm based on the full waveform in the time domain and the alarm information including the actual discharge position of the partial discharge source.

[0177] This embodiment detects the digital signal converted from the electromagnetic field signal based on the strong correlation characteristics obtained by pre-screening as indicators, and when it is determined that the local discharge signal is monitored, it realizes the acquisition of a high-precision full-time domain waveform of the GIS local discharge; then, the local discharge source is roughly located based on the full-time domain waveform of the GIS local discharge, and the rough positioning result is injected into the simulator to simulate the local discharge phenomenon of the GIS equipment and perform simulation enhancement calculation of the simulation enhancement data of the local discharge source, so as to realize fast and high-precision GIS electromagnetic field simulation; finally, iterative search is performed based on the full-time domain waveform and simulation enhancement data collected in real time to obtain the actual discharge position of the local discharge source; and early warning is realized based on the actual discharge position and other information. Therefore, the present invention realizes high-performance real-time status perception and early warning based on GIS field-electricity fusion.

[0178] As a further preferred technical solution, the signal synchronous acquisition and measurement embedded device includes a signal acquisition position determination module, a signal acquisition module, a signal processing front end, an analog-to-digital converter and a signal measurement module, wherein:

[0179] The signal acquisition position determination module is used to determine the installation position of each sensor in the signal acquisition module;

[0180] The signal acquisition module is used to collect electromagnetic field signals inside the GIS;

[0181] The signal processing front end is used to perform signal conditioning on the electromagnetic field signal transmitted by the signal acquisition module, and transmit the conditioned signal to the analog-to-digital converter;

[0182] The analog-to-digital converter is used to convert the conditioned signal from an analog signal to a digital signal and then transmit it to the signal measurement module;

[0183] The signal measurement module is used to monitor the digital signal based on the strongly correlated features obtained in advance, and when it is determined that a local discharge signal is monitored, a full time domain waveform diagram including a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal is obtained, wherein the strongly correlated features include time domain strong correlation features and frequency domain strong correlation features.

[0184] This embodiment realizes multi-channel acquisition of electromagnetic field signals inside the GIS by setting up a multi-channel acquisition module, and conditions the acquired GIS internal electric field signals to obtain high-performance electromagnetic field signals. Then, the digital signal converted from the electromagnetic field signal is detected based on the strong correlation characteristics obtained in advance as an indicator, and when it is determined that a local discharge signal is monitored, the time domain waveform diagram and the frequency domain waveform diagram are respectively obtained according to the time domain strong correlation characteristics and the frequency domain strong correlation characteristics, and the time domain waveform diagram and the frequency domain waveform diagram are integrated to obtain a full time domain waveform diagram; therefore, the present invention can realize the acquisition of a high-precision full time domain waveform diagram of GIS local discharge.

[0185] As a further preferred technical solution, Figure 2 As shown, the signal processing front end includes amplifying and filtering circuits having the same number as the sensors in the signal acquisition module, and the amplifying and filtering circuits are connected to the sensors in a one-to-one correspondence;

[0186] The amplifying and filtering circuit includes a radio frequency gain circuit, a bandpass filter and a bandstop filter which are connected in sequence, wherein the frequency of the bandpass filter is 300MHz~1500MHz, and the frequency of the bandstop filter is 800MHz~1000MHz.

[0187] It should be noted that this embodiment performs bandpass and band-stop filtering according to the effective frequency band and common interference frequency band of the UHF signal in GIS, and is more suitable for UHF signal acquisition in GIS. The set bandpass filter is used to extract useful signals in a specific frequency band and filter out irrelevant interference signals and noise, and a low-noise, wide-band, cascadeable and high-linearity signal amplifier can be connected after the band-stop filter.

[0188] As a preferred technical solution, the analog-to-digital converter adopts an ADC chip with a data conversion rate of 4 channels at 4Gsps or 1 channel at 5Gsps.

[0189] It should be noted that the synchronous acquisition and measurement embedded device provided in this embodiment can realize 4GSPS 4-channel high-speed data acquisition and real-time processing, and supports multi-channel sampling, which can realize simultaneous acquisition of 4-channel data, with a signal input range of 1.4V and 12-bit data at a sampling rate of 4GSPS. At the same time, the signal processing logic with fixed functions is placed in the hardware FPGA to improve the parallel speed of the system and realize multi-channel, high-precision GIS internal electromagnetic field signal acquisition.

[0190] As a further preferred technical solution, the signal measurement module includes a cache unit, a first monitoring unit, a second monitoring unit and a waveform synthesis unit, wherein:

[0191] The cache unit is used to cache the digital signal;

[0192] The first monitoring unit is used to extract the time domain strong correlation feature from the digital signal, determine that a partial discharge signal is monitored based on the time domain strong correlation feature, and output a first time domain waveform diagram during discharge;

[0193] The second monitoring unit is used to extract the frequency domain strong correlation feature from the digital signal, determine that a partial discharge signal is monitored based on the frequency domain strong correlation feature, and output a frequency domain waveform diagram during discharge;

[0194] The waveform synthesis unit is used to obtain a full time domain waveform diagram of the partial discharge signal based on the first time domain waveform diagram and the frequency domain waveform diagram.

[0195] As a further preferred technical solution, the first monitoring unit is used to analyze the time domain strong correlation feature to determine whether it exceeds a first discharge threshold, and if so, determine that a partial discharge signal is monitored, and if not, determine that no partial discharge signal is monitored;

[0196] The second monitoring unit is used to analyze the frequency domain strong correlation feature to determine whether it exceeds a second discharge threshold, and if so, determine that a partial discharge signal is monitored; otherwise, determine that no partial discharge signal is monitored.

[0197] As a further preferred technical solution, when no partial discharge signal is detected, the digital signal stored in the cache unit is deleted.

[0198] It should be noted that the collected time domain data is completely sent to the high-performance storage cache unit (Buffer). If it is determined that no partial discharge UHF signal is detected, the detected time domain data is deleted from the Buffer to release storage space for the next collected time domain data.

[0199] As a further preferred technical solution, the second monitoring unit includes:

[0200] A Fourier transform subunit, used for performing at least two fast Fourier transforms on the electromagnetic field signal to obtain frequency domain data of at least two frequency bands;

[0201] The normalization subunit is used to extract the frequency domain strong correlation features from the frequency domain data of each frequency band, and after analyzing and matching the frequency domain strong correlation features, when it is determined that the frequency domain data of each frequency band exceeds the second discharge threshold, the frequency domain waveform corresponding to each frequency band is normalized to obtain a frequency domain waveform diagram during discharge.

[0202] When the amount of collected signal data is large, if FFT calculation is performed directly, the processing speed is slow due to the large amount of data, which is not conducive to real-time signal detection and processing. Therefore, this application adopts array processing, such as Figure 3 As shown, the FFT results are divided into four frequency bands for parallel processing (the frequency bands are 200M-525M, 525M-850M, 850M-1175M, and 1175M-1500M respectively), and the correlations in the four frequency bands are calculated respectively. This can increase the number of correlation determination bases to multiple, thereby greatly increasing the amount of information reflecting the relative deformation of the waveform. At the same time, it can also fully reflect the changes in the frequency response curve in different frequency ranges due to different degrees of discharge, and finally the waveforms of multiple frequency bands are fused.

[0203] Therefore, the array processing mechanism of the present application is a frequency domain segmentation and combination mechanism. Each part in the array only processes a part of the frequency domain, and then merges them together to restore the complete frequency domain waveform. This mechanism disperses the noise signal superimposed on the processed frequency domain data to the segmented parts, reducing the stacking of noise signals, which will help to achieve high-sensitivity detection of local discharge UHF signals.

[0204] As a further preferred technical solution, the waveform synthesis unit includes:

[0205] A waveform conversion subunit, used for converting the frequency domain waveform diagram into a second time domain waveform diagram during discharge;

[0206] The waveform synthesis subunit is used to synthesize the first time domain waveform diagram and the second time domain waveform diagram to obtain a full time domain waveform diagram of the partial discharge signal.

[0207] Furthermore, this embodiment selects Xilinx UltraScale+MPSoC FPGA chip to process digital signals.

[0208] It should be noted that this embodiment receives sampling data from the analog-to-digital converter ADC based on the cache unit buffer, calculates characteristic indicators of the transmitted time domain waveform data, matches the data, and simultaneously performs FFT to convert the time domain waveform into a frequency domain waveform, and calculates characteristic indicators after the frequency domain waveform is arrayed, and performs data matching, normalizes the obtained frequency domain waveform, and performs IFFT after normalization to obtain the time domain waveform, and after combining it with the waveform graph directly obtained in the time domain, outputs the full time domain waveform graph.

[0209] This embodiment proposes a frequency domain array method and a real-time digital matching method, and develops an embedded device for synchronous acquisition and measurement of GIS partial discharge UHF signals based on ultra-high-speed ADC and FPGA architecture to obtain a high-precision full-time domain waveform diagram of the partial discharge UHF signal.

[0210] It should be noted that since there are multiple time-frequency domain indicators for detecting partial discharge signals, if all time-frequency domain indicators are calculated to detect partial discharge signals, the detection results are theoretically very accurate. However, if detection and judgment are based on all time-frequency domain indicators, it will take up a lot of computing resources, and the computing resources of the current hardware cannot perform the calculation. Randomly selecting certain features for calculation cannot meet the accuracy requirements.

[0211] This embodiment designs a feature screening module to use a neural network to screen out strongly correlated features for subsequent partial discharge signal detection, thereby improving the accuracy of partial discharge signal detection while greatly reducing the amount of calculation.

[0212] Specifically, the time-frequency domain features include margin, skewness, root mean square, mean, variance, peak-to-peak value, shape factor, kurtosis, impulse factor and C index; the frequency domain features include mean frequency, frequency concentration, frequency center, root mean square frequency and main frequency band position change.

[0213] Specifically, the time-frequency domain characteristics of detecting partial discharge signals are shown in Table 1:

[0214] Table 1 Proposed time-frequency domain features

[0215]

[0216] Where: is the maximum absolute value (peak value), is the square root amplitude, is the skewness, Time domain waveform data The third-order center distance, Time domain waveform data The standard deviation of is the root mean square, For the Time domain data, is the total number of time domain data points, is the shape factor, For the deviation: is the kurtosis, is the pulse factor, is the time domain waveform data, is the mean frequency, is the amplitude of each frequency on the spectrum, is a fixed coefficient used to adjust the ratio in the formula. is the frequency concentration index 1, is the frequency concentration index 2, is the frequency concentration index 3, is the frequency center index, is the frequency of each point on the spectrum, is the frequency concentration index 4, is the root mean square frequency index, is the main frequency band position change index 1, is the main frequency band position change index 2, is the frequency concentration index 5, is the frequency concentration index 6, is the frequency concentration index 7, is the frequency concentration index 8, For the multiplication sign.

[0217] It should be noted that the indicators listed in Table 1 are common time-frequency domain indicators, some of which contain certain characteristics of UHF and are common UHF signal indicators. These characteristic indicators have more or less information that can be used for UHF signal detection, but this information may be repeated or complementary in multiple features. Therefore, this embodiment uses a graph neural network to screen out indicators with strong correlation among these characteristic indicators.

[0218] As a further preferred technical solution, the feature screening module includes:

[0219] Feature map construction unit for complete feature maps in graph neural networks middle, is a node set, is the edge set, let Respectively represent nodes The eigenvectors of the edges, the eigenvectors of the nodes and the state vectors of its surrounding nodes and nodes The feature vectors of the surrounding nodes;

[0220] The aggregation update unit is used to aggregate and update the information of input nodes and edges according to the local transfer function of updating the node state, and output the node label. The formula is expressed as:

[0221]

[0222]

[0223] In the formula, is the local transition function for updating the node state, is the local output function, is the state vector learned by the graph neural network iteration, is the node label;

[0224] The iteration unit is used to calculate the correlation between a node and its adjacent nodes in the continuous iteration of the graph neural network, and obtain the strong correlation features in the time domain and the strong correlation features in the frequency domain.

[0225] As a further preferred technical solution, the iteration unit is specifically used for:

[0226] set up and are the vectors constructed by superimposing the state vectors of all nodes, all output labels, the feature vectors of all nodes and edges, and the feature vectors of all nodes. The formula can be written in a more compact form as follows:

[0227]

[0228]

[0229] in, express No. Iterations, Represented by the eigenvector and the The state vector of the iteration is obtained by the global transfer function The state vector of all nodes in the iteration, and are the global transfer function and the global output function, respectively, and are the local transfer functions of all nodes. and local output function Stacking;

[0230] During the iteration process, nodes with similar states, nodes with complementary states, and the influence of each node on the overall graph neural network are determined. One of the nodes with similar states is selected, nodes with complementary states are merged into one node, and the node with the greatest influence on the overall graph neural network is used as a strong correlation feature.

[0231] It should be noted that the characteristic vector of a node is some inherent state of the node, which will not change during the iteration process of the graph neural network. Specifically in this embodiment, it is some parameter indicators in the time-frequency domain feature calculation formula of the electromagnetic field signal, such as standard deviation, peak value, etc. The characteristic vector of an edge refers to the connection between two nodes. Specifically in this embodiment, it refers to the same parameter indicators or similar processing processes in the calculation formula between nodes. Connecting these similar feature nodes helps the model learn the relationship and mutual influence between these features. The state vector refers to the continuously updated node vector obtained according to the calculation during the iteration process; the node label is the target value used for supervised learning tasks. In this embodiment, each node represents a signal feature, and there is a corresponding label to indicate the category and state of these signal features.

[0232] This embodiment uses the graph neural network GNN to calculate the connection between a node and its adjacent nodes. Through multiple iterations, the relationship between the adjacent nodes of the adjacent nodes and the current node can also be found, which is very suitable for screening strong correlation features. This embodiment imports all the features of the GIS partial discharge time-frequency domain signal (as shown in Table 1 above) into the graph neural network as nodes in the network, and constructs a complete feature map as shown in Figure 4 As shown, Figure 4 Each node in the ,represents a single feature, and the lines between nodes represent the ,relationship between the features represented by the nodes. Some features are highly ,related, such as and The Chinese Communist Party , while the relationship between some features is weak, so no connection is made. Blank nodes represent features that are not closely related to other features.

[0233] like Figure 5 As shown in the figure, during the iteration process of the graph neural network, the feature vector of the node itself will not change with the network iteration; while the state vector between nodes changes with the iteration of the GNN network. Therefore, the graph neural network can calculate the connection between a node and its adjacent nodes. Through multiple iterations, the connection between the adjacent nodes of the adjacent nodes and the current node can also be found. Through the continuous iteration of the graph neural network, it can be found that the states of some nodes in the graph neural network are becoming more and more similar, the states of some nodes are complementary, and the influence of some nodes on the overall graph neural network becomes smaller. Select one of the similar nodes, combine the complementary nodes into one, and then select the node that has the greatest impact on the overall system, that is, obtain the strong correlation characteristics of the system.

[0234] The graph neural network used in this embodiment constrains the model in advance through the loss function, and finally uses the gradient descent algorithm to minimize the loss, thereby obtaining the optimal parameters of the network model and obtaining a trained graph neural network. This embodiment substitutes the features in Table 1 above into the graph neural network, performs correlation strength screening, and obtains the features with the highest correlation, that is, the results of the strong correlation features are the root mean square, pulse factor and margin index in the time domain, the root mean square frequency index, main frequency band position change index 1 and main frequency band position change index 2 in the frequency domain. Then, the local discharge signal is detected based on these screened strong correlation features. Although the amount of calculation is reduced, the accuracy of the local discharge signal detection is not affected because the correlation between the indicators is considered. Therefore, the present invention improves the accuracy of the local discharge signal detection under the premise of greatly reducing the amount of calculation.

[0235] Embodiment 2

[0236] Based on the contents disclosed in the above embodiment 1, the process of realizing GIS electromagnetic field distribution simulation enhancement by the simulator in the embodiment 1 through finite difference time domain is described in detail. The simulator includes:

[0237] A simulation module, used for using the coarse positioning result of the local discharge source obtained based on the full time domain waveform diagram as the initial injection point of the three-dimensional simulation model of the GIS equipment to simulate the local discharge phenomenon of the GIS equipment, and performing simulation calculation to obtain a coarse grid simulation result;

[0238] The enhancement module is used to enhance the coarse grid simulation result by using an enhanced model to obtain simulation enhanced data.

[0239] As a further preferred technical solution, Figure 6 As shown, the simulation module includes:

[0240] The spatial discretization unit is used to discretize the three-dimensional simulation model of the GIS equipment into a spatial grid and determine the corresponding relationship between the number and spatial coordinate data of each grid node;

[0241] The differential simulation unit is used to inject the coarse positioning results as the initial injection point into the corresponding spatial grid to simulate the partial discharge phenomenon of the GIS equipment, convert the Maxwell equations used to describe the electromagnetic field of the GIS into a fourth-order matrix, and use the four-step HIE-FDTD algorithm to solve the fourth-order matrix and calculate the field strength data of each grid node.

[0242] As a further preferred technical solution, the differential simulation unit includes:

[0243] The matrix reconstruction subunit is used to convert the Maxwell equations used to describe the electromagnetic field of GIS into a sixth-order matrix form:

[0244]

[0245] In the formula, is the vector consisting of the components of the electric and magnetic fields in the rectangular coordinate system, ,in:

[0246]

[0247]

[0248]

[0249]

[0250]

[0251]

[0252] in, is a sixth-order matrix, expressed as:

[0253]

[0254] In the formula, is the dielectric constant, is the magnetic permeability.

[0255] It should be noted that this embodiment expresses the Maxwell equations in matrix form so that the electromagnetic field components are expressed in the form of vectors, and the differential operations in the equations are described by matrix operations. This form provides a more unified and compact description method, making the equations easier to understand and derive, and in the numerical solution process, numerical methods usually involve solving discretized equations, and the matrix form makes these solution methods more direct and effective.

[0256] The matrix conversion subunit is used to convert the sixth-order matrix form into a fourth-order rectangular form:

[0257]

[0258] In the formula, and The representation forms are:

[0259]

[0260]

[0261] The solving subunit is used to solve the fourth-order matrix form using the four-step HIE-FDTD algorithm and calculate the field strength data of each grid node.

[0262] This embodiment can significantly improve the accuracy and stability of the numerical solution by introducing a higher-order time integration scheme and a space integration scheme. The traditional FDTD method may be affected by numerical dissipation and numerical dispersion, and the present invention can reduce these problems through a more accurate integration method, which is particularly effective when dealing with long-term simulations or high-frequency electromagnetic wave propagation. For electromagnetic field simulations of complex structures, such as those with media, inhomogeneous media or inhomogeneous structures, this embodiment uses a four-step hybrid implicit explicit finite-difference time-domain (HIE-FDTD) method to better handle these complexities. Its accurate integration and numerical methods can effectively simulate electromagnetic field propagation and interactions between different media.

[0263] As a further preferred technical solution, the solving subunit is used to perform the following steps:

[0264] 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;

[0265] The pursuit method is used to solve the three diagonal implicit equations and obtain the grid node field strength data.

[0266] Specifically, the four-step HIE-FDTD algorithm is decomposed in the time domain into , , , Four steps, including:

[0267] (1) :

[0268]

[0269] (2) :

[0270]

[0271] (3) :

[0272]

[0273] (4) Step by step:

[0274]

[0275] for Step by step, using a semi-implicit difference scheme to introduce auxiliary vectors , keeping the stability condition of the algorithm unchanged, we can get:

[0276]

[0277] in and is the mutual coupling equation, and the substitution method is used to eliminate The tridiagonal implicit equations are as follows:

[0278]

[0279] In the formula, is the point loss parameter of the medium, , , is the time step, represents the electric field components in the x, y, and z directions, The magnetic field components in the x, y, and z directions. The superscripts indicate the time step, e.g. Represents the time step n E x , represents the electric field Ex at the 1 / 4 position between time steps n and n+1, is the sixth-order identity matrix.

[0280] for , the auxiliary vector Substitute the expression into the expansion and get:

[0281]

[0282] for , the auxiliary vector Substitute the expression into the expansion and get:

[0283]

[0284] for , the auxiliary vector Substitute the expression into the expansion and get:

[0285]

[0286] According to the above equations, the components of the electric field and magnetic field in the x, y, and z directions at each step can be solved.

[0287] This embodiment can significantly improve the accuracy and stability of the numerical solution by introducing a higher-order time integration scheme and a space integration scheme. The traditional FDTD method may be affected by numerical dissipation and numerical dispersion, and this embodiment can reduce these problems through a more accurate integration method, which is particularly effective when dealing with long-term simulations or high-frequency electromagnetic wave propagation. For electromagnetic field simulations of complex structures, such as those with media, inhomogeneous media or inhomogeneous structures, this embodiment can better handle these complexities. Its accurate integration and numerical methods can effectively simulate electromagnetic field propagation and interaction between different media.

[0288] It should be noted that the present embodiment can obtain simulation calculation results through calculation, and the result table shows the spatial coordinates corresponding to each grid node and the distribution data of the electromagnetic field at that point. Since the coarseness of the simulation grid division is proportional to the calculation accuracy and inversely proportional to the calculation time, in order to solve the contradiction between accuracy and time, a real-time enhanced simulation method for grid data is designed, and grid enhancement is implemented for the coarse grid simulation electromagnetic field distribution results based on graph machine learning, and the result of finer grid simulation is also achieved. Furthermore, the step S40: using the node interpolation enhancement strategy to generate new grid nodes and feature values ​​of the new grid nodes between adjacent grid nodes is specifically used for:

[0289] Synthetic nodes are created by combining existing nodes through Node Interpolation, which is a generative enhancement strategy that generates new samples by interpolating between samples of the minority class and their nearest neighbors. The training distribution is as follows:

[0290]

[0291]

[0292]

[0293] In the formula, and are two adjacent grid nodes and both come from , is the new grid node, , that is, by the point and Point Generate a new node between , the new node position is given by The value of is determined.

[0294] As a further preferred technical solution, the field strength data of the new grid node is:

[0295]

[0296]

[0297] In the formula, and is the field strength data of two adjacent grid nodes, is the field strength data of the new grid node, is an undetermined coefficient. Δ is determined by λ and the field strength attribute relationship between the two nodes. The specific value needs to be confirmed in combination with the corresponding field strength, because the gradient of distance change is different from that of field strength change.

[0298] In this way, new nodes can be generated between adjacent nodes. The new node eigenvalues ​​also contain spatial data and field strength data, thereby achieving grid enhancement of the coarse grid simulation electromagnetic field distribution results and achieving the same results as the finer grid simulation results.

[0299] It should be noted that this embodiment involves spatial data and electromagnetic field data of existing nodes through an algorithm based on node proliferation between adjacent nodes. For grid data in GIS simulation, it has obvious topological structure characteristics. This embodiment can effectively capture the relationship between grids and retain the topological structure information of grid data. In addition, the algorithm of this embodiment generates the representation of new nodes through the characteristics of adjacent nodes, so that the algorithm can take into account the local characteristics of the nodes. In addition, this embodiment is an algorithm enhancement based on a graph structure for the coarse grid data obtained by simulation, which can adapt to grid data of different scales, including small-scale and large-scale grids, and is adaptable to GIS simulation data under different voltage levels of different devices, thereby realizing GIS simulation data enhancement and improving the quality of grid data.

[0300] Embodiment 3

[0301] Based on the contents disclosed in the above embodiment 1, Figure 7 As shown, the process of realizing GIS electromagnetic field distribution simulation enhancement by the simulator in the first embodiment through finite difference time domain is described in detail, and the simulator includes:

[0302] A simulation module, used for using the coarse positioning result of the local discharge source obtained based on the full time domain waveform diagram as the initial injection point of the three-dimensional simulation model of the GIS equipment to simulate the local discharge phenomenon of the GIS equipment, and performing simulation calculation to obtain a coarse grid simulation result;

[0303] The enhancement module is used to enhance the coarse grid simulation result by using an enhanced model to obtain simulation enhanced data.

[0304] Among them, the simulation module is specifically used to use the coarse positioning result of the local discharge source obtained based on the full time domain waveform diagram as the initial injection point of the three-dimensional simulation model of the GIS equipment to simulate the local discharge phenomenon of the GIS equipment, and use the time domain finite element method to perform simulation calculations to obtain coarse grid simulation results. Specifically, according to the structural parameters of the GIS equipment, a simulation model of three-dimensional data can be established using simulation software such as COMSOL Multiphysics, and the three-dimensional data model is discretized into multiple tetrahedral units through the finite element method, each of which includes a number of nodes, and simulation calculations are performed to obtain the coarse grid simulation data that needs to be processed.

[0305] The enhancement module is specifically used to perform simulation enhancement on the coarse grid simulation data based on the GraphSAGE algorithm, including:

[0306] Network abstraction unit, used to abstract the simulation model of GIS equipment into a network topology structure , 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;

[0307] The primary aggregation unit is used to use the simulation feature data of the node as the initial benchmark feature, and use the GraphSAGE algorithm to perform a primary aggregation process on the neighbor nodes of each node in each slice unit to generate the initial benchmark feature of the newly added node;

[0308] The secondary aggregation unit is used to perform secondary aggregation processing on the neighbor nodes of the newly added node using the GraphSAGE algorithm to generate a feature vector representation of the newly added node.

[0309] It should be noted that, for the newly added nodes generated by the overlapping faces of any two adjacent free tetrahedrons, each node in the sheet unit composed of the two free tetrahedrons can be used as the neighbor nodes of the newly added node. The GraphSAGE algorithm is used 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, that is, the three-dimensional spatial coordinate data and electric field strength of the newly added node are obtained, thereby generating more new node data based on the original grid node data, completing the enhanced refinement of the grid node data generated by the simulation calculation.

[0310] In detail, in the GraphSAGE algorithm, random sampling of neighbor nodes is usually used to update the representation of the target node. However, in GIS simulation, the selection of neighbor nodes needs to consider more factors, such as the physical distance between nodes, the similarity of nodes, etc. Therefore, this embodiment regards two free tetrahedral units as a sheet unit, and aggregates the neighbors of the newly added node in a sheet unit, so that the similarity and correlation between nodes are higher. At the same time, considering that there are many types of node features in GIS simulation data, the spatial coordinates and electric field strength of the node are used as node features, which is more in line with the data connection of GIS simulation. In addition, this embodiment adjusts the depth of the model of the GraphSAGE algorithm, that is, the number of aggregation layers, according to the complexity and scale of GIS simulation data, and improves the algorithm's ability to express data in order to obtain better performance.

[0311] As a further preferred technical solution, the primary polymerization unit is used to perform the following steps:

[0312] (1) Using the simulation feature data of each node as the initial benchmark feature of each node;

[0313] (2) 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 of , 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 ;

[0314] Specifically, each node in the network topology ( ) is the characteristic vector of the simulation feature data as the reference feature corresponding to each node For each slice unit, traverse each node in the slice unit in the first aggregation ( ), for the node , first get its neighbor set ,node The feature vector of .

[0315] Specifically, the nodes are aggregated using the aggregation function through the aggregator Corresponding to the initial baseline features of all nodes in the neighbor set, an aggregated neighbor feature is obtained for:

[0316]

[0317] In the formula, Represents an aggregate function.

[0318] (3) Connect the initial baseline 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 ;

[0319] Specifically, this embodiment uses the feature of each node after a traversal As the new feature vector representation of the node, and as the benchmark feature of each node in the secondary traversal process, where:

[0320]

[0321] In the formula, is the concatenation function, is the first weight matrix, yes function.

[0322] Furthermore, in this embodiment, the aggregation function uses the average aggregation function MEAN ,Right now .

[0323] (4) Use the aggregation function to aggregate the initial baseline features of the common nodes of the two free tetrahedrons in each sheet element to obtain the initial baseline features of the newly added nodes.

[0324] Specifically, since the newly added node generated by the common face of two adjacent free tetrahedrons has no initial reference feature, this embodiment aggregates the initial reference features of the common nodes of the two adjacent free tetrahedrons using an aggregation function to obtain the initial reference feature of the newly added node P. , the formula is:

[0325]

[0326] In the formula, is the initial datum feature of the common node of two adjacent free tetrahedrons,

[0327] Furthermore, the aggregation function in this embodiment adopts the average aggregation function, so .

[0328] As a further preferred technical solution, the secondary polymerization unit is specifically used to perform the following steps:

[0329] (1) The node features after a traversal As a new benchmark feature for the corresponding node;

[0330] (2) For 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 ;

[0331] Specifically, this embodiment converts all nodes in the slice unit into The neighbor set of the newly added node generated by the slice unit , through the second aggregator, the aggregation function is used to aggregate the new baseline features of all nodes in the neighbor set, that is, ( ), and obtain the secondary aggregated neighbor features , the formula is:

[0332]

[0333] In the formula, is an aggregate function, is the new reference feature for each node in the sheet element.

[0334] (3) Connecting the initial baseline feature of the newly added node and the secondary aggregated neighbor feature to obtain a second joint feature, performing a matrix multiplication operation on the second weight matrix and the second joint feature, and then performing a nonlinear transformation to obtain a feature vector representation of the newly added node.

[0335] Specifically, the feature vector of the newly added node obtained in this embodiment is expressed as:

[0336]

[0337] In the formula, is the concatenation function, is the second weight matrix, yes function.

[0338] Furthermore, since there is no natural order for the neighbors of a node, the aggregator in this embodiment Using the average aggregation function, we have Thus, a new node feature vector will be generated between any sheet unit, that is, the three-dimensional space coordinates and electric field strength of the new node, thereby enhancing the grid node data of GIS finite element method simulation.

[0339] As a further preferred technical solution, the complexity of the GraphSAGE algorithm used in this embodiment is:

[0340]

[0341] in, It is the number of aggregators, the number of weight matrices, and the number of layers in the network. Aggregators , used to aggregate node neighbor information, and Weight Matrix , used to propagate information between different layers. In order to achieve better sampling aggregation effect, , and perform two polymerizations.

[0342] As a further preferred technical solution, in this embodiment, the sampling of nodes is of fixed length. Indicates The number of sampling neighbors of the layer is defined in advance. , repeated sampling or negative sampling method with replacement is used to ensure that the complexity becomes stable. When the number of defined sampling neighbors is greater than the actual number of neighbors, repeated sampling is used to make up the actual number of sampling neighbors. ; When the number of defined sampled neighbors is less than the actual number of neighbors, negative sampling is used to reduce the actual number of sampled neighbors to .

[0343] Embodiment 4

[0344] Based on the content disclosed in the above embodiment 1, this embodiment specifically uses a differential deep learning network model to enhance the coarse grid simulation results to obtain simulation enhanced data. The specific working principle of the differential deep learning network model is as follows: Figure 8 As shown: the differential deep learning network model includes an enhancement network and a structural similarity network connected in sequence, the enhancement network includes a self-adjusting module and a differential convolution module, and the coarse grid simulation result includes coarse grid structure data and coarse grid field strength data; 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 enhancement grid structure characteristics and fine grid enhancement field strength characteristics; the structural similarity network is used to match the fine grid enhancement grid structure characteristics and the fine grid enhancement field strength characteristics to calculate the similarity characteristics; based on the similarity characteristics, the simulation enhancement data is calculated.

[0345] This embodiment designs and constructs a differential deep learning network model, in which convolution combined with a differential algorithm is used to process coarse grid field strength data, which can accurately simulate the electromagnetic field distribution of GIS, especially when dealing with complex geometric structures and boundary conditions; and uses a deep learning network to optimize the simulation process, improve the simulation accuracy, and accelerate the simulation process, effectively shortening the design and evaluation cycle, thereby taking into account both the accuracy and efficiency of GIS electromagnetic simulation.

[0346] As a further preferred technical solution, Fig. 9 As shown, the self-adjusting module includes a first convolutional layer Conv1 and a second convolutional layer Conv2 connected in sequence, and the outputs of the first convolutional layer Conv1 and the second convolutional layer Conv2 are connected through a first addition operation and then output to the activation function layer;

[0347] The coarse grid structure data is used as the input of the first convolution layer Conv1, the grid size supplementary information of the coarse grid structure data is used as the input of the second convolution layer Conv2, the deviation 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 connected via residual Output to the first adding operation.

[0348] It should be noted that this embodiment realizes the refined enhancement of the coarse grid structure data by designing a self-adjusting module. The self-adjusting module realizes the refined enhancement of the electromagnetic field coarse grid data of the GIS equipment by combining convolution operations, residual connections and sigmoid activation functions. This method not only improves the detail expression of the grid data, but also can adaptively adjust the enhancement strategy according to the original grid data through the self-adjusting mechanism, thereby ensuring the accuracy and adaptability of the enhanced grid structure.

[0349] As a further preferred technical solution, the function of the self-adjusting module is to perform fine enhancement of the grid structure. GIS equipment electromagnetic field coarse grid data Input the self-adjusting module in the enhancement module to obtain the fine grid enhancement grid structure characteristics , the formula is:

[0350]

[0351] In the formula, Enhance the mesh structure features for fine meshes, represents the grid-enhanced convolution operation, represents the fine-tuning convolution operation, is the coarse grid structure data, Supplementary information for the grid size of the coarse grid structure data, is the deviation vector of the coarse grid structure data, is the residual link, is the activation function.

[0352] As a further preferred technical solution, Fig. 9 As shown, the differential convolution module includes differential convolution layers connected in sequence , Size Integration Layer , self-attention mechanism layer And the third convolutional layer Conv3, the third convolutional layer Conv3 is connected with Activation function;

[0353] The coarse grid field strength data is used as the differential convolution layer The input of the differential convolution layer Used to calculate field intensity characteristics of different sizes of the coarse grid field intensity data by using a differential algorithm;

[0354] The size integration layer Used to sum the field intensity features of different sizes calculated by the differential convolution layer to obtain a reshaped field intensity feature;

[0355] The reshaped field strength features are passed through the self-attention mechanism layer , after the third convolutional layer Conv3, the fine grid enhanced field strength features are output.

[0356] It should be noted that this embodiment designs a differential convolution module for accurate extraction of field intensity data. The differential convolution module can accurately extract electromagnetic field intensity features through specially designed differential convolution operations, effectively simulate the physical distribution of the electromagnetic field, and combine the self-attention mechanism. This module can highlight important field intensity features and improve the extraction accuracy of field intensity data and the attention concentration of the model. And by extracting and summing features with convolution kernels of different sizes, it is possible to capture the multi-scale features of electromagnetic field data, which is particularly important for understanding and simulating the complex changes of electromagnetic fields. This method improves the generalization ability of the model, enabling it to better adapt to changes in electromagnetic fields of different scales.

[0357] As a further preferred technical solution, 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.

[0358] It should be noted that those skilled in the art may also select other differential algorithms to be combined with the deep learning network according to actual application requirements, and this embodiment does not specifically limit this.

[0359] Further, taking the five-point difference as an example, Fig.10 As shown, the differential convolution layer The design is as follows:

[0360] Design the basic convolution kernel based on the five-point difference algorithm :

[0361]

[0362] Considering the data enhancement for 3D GIS electromagnetic field simulation, the dimensionality transformation kernel Get the three-dimensional difference convolution kernel :

[0363]

[0364]

[0365] Among them, ⊙ represents the corresponding multiplication of elements.

[0366] Using 3D differential convolution kernel Extraction of coarse grid field intensity gradient information (extracting coarse grid electromagnetic field gradient changes to guide fine grid data generation):

[0367]

[0368] Among them, ∗ represents the convolution operation, For application The gradient information data obtained later;

[0369] The coarse grid field intensity data is enhanced by upsampling to obtain the upsampled fine grid field intensity data U (obtaining refined coarse grid data):

[0370]

[0371] in, The upsampling convolution kernel is obtained with network training;

[0372] Combined with the coarse grid field intensity gradient information G, preliminary fine grid field intensity data is obtained

[0373] .

[0374] This embodiment directly uses the form of the differential algorithm to define the convolution kernel, ensuring that the physical principles can be accurately mapped to the convolution operation. This is achieved by taking the core calculation formula of the differential algorithm and converting it into the form of the convolution kernel, thereby ensuring the correct application of continuous physical laws to discrete data. In addition, this embodiment takes scale transformation into account when 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 to capture multi-scale features.

[0375] By extending the two-dimensional convolution kernel to three-dimensional space through a certain transformation, the key step of applying the traditional differential method to the three-dimensional GIS electromagnetic field simulation is that this dimensional transformation takes into account the physical effects in each direction in the three-dimensional space. The rotational invariance problem can be solved by designing a symmetrical three-dimensional convolution kernel. A symmetrical kernel will produce a consistent response in all directions, which can solve the isotropy problem in electromagnetic field simulation.

[0376] It should be noted that the special differential convolution operation designed in this implementation is based on a method for solving numerical partial differential equations, such as the five-point difference method, which is often used to approximate the solution of partial differential equations in numerical analysis. In deep learning, traditional convolution layers are usually used to extract features of input data, while special differential convolution kernels can be designed to be more suitable for the simulation of specific physical phenomena, such as electromagnetic field simulation. The design of such convolution kernels is usually based on the discrete approximation of the second-order derivative of the function space in the differential algorithm, so that the network can capture the physical characteristics of the data while learning the data.

[0377] Compared with traditional convolutional layers, differential convolution can maintain high-quality data while ensuring that the data conforms to specific physical laws, making the data generated by the network more physically credible. This combination makes the model perform better when dealing with problems that require precise physical modeling, such as electromagnetic field simulation and climate change simulation.

[0378] Furthermore, the function of the differential convolution module set in this embodiment is to obtain accurate fine grid field strength data, and use the differential convolution module to Coarse grid field strength data of electromagnetic field of GIS equipment Processing is performed to obtain the fine grid enhanced field strength characteristics , the formula is:

[0379]

[0380]

[0381]

[0382] In the formula, is the field intensity feature generated by differential convolution, is the differential convolution operation, is the coarse grid field strength data, is the field strength characteristic after reorganization, Indicates the application of convolution kernels of different sizes. Enhance the field strength characteristics for the fine grid, For the self-attention mechanism operation, is the convolution operation, is the activation function.

[0383] Furthermore, this embodiment optimizes the computational efficiency and accuracy of the simulation by combining the self-adjusting module and the differential convolution module. The self-adjusting module reduces unnecessary calculations through intelligent enhancement strategies, while the differential convolution module ensures high-precision extraction of field strength data. The two work together to achieve 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, precise feature extraction and efficient calculation methods.

[0384] As a further preferred technical solution, Fig.11 As shown, the structural similarity network includes a first branch network, a second branch network, a second addition operation and a first multilayer perceptron, the outputs of the first branch network and the second branch network are both connected to the second addition operation, and the output of the second addition operation is connected to the first multilayer perceptron , the multilayer perceptron Followed by an activation function .

[0385] Specifically, the first branch network includes a convolutional neural network layer CNN and a batch normalization operation connected in sequence. , the regularization operation Then there is an activation function ;

[0386] The second branch network includes a second multilayer perceptron , the second multi-layer perceptron Then there is an activation function .

[0387] Furthermore, the function of the structural similarity network is to match the enhanced grid data with the field strength data for the first The fine grid structure characteristics of the electromagnetic field of GIS equipment with coarse grid number Enhanced field strength characteristics with fine grids The field strength and structural characteristics of:

[0388]

[0389]

[0390]

[0391]

[0392] In the formula, For the Extraction information of fine grid structure features corresponding to the coarse grid number of electromagnetic field of GIS equipment; For the Extraction information of the fine grid-enhanced field strength characteristics corresponding to the coarse grid number of the electromagnetic field of the GIS equipment; and Respectively represent the fine grid enhanced grid structure feature extraction information Enhanced field strength feature extraction information with fine grid The modulation weight of It is the comprehensive feature of information extraction of fine grid enhanced grid structure feature and fine grid enhanced field strength feature; For similarity characteristics; is the batch normalization operation, is the convolution operation, is a multi-layer perceptron of the structural similarity module, It is the combined characteristic of field strength and structure.

[0393] It should be noted that the functions of the structural similarity network designed in this embodiment are: (1) Preserving physical structure: The structural similarity module ensures that the enhanced data retains its physical accuracy while maintaining structural integrity and coherence by focusing on the structural properties of the data. This is particularly important in the field of simulation because the physical behavior of the electromagnetic field strongly depends on its spatial structural characteristics. (2) Improving the quality of data enhancement: This module helps to generate higher quality data, especially in restoration and super-resolution tasks, and can ensure that the details of the generated data have a high structural similarity with the original coarse grid data. (3) Enhanced feature representation: The structural similarity module provides an effective mechanism to enhance and significantly characterize key features in electromagnetic field data, especially in terms of field strength and grid structure, by fusing features extracted by CNN and MLP.

[0394] As a further preferred technical solution, the calculation of GIS fine grid enhanced data based on the similarity feature specifically includes the following steps:

[0395] (1) 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;

[0396] It should be noted that this embodiment specifically uses a field strength consistency module to process the coarse grid structure data, the coarse grid field strength data, the fine grid enhanced grid structure features, the fine grid enhanced field strength features and the similarity features, so as to converge the difference between the fine grid enhanced data and the coarse grid data.

[0397] Use the following formula to calculate the Electromagnetic field fine grid enhancement grid structure characteristics of strip GIS equipment and fine grid enhanced field strength characteristics With GIS equipment electromagnetic field coarse grid data and coarse grid field strength data Fine grid enhanced data after comprehensive compensation :

[0398]

[0399]

[0400]

[0401]

[0402] In the formula, , , and They are regularized electromagnetic field fine grid enhanced grid structure characteristics, regularized fine grid enhanced field strength characteristics, regularized electromagnetic field coarse grid data and regularized coarse grid field strength data; is the matrix 2 norm.

[0403] (2) calculating the grid structure consistency based on the regularized coarse grid structure data and the regularized fine grid enhanced grid structure characteristics;

[0404] Specifically, the calculation formula for grid structure consistency is:

[0405]

[0406] In the formula, The grid structure is consistent.

[0407] (3) calculating the grid field strength consistency based on the regularized coarse grid field strength data and the regularized fine grid enhanced field strength characteristics;

[0408] Specifically, the calculation formula for grid field strength consistency is:

[0409]

[0410] In the formula, is the grid field strength consistency.

[0411] (4) calculating grid field strength loss compensation based on the grid structure consistency and the grid field strength consistency;

[0412] Specifically, the calculation formula for grid field strength loss compensation is:

[0413]

[0414] In the formula, Compensate for grid field strength loss.

[0415] (5) 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;

[0416] Specifically, the calculation formula for fine grid enhanced structure data is:

[0417]

[0418] In the formula, Enhance structural data for fine meshes.

[0419] (6) calculating fine grid enhanced field strength data based on the fine grid enhanced field strength characteristics, the grid field strength loss compensation and the similarity characteristics;

[0420] Specifically, the calculation formula for fine grid enhanced field strength data is:

[0421]

[0422] In the formula, Enhance field strength data for fine grids; and is the scale factor.

[0423] As a further preferred technical solution, the differential deep learning network model used in this embodiment is a pre-trained model that can be used to calculate the similarity between the grid structure and the grid field strength. The training process includes the following steps:

[0424] A GIS electromagnetic field coarse grid historical simulation data set is used, and the differential deep learning network model is trained using the historical simulation data set until the total loss function of the model is minimized, wherein the total loss function includes a constraint loss function based on Maxwell's equations and a field strength and structural consistency loss function, and the formula is expressed as:

[0425]

[0426] In the formula, is the total loss function, is the field strength and structure consistency loss function, is the Maxwell constraint loss function, and is the weight parameter.

[0427] It should be noted that this embodiment has carried out a multi-dimensional loss function design. By combining structural similarity loss, field strength consistency loss and physical constraint loss, multiple important characteristics of simulation data are comprehensively considered. The design of this multi-dimensional loss function is beneficial to ensure the accuracy of the simulation while being able to adjust the weights between different losses to adapt to specific simulation needs and optimization goals.

[0428] Furthermore, the formula of the field strength and structure consistency loss function is expressed as:

[0429]

[0430]

[0431]

[0432] In the formula, represents the gradient obtained by applying the difference method, is the structural similarity loss, is the field strength consistency loss, and is the scale factor, Enhance the mesh structure features for fine meshes, is the coarse grid structure data, To enhance the field strength characteristics for fine grids, is the coarse grid field strength data, is the 2-norm of the vector.

[0433] It should be noted that this embodiment ensures that the enhanced electromagnetic field data is highly consistent with the original data in structure and field strength through structural similarity loss and field strength consistency loss. This consistency not only promotes the convergence of simulation data, but also ensures the physical authenticity of simulation results and improves the reliability of simulation.

[0434] And by introducing the scaling factor, a flexible loss function adjustment mechanism is provided. This mechanism can not only adjust the importance of different parts of the loss function according to specific tasks, but also improve the generalization ability and adaptability of the model when processing different types of electromagnetic field data.

[0435] Furthermore, the formula of the constraint loss function based on Maxwell equations is expressed as:

[0436]

[0437]

[0438]

[0439]

[0440]

[0441]

[0442] In the formula, is the magnetic field component, is the Gaussian law loss, is Gauss's magnetic law loss, is the loss due to Faraday's law of electromagnetic induction, is the Ampere's law loss, The proportionality constant, , , , is the scale factor, represents the rotation, is the vector 2 norm; is the magnetic permeability of vacuum, is the permittivity of vacuum, It is the fine grid enhanced field strength data.

[0443] It should be noted that the physical constraint loss function ensures that the enhanced electromagnetic field data strictly follows the basic laws of electromagnetism through the key components of Maxwell's equations - Gauss's law, Gauss's law of magnetism, Faraday's law of electromagnetic induction, and Ampere's law. This constraint makes the electromagnetic field data output by the model not only numerically accurate, but also strict in physical laws, which increases the scientificity and practicality of the model.

[0444] It should be noted that when the differential method is combined with the generative adversarial network to form a differentially enhanced generative adversarial network, the differential convolution kernel is directly integrated into the generator architecture, thereby introducing differential enhancement of spatial features. The adversarial here is mainly reflected in the loss of the model, that is, when using the generative adversarial network, the total loss function of the model includes not only the physical constraint loss function and the field strength and structural consistency loss function, but also the adversarial loss part.

[0445] In summary, the loss function designed in this embodiment combines the advantages of traditional physical models and deep learning. Through precise mathematical expressions and physical constraints, it helps to optimize the stability and convergence speed of the algorithm and improve training efficiency. The total loss function combines electromagnetic field theory and deep learning technology, which not only enhances the model's ability to process electromagnetic field simulation data, but also ensures the accuracy and physical authenticity of the simulation results, providing an effective optimization strategy for electromagnetic field simulation.

[0446] Embodiment 5

[0447] Based on the content disclosed in the above embodiment 1, this embodiment specifically adopts a generative adversarial network model to enhance the coarse grid simulation results to obtain enhanced fine grid data. Fig.12 As shown:

[0448] The enhancement model adopts a generative adversarial network model, which 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.

[0449] The discriminator includes a structural similarity module, a field strength consistency module and an accuracy discrimination module, wherein:

[0450] 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;

[0451] The field intensity consistency module is used to calculate a second confidence value of the field intensity information contained in the enhanced fine grid data and the real fine grid data obtained by simulation;

[0452] 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.

[0453] It should be noted that in this embodiment, the real coarse grid data and the real fine grid data are input into the generative adversarial network for training in advance until the overall optimization loss function is minimized, and the trained generator is used as a data enhancement model to enhance the real coarse grid data generated in real time, wherein the generator network is responsible for learning the potential distribution characteristics of the real samples and synthesizing new samples; the role of the discriminator network is to distinguish between the real samples and the generated new samples. The generator and the discriminator achieve the adversarial effect through alternating training, continuously improving their respective generation and discrimination capabilities, and finally making the adversarial generator network and the discriminator network reach a Nash equilibrium.

[0454] Among them, the discriminator designed in this embodiment is used to determine whether the input data meets the requirements. The discriminator consists of a structural similarity module, a field strength consistency module and an accuracy discrimination module. The grid data is divided into structural data and field strength data for comparative analysis, which can effectively improve the discrimination accuracy.

[0455] Since GIS equipment has a certain physical structure, the data points of electromagnetic field simulation cannot exceed a certain range. The PU-GAN (Projection Upsampling Generative Adversarial Network) designed in the embodiment of the present invention multiplies the coarse grid data points by upsampling. The generator adopts cyclic multiplication enhancement to enhance the electromagnetic field between coarse grid data points, which ensures that it does not exceed the physical structure of GIS to a certain extent, thereby improving the accuracy of GIS electromagnetic field simulation data enhancement.

[0456] As a further preferred technical solution, Fig.13 As shown, the generator includes several cascaded feature extraction and expansion networks, each of which is connected to a multi-layer perceptron;

[0457] The feature extraction and expansion network includes a feature extraction module Featuer Extraction and a feature expansion module connected in sequence, and the output of the feature extraction module Featuer Extraction in the feature extraction and expansion network of the previous level is connected to the output of the feature extraction module Featuer Extraction in the feature extraction and expansion network of the next level;

[0458] like Fig.14 As shown, the feature extraction module Featur Extraction includes a first MLPs layer, a KNN layer, a second MLPs layer, a maximum pooling layer maxpooling and a data compression layer compression which are 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 compression, and the output of the data compression layer compression is connected to the feature expansion module;

[0459] 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.

[0460] It should be noted that the function of the generator designed in this embodiment is to enhance the GIS coarse grid data and generate the required GIS fine grid data; the designed generator adopts a cyclic multiple enhancement design, and can autonomously adjust the coarse grid enhancement multiple according to demand, and the multiple enhancement can better retain the characteristics of the coarse grid data.

[0461] Specifically, the working process of the feature extraction module in this embodiment is as follows: first, the three-dimensional space coordinates of each sampling point are input and field strength values The N×4-dimensional information composed of the information is extracted as a fixed-scale feature C' through the first MLPs layer, and then the features are grouped into N×K groups using the KNN layer, and the group features are further optimized using the second MLPs layer to obtain the features of the G channel. The features output by the second MLPs layer are combined with the features output by the first MLPs layer to obtain a new G channel feature (a total of N×K×(2G+C')). Finally, the group information is condensed by maxpooling to obtain the final N×(2G+C'), and the N×4-dimensional features of the sampling points initially output are added. This feature will be fed into the subsequent modules as a module multi-scale feature aggregation to improve the reconstruction accuracy while also improving the parameter efficiency and reducing the model size. Under the compression of the last layer of data compression layer compression, N×C features are obtained. After the obtained N×C features are copied, a vector encoded as 1 or -1 is added after each feature to generate position perturbations, and then the subsequent multi-layer perceptron MLP is used to regress the 2N×(C+1) features into 2N×4 data.

[0462] It should be noted that after multiple cycles of feature extraction and expansion, a 2 r The N×C features are finally regressed to four-dimensional data by a multi-layer perceptron MLP. .

[0463] It should be noted that The value of depends on the accuracy difference between the input simulation coarse grid and fine grid, which is not specifically limited in this embodiment.

[0464] Specifically, different modules target data accuracy. If the previous feature needs to be passed to the next layer, good interpolation technology is required to match the features of different levels. In this embodiment, when copying the features output by the feature extraction module in the feature extraction and expansion network of the previous level 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 the features of different levels.

[0465] As a further preferred technical solution, this embodiment uses the bilateral interpolation technology to interpolate the points using the features around the target point. The interpolation of the feature values ​​output by the previous feature extraction module is:

[0466]

[0467] in, It is The coordinates and field strength of each point, It is The coordinates and field strength of the points, It is The eigenvalues ​​corresponding to the points are It is The eigenvalues ​​corresponding to the points are is the set of data points that need to be inserted, where the joint weighted function is expressed as follows:

[0468]

[0469]

[0470] Among them, the width parameter and is the average distance from a point to its nearest neighbor, Represents the magnitude of a vector.

[0471] This embodiment is for coarse grid data enhancement. The use of bilateral interpolation can effectively avoid the overlap of generated grid data with source grid data, thus avoiding waste.

[0472] As a further preferred technical solution, Fig.15 As shown, the structural similarity module includes a first structural branch network, a second structural branch network and a first regression network, the enhanced structural information contained in the fine grid generation 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;

[0473] 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.

[0474] Specifically, Fig.15 As shown, the first structure branch network and the second structure branch network both include a multi-layer perceptron and a maximum pooling layer connected in sequence, and the output of the first structure branch network and the second structure branch network is subjected to a matrix multiplication operation to output a structural feature matrix:

[0475]

[0476]

[0477]

[0478] In the formula, is the structural feature matrix, The structural characteristics of the data for the fine grid generation, is the structural feature of the real fine grid data, , respectively , The modulation weight, represents the multi-layer perceptron operation, represents the maximum pooling operation, Generate data for the fine grid The structural information of each point, is the real fine grid data The structural information of each point.

[0479] Specifically, Fig.15 As shown, the first regression network includes a first self-attention unit, a multi-layer perceptron and a fully connected layer connected in sequence, the first self-attention unit is used 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.

[0480] As a further preferred technical solution, Fig.16 As shown, the field strength consistency module includes a first enhanced branch network, a second enhanced branch network and a second regression network, the enhanced field strength information contained in the fine grid generation 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 a field strength feature matrix after matrix multiplication operation;

[0481] 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.

[0482] Specifically, the first enhanced branch network and the second enhanced branch network both include a multi-layer perceptron and an activation function connected in sequence, and the output of the first enhanced branch network and the second enhanced branch network is subjected to a matrix multiplication operation to output a field intensity feature matrix as follows:

[0483]

[0484]

[0485]

[0486] In the formula, is the field strength characteristic matrix, The field strength characteristics of the data generated for the fine grid, is the field intensity characteristic of the real fine grid data, '、 respectively , The modulation weight, Generate data for the fine grid Field strength information of each point, is the real fine grid data Field strength information of each point.

[0487] Specifically, the second regression network includes a second self-attention unit, a multi-layer perceptron and a fully connected layer connected in sequence, the second self-attention unit is used to generate a field intensity attention weight based on the field intensity feature matrix, and the field intensity attention weight outputs a second confidence value after passing through the multi-layer perceptron and the fully connected layer.

[0488] As a further preferred technical solution, Fig.17 As shown, the self-attention unit used in this embodiment is used to perform enhanced feature integration. Specifically, the input features are converted into G, H and F through three independent MLPs, and then the attention weights are generated by G and H:

[0489] M

[0490] in, Represents the softmax function.

[0491] Then the weighted feature M is obtained, and finally the output feature is generated, which is the sum of the input feature and the weighted feature.

[0492] 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 the accuracy discrimination module, which is used to control the generated grid structure and field strength data and distinguish the accuracy of the generated results. The accuracy discrimination module can be defined as a regional-level fully convolutional binary classifier, which can narrow the distance between the generated fine grid simulation data and the fine grid simulation data distribution of the numerical method, and use the least squares adversarial loss to ensure the stability of the training process.

[0493] The adversarial loss between the fine mesh generated data and the true fine mesh data is defined as follows:

[0494]

[0495] In the formula, and They are respectively the discriminator D generating data from the fine grid output of the generator With real fine grid data Confidence value of the prediction in .

[0496] As a further preferred technical solution, the formula of the overall optimization loss function is expressed as:

[0497]

[0498] In the formula, For the overall optimization loss function, A first confidence value of the structural information contained in the fine grid generation data and the true fine grid data, a second confidence value for the field strength information contained in the fine grid generation data and the real fine grid data, is the adversarial loss between the fine grid generated data and the true fine grid data, and is the weight parameter.

[0499] The overall optimization loss function set in this embodiment comprehensively considers the structure and field strength confidence of the generated fine grid and the real fine grid, and has the advantages of: (1) Promoting the generator to generate realistic data: By optimizing the loss function, the generator is forced to generate samples that are as close to the real data distribution as possible. (2) Improving the accuracy of the discriminator: The discriminator improves its ability to distinguish between real data and generated data by optimizing the loss function. This includes enabling the discriminator to more accurately identify the data generated by the generator, and better distinguishing between generated data and real data. (3) Achieving a balance between the generator and the discriminator: Optimizing the loss function ensures that there is a balance between the generator and the discriminator, avoiding over-optimization of one of them and causing unstable training. This balance helps the generator generate more realistic data while ensuring that the discriminator can effectively distinguish between real data and generated data.

[0500] Embodiment 7

[0501] Based on the contents disclosed in the above-mentioned embodiment 1, the specific implementation of the position searcher in the embodiment 1 is described as follows in this embodiment: the simulation enhancement data includes an enhanced discharge position and an enhanced discharge intensity, and the position searcher includes:

[0502] The state transition model building module uses the Markov decision model to reconstruct the state transition model corresponding to the process of the ant colony algorithm exploring the actual discharge position of the local discharge source with the coarse positioning result of the local discharge source as the center, according to the enhanced discharge position and enhanced discharge intensity at each moment;

[0503] An heuristic space parameterization module generates a heuristic metric using a heuristic learner based on a graph neural network, and converts the state transition model into a location exploration model that is affected by the heuristic metric and requires graph traversal;

[0504] The iterative search module is used to iteratively search the position exploration model using a neural-guided perturbation interleaved local search algorithm to obtain an actual discharge position of a local discharge source.

[0505] like Fig.18 As shown, in this embodiment, the heuristic rules in the ant colony algorithm are automatically designed and enhanced in a data-driven manner to more effectively solve the rapid iterative search for the optimal matching position of the discharge source. The deep ant colony algorithm uses a graph neural network to generate heuristic metrics, reduces the need for expert knowledge, and combines probabilistic local search to achieve better performance to ensure efficient solution.

[0506] As a further preferred technical solution, the state transition model construction module is specifically used for:

[0507] According to the simulated discharge position and simulated discharge intensity at each moment, the state space corresponding to the actual discharge position of the local discharge source is constructed in the simulator :

[0508]

[0509]

[0510] Where: Indicates that in the simulator The local discharge source node is explored at The corresponding status, Indicates The simulated discharge position at each moment, Indicates The simulated discharge intensity at each moment, It represents the initial discharge intensity of the local discharge source obtained based on the full time domain waveform of GIS local discharge. and initial discharge position As the initial state, represents the exploration cycle, ;

[0511] Constructing the action space corresponding to the actual discharge location of the PD source in the simulator :

[0512]

[0513] Where: Indicates The action of the moment Greedy strategy is the standard choice of action ;

[0514] The state transition model corresponding to the process of the reconstructed ant colony algorithm exploring the actual discharge position of the local discharge source with the initial discharge position of the local discharge source as the center is:

[0515]

[0516] Where: Indicates in action The discharge node To the discharge node The probability of transfer, Indicates that in the simulator The local discharge source node is explored at The corresponding state, Represented by the node Transfer to Node The corresponding pheromone concentration, Represented by the node Transfer to Node The corresponding heuristic function is, Represented by the node Transfer to the included Nodes in The corresponding pheromone concentration, Represented by the node Transfer to the included Nodes in The corresponding heuristic function is, and Represent the pheromone concentration and the weight of the heuristic function, Represents the set of non-uniform grid local discharge source nodes that are allowed to be selected at the next moment.

[0517] Furthermore, by the node To Node The reward function for the transfer is , Indicates The measured discharge intensity is obtained by processing the full waveform in the time domain collected by the sensor at each moment. Indicates The partial discharge intensity obtained by simulation at each moment.

[0518] It should be noted that this embodiment uses the Markov model to formalize and model the state transition process in the ant colony algorithm. This modeling can help understand and analyze the search process of the ant colony algorithm in the solution space, including key steps such as ant path selection and pheromone update.

[0519] As a further preferred technical solution, the heuristic space parameterization module is used to perform the following steps:

[0520] Use multi-layer perceptron to extract the graph neural network The layer is The node and Edge features connecting nodes and mapped to the heuristic metric ;

[0521] The state transition model is transformed into a state transition model that is influenced by the heuristic metric and requires The location exploration model of step graph traversal is:

[0522]

[0523] Where: represents the location exploration model, Indicates that in the simulator The local discharge source node is explored at The corresponding status, Indicates that in the simulator The local discharge source node is explored at The corresponding status, represents the exploration cycle, .

[0524] It should be noted that this embodiment constructs a heuristic learner based on graph neural network, parameterizes the heuristic space for exploring GIS local discharge sources, generates heuristic metrics by using graph neural network, and combines probabilistic local search to achieve better performance to ensure efficient solution.

[0525] Furthermore, the graph neural network The propagation characteristics of the layer are:

[0526]

[0527]

[0528] Where: Indicates Layer The characteristics of the nodes, Indicates Layer The characteristics of the nodes, Indicates Layer The characteristics of the nodes, Indicates The layer is The node and The edge features connecting the nodes are Indicates The layer is The node and The edge features connecting the nodes are , , , as well as Indicates The learnable parameters of the layer, represents the activation function, represents batch normalization processing, Expressing the The neighborhood of each node is aggregated and calculated. Indicates The neighborhood of a node, express function, represents the Hadamard product, .

[0529] It should be noted that when the ant colony algorithm explores the location of local discharge, it constructs a graph by traversing the discharge nodes, and the nodes of the graph neural network correspond to the discharge nodes.

[0530] As a further preferred technical solution, the location searcher further includes a training module for performing the following steps:

[0531] The gradient strategy is used to train the heuristic learner based on the graph neural network. The objective function used in the training process is for:

[0532]

[0533] Where: represents the objective function corresponding to the actual discharge location of the PD source using the neural-guided perturbation interleaved local search algorithm, Indicates balance and Parameters, In the heuristic metric The expected value of the actual discharge position of the local discharge source under the influence of represents the state space corresponding to the exploration of the actual discharge position of the PD source, represents the objective function;

[0534] Among them, the objective function The gradient of , the formula is:

[0535]

[0536] Where: represents the average target value for directly exploring the actual discharge location of the PD source, represents the average target value of the actual discharge location of the PD source explored using the local search algorithm with neural-guided perturbation interleaving, In the heuristic metric Explore the gradient of the actual discharge location of the partial discharge source under the influence;

[0537] When the maximum number of iterations is reached Or the maximum number of iterations has not been reached But the objective function The training ends when Indicates the minimum threshold corresponding to the objective function.

[0538] As a further preferred technical solution, the iterative search module is used to perform the following steps:

[0539] Iteratively searching the location exploration model using a local search algorithm to obtain a local optimal solution;

[0540] Specifically, the local optimal solution The formula is:

[0541]

[0542]

[0543] Where: represents the objective function, represents the local search operator, represents the state space corresponding to the exploration of the actual discharge position of the PD source, Indicates that the local search for the discharge position is performed under the local search operator. Indicates The measured discharge intensity is obtained by processing the full waveform in the time domain collected by the sensor at each moment. Indicates The local discharge intensity obtained by simulation at each moment is represents the exploration cycle, .

[0544] Perform neural-guided perturbations on the local optimal solution obtained by the current iterative search, and obtain the optimal exploration scheme for the actual discharge position of the local discharge source of the current iterative search through the local search interleaved with neural-guided perturbations;

[0545] Specifically, the optimal exploration scheme of the actual discharge position of the local discharge source in the previous iteration is obtained, and the formula is expressed as:

[0546]

[0547] Where: represents the optimal exploration scheme of the actual discharge position of the local discharge source in the current iteration obtained by performing neural-guided perturbation on the local optimal solution, is the local optimal solution; represents the number of perturbation moves, Represents a heuristic metric.

[0548] It should be noted that the optimal exploration scheme for the actual discharge position of the local discharge source in the current iteration obtained by the disturbance is then locally searched to obtain the optimal exploration scheme for the actual discharge position of the local discharge source in the current iteration. Local search to obtain the local optimal solution :

[0549]

[0550] go through The optimal exploration scheme of GIS local discharge sources obtained by local search with neural-guided perturbation interleaving :

[0551]

[0552] Where: Represents a set of parameters or variables that minimizes the function.

[0553] It should be noted that when a local search algorithm is used to iteratively search the position exploration model to obtain a local optimal solution, a local optimal situation may occur. This embodiment perturbs the local optimal solution by utilizing the heuristic metric obtained by graph neural network training. The heuristic metric can help the algorithm quickly locate and jump out of the local optimal solution, guide the search process to approach the global optimal solution, and guide the search process to explore the solution space more effectively.

[0554] After all the local discharge source nodes are selected, the pheromone concentrations between the local discharge source nodes are updated;

[0555] Specifically, the update formula of pheromone concentration is expressed as:

[0556]

[0557] Where: represents the update coefficient, Indicates The pheromone concentration during the exploration cycle, Indicates The pheromone concentration during the exploration cycle, Indicates Exploration cycle to The change in pheromone concentration during an exploration cycle.

[0558] When the iterative search reaches an iterative convergence condition, an optimal exploration scheme for the actual discharge position of the partial discharge source is determined, and matching of the actual discharge position of the partial discharge source is achieved based on the optimal exploration scheme.

[0559] The ant colony algorithm in this embodiment simulates the search process of ants in a graph to find the possible location of the local discharge source. The graph neural network can effectively propagate the information of the local discharge signal through message passing and feature aggregation, and infer and locate the location in combination with the search results of the ant colony algorithm. The parallel computing capability and efficient feature learning of the graph neural network enable it to quickly process large-scale graph data, which is suitable for real-time or near-real-time local discharge location tasks. As a heuristic optimization method, the ant colony algorithm, combined with the feature learning ability of the graph neural network, can intelligently guide the search process and quickly discover the possible location of the discharge source. This method can effectively avoid falling into the local optimal solution and improve the global search capability, and can achieve fast real-time high-precision positioning.

[0560] It should be noted that if Fig.19 As shown, in this embodiment, the initial discharge position is taken as the center of the circle and the radius of the GIS cross section is taken as the radius of the spherical area for local discharge positioning to determine the optimal exploration scheme, and the actual discharge position of the local discharge source is matched according to the optimal exploration scheme path, so that the local discharge source position can be quickly positioned.

[0561] Embodiment 8

[0562] Based on the contents disclosed in the above embodiment 1, this embodiment describes the warning device mentioned in the embodiment 1 in detail as follows:

[0563] The early warning device is equipped with a fault classification module, which includes:

[0564] A deep feature extraction network is used to obtain deep features corresponding to the full time domain waveform, wherein the deep feature extraction network includes a feature extraction network and an output network, the feature extraction network is formed by superimposing a plurality of feature extraction layers, and each of the feature extraction layers includes a channel attention module and a spatial attention module connected in sequence;

[0565] A reinforcement learning unit, used to construct a data set using the deep features, and train a GIS partial discharge feature matching Markov model using a deep reinforcement learning algorithm to obtain an optimal GIS partial discharge feature matching result;

[0566] The early warning unit generates early warning information based on the optimal GIS local discharge feature matching result and the actual discharge position of the local discharge source to perform fault early warning.

[0567] Specifically, Fig. 20As shown, the deep feature extraction network includes a feature extraction network and an output network. The feature extraction network is formed by superimposing several feature extraction layers. Each of the feature extraction layers includes a channel attention module and a spatial attention module connected in sequence. The present embodiment adopts a dual attention mechanism, that is, the spatial attention is integrated with the channel attention to extract features of the full time domain waveform signal, which can not only improve the discrimination of GIS partial discharge features and reduce redundancy, but also consider features from a spatial perspective. Features can be considered from different perspectives, which is very helpful for the extraction of GIS partial discharge features. At the same time, by superimposing multiple layers, sample information can be fully and effectively utilized. In addition, a GIS partial discharge feature matching Markov model is trained based on a deep reinforcement learning algorithm for extracting GIS partial discharge features to obtain the optimal GIS partial discharge feature matching strategy. When partial discharge occurs in GIS, feature matching of GIS partial discharge signals is achieved to determine the GIS partial discharge type.

[0568] Embodiment 9

[0569] This embodiment proposes an implementation method corresponding to the GIS field-electricity fusion real-time state perception and early warning system in the above embodiment 1, such as Fig.21 As shown, the method comprises the following steps:

[0570] When the electromagnetic field signal inside the GIS is monitored according to the strong correlation characteristics obtained in advance, including the partial discharge signal, a full time domain waveform diagram including the time domain waveform diagram and the frequency domain waveform diagram of the partial discharge signal is obtained;

[0571] Based on the full time domain waveform of GIS partial discharge, the partial discharge source is roughly located, and the simulation enhancement calculation is performed using the rough positioning result as the initial injection point to obtain simulation enhancement data of the partial discharge source;

[0572] Iteratively search the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement result to obtain the actual discharge position of the partial discharge source;

[0573] An early warning is provided based on the full waveform in the time domain and the early warning information including the actual discharge position of the partial discharge source.

[0574] It should be noted that the method described in the present invention is used to implement GIS fault warning using the GIS field-electricity fusion real-time status perception and warning system as described in the above embodiments. Other embodiments can refer to the above embodiments and will not be repeated here.

[0575] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples", or "some examples" means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representation of the above terms does not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner.

[0576] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description of the present invention, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0577] Although the embodiments of the present invention have been shown and described above, it is to be understood that the above embodiments are exemplary and are not to be construed as limitations of the present invention. A person skilled in the art may change, modify, replace and vary the above embodiments within the scope of the present invention.

Claims

1. A GIS field-electricity fusion real-time status perception and early warning system, characterized in that: The system includes a signal synchronous acquisition and measurement embedded device and an industrial computer, wherein a simulator, a location searcher and an early warning device are deployed in the industrial computer; The signal synchronous acquisition and measurement embedded device is used to monitor the electromagnetic field signal inside the GIS according to the pre-screened strong correlation characteristics, and obtain a full time domain waveform diagram including a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal when the signal contains a local discharge signal; The simulator is used to roughly locate the local discharge source based on the full time domain waveform of the GIS local discharge, and use the rough positioning result as the initial injection point to perform simulation enhancement calculation to obtain simulation enhancement data of the local discharge source; The position searcher is used to iteratively search the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement data to obtain the actual discharge position of the partial discharge source; The early warning device is used to issue an early warning based on the full time domain waveform diagram and the early warning information including the actual discharge position of the partial discharge source; The simulator includes a simulation module and an enhancement module. The enhancement module is used to enhance the coarse grid simulation result obtained by simulating the coarse positioning result of the simulation module as the initial injection point to obtain simulation enhanced data; the enhancement module is specifically used to enhance the coarse grid simulation result based on the GraphSAGE algorithm, or to enhance the coarse grid simulation result using a differential deep learning network model, or to enhance the coarse grid simulation result using a generative adversarial network model.

2. The GIS field-electricity fusion real-time status perception and early warning system according to claim 1 is characterized in that: The signal synchronous acquisition and measurement embedded device comprises: A signal acquisition position determination module is used to determine the installation position of each sensor in the signal acquisition module; Signal acquisition module, used to collect electromagnetic field signals inside GIS; A signal processing front end is used to perform signal conditioning on the electromagnetic field signal transmitted by the signal acquisition module, and transmit the conditioned signal to an analog-to-digital converter; An analog-to-digital converter, used to convert the conditioned signal from an analog signal to a digital signal and transmit it to a signal measurement module; The signal measurement module is used to monitor the digital signal based on the strong correlation features obtained in advance, and when it is determined that a local discharge signal is monitored, a full time domain waveform diagram including a time domain waveform diagram and a frequency domain waveform diagram of the local discharge signal is obtained, wherein the strong correlation features include time domain strong correlation features and frequency domain strong correlation features.

3. The GIS field-electricity fusion real-time status perception and early warning system according to claim 2 is characterized in that: The signal synchronous acquisition and measurement embedded device also includes a feature screening module, which is used to use the time-frequency domain features used to detect local discharge signals as nodes of the graph neural network to screen the correlation strength, so as to obtain the time-domain strong correlation features and the frequency-domain strong correlation features, wherein the lines between the nodes represent the relationship between the features represented by the nodes.

4. The GIS field-electricity fusion real-time status perception and early warning system according to claim 3 is characterized in that: The feature screening module comprises: Feature map construction unit for complete feature maps in graph neural networks middle, is a node set, is the edge set, let Respectively represent nodes The eigenvectors of the edges, the eigenvectors of the nodes and the state vectors of its surrounding nodes and nodes The feature vectors of the surrounding nodes; The aggregation update unit is used to aggregate and update the information of input nodes and edges according to the local transfer function of updating the node state, and output the node label. The formula is expressed as: In the formula, is the local transition function for updating the node state, is the local output function, is the state vector learned by the graph neural network iteration, is the node label; The iteration unit is used to calculate the correlation between a node and its adjacent nodes in the continuous iteration of the graph neural network, and obtain the strong correlation features in the time domain and the strong correlation features in the frequency domain.

5. The GIS field-electricity fusion real-time status perception and early warning system according to claim 4 is characterized in that: The iteration unit is specifically used for: set up and are the vectors constructed by superimposing the state vectors of all nodes, all output labels, the feature vectors of all nodes and edges, and the feature vectors of all nodes. The formula can be written in a more compact form as follows: in, express No. Iterations, Represented by the eigenvector and the The state vector of the iteration is obtained by the global transfer function The state vector of all nodes in the iteration, and are the global transfer function and the global output function, respectively, and are the local transfer functions of all nodes. and local output function Stacking; During the iteration process, nodes with similar states, nodes with complementary states, and the influence of each node on the overall graph neural network are determined. One of the nodes with similar states is selected, nodes with complementary states are merged into one node, and the node with the greatest influence on the overall graph neural network is used as a strong correlation feature.

6. The GIS field-electricity fusion real-time status perception and early warning system according to claim 2 is characterized in that: The signal measurement module comprises: A cache unit, used for performing cache processing on the digital signal; A first monitoring unit is used to extract the time domain strong correlation feature from the digital signal, determine that a partial discharge signal is monitored based on the time domain strong correlation feature, and output a first time domain waveform diagram during discharge; A second monitoring unit is used to extract the frequency domain strong correlation feature from the digital signal, determine that a partial discharge signal is monitored based on the frequency domain strong correlation feature, and output a frequency domain waveform diagram during discharge; A waveform synthesis unit is used to obtain a full time domain waveform diagram of a partial discharge signal based on the first time domain waveform diagram and the frequency domain waveform diagram.

7. The GIS field-electricity fusion real-time status perception and early warning system according to claim 6 is characterized in that: The second monitoring unit comprises: A Fourier transform subunit, used for performing at least two fast Fourier transforms on the electromagnetic field signal to obtain frequency domain data of at least two frequency bands; The normalization subunit is used to extract the frequency domain strong correlation features from the frequency domain data of each frequency band, and after analyzing and matching the frequency domain strong correlation features, when it is determined that the frequency domain data of each frequency band exceeds the second discharge threshold, the frequency domain waveform corresponding to each frequency band is normalized to obtain a frequency domain waveform diagram during discharge.

8. The GIS field-electricity fusion real-time status perception and early warning system according to claim 6 is characterized in that: The waveform synthesis unit comprises: A waveform conversion subunit, used for converting the frequency domain waveform diagram into a second time domain waveform diagram during discharge; The waveform synthesis subunit is used to synthesize the first time domain waveform diagram and the second time domain waveform diagram to obtain a full time domain waveform diagram of the partial discharge signal.

9. The GIS field-electricity fusion real-time status perception and early warning system according to claim 1, characterized in that: The simulation module is used to use the coarse positioning result of the local discharge source obtained based on the full time domain waveform diagram as the initial injection point of the three-dimensional simulation model of the GIS equipment to simulate the local discharge phenomenon of the GIS equipment, and perform simulation calculation to obtain a coarse grid simulation result.

10. The GIS field-electricity fusion real-time status perception and early warning system according to claim 9, characterized in that: The simulation module comprises: The spatial discretization unit is used to discretize the three-dimensional simulation model of the GIS equipment into a spatial grid and determine the corresponding relationship between the number and spatial coordinate data of each grid node; The differential simulation unit is used to inject the coarse positioning results as the initial injection point into the corresponding spatial grid to simulate the partial discharge phenomenon of the GIS equipment, convert the Maxwell equations used to describe the electromagnetic field of the GIS into a fourth-order matrix, and use the four-step HIE-FDTD algorithm to solve the fourth-order matrix and calculate the field strength data of each grid node.

11. The GIS field-electricity fusion real-time status perception and early warning system according to claim 10, characterized in that: The differential simulation unit comprises: The matrix reconstruction subunit is used to convert the Maxwell equations used to describe the electromagnetic field of GIS 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; The matrix conversion subunit is used to convert the sixth-order matrix form into a fourth-order rectangular form: In the formula, and is a sixth-order matrix; The solving subunit is used to solve the fourth-order matrix form using the four-step HIE-FDTD algorithm and calculate the field strength data of each grid node.

12. The GIS field-electricity fusion real-time status perception and early warning system according to claim 11, characterized in that: The solving subunit is used to perform the following steps: 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.

13. The GIS field-electricity fusion real-time status perception and early warning system according to claim 9, characterized in that: The enhanced model adopts a differential deep learning network model, and the coarse grid simulation results include coarse grid structure data and coarse grid field strength data; 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, the simulation enhancement data is calculated.

14. The GIS field-electricity fusion real-time status perception and early warning system according to claim 13, characterized in that: The self-adjusting module includes 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.

15. The GIS field-electricity fusion real-time status perception and early warning system according to claim 13, characterized in that: The differential convolution module includes 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.

16. The GIS field-electricity fusion real-time status perception and early warning system according to claim 15, 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.

17. The GIS field-electricity fusion real-time status perception and early warning system according to claim 13, characterized in that: The structural similarity network includes a first branch network, a second branch network, a second addition operation and a first multilayer 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 multilayer perceptron, and the multilayer perceptron is followed by an activation function.

18. The GIS field-electricity fusion real-time status perception and early warning system according to claim 17, characterized in that: The first branch network includes 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.

19. The GIS field-electricity fusion real-time status perception and early warning system according to claim 18, characterized in that: The using the structural similarity network to match the fine grid enhanced grid structure feature and the fine grid enhanced field strength feature to calculate the similarity feature includes: 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.

20. The GIS field-electricity fusion real-time status perception and early warning system according to claim 13, characterized in that: The step of calculating the simulation enhancement 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 grid field strength loss compensation based on the grid structure consistency and the grid field strength consistency; 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; The simulation enhancement data is calculated based on the fine grid enhanced field strength characteristics, the grid field strength loss compensation and the similarity characteristics.

21. The GIS field-electricity fusion real-time status perception and early warning system according to claim 1, characterized in that: The simulation enhancement data includes an enhanced discharge position and an enhanced discharge intensity, and the position searcher includes: The state transition model building module uses the Markov decision model to reconstruct the state transition model corresponding to the process of the ant colony algorithm exploring the actual discharge position of the local discharge source with the coarse positioning result of the local discharge source as the center, according to the enhanced discharge position and enhanced discharge intensity at each moment; An heuristic space parameterization module generates a heuristic metric using a heuristic learner based on a graph neural network, and converts the state transition model into a location exploration model that is affected by the heuristic metric and requires graph traversal; The iterative search module is used to iteratively search the position exploration model using a neural-guided perturbation interleaved local search algorithm to obtain an actual discharge position of a local discharge source.

22. The GIS field-electricity fusion real-time status perception and early warning system according to claim 21, characterized in that: The state transition model construction module is used to: According to the simulated discharge position and simulated discharge intensity at each moment, the state space corresponding to the actual discharge position of the local discharge source is constructed in the simulator : Where: Indicates that in the simulator The local discharge source node is explored at The corresponding status, Indicates The simulated discharge position at each moment, Indicates The simulated discharge intensity at each moment, It represents the initial discharge intensity of the local discharge source obtained based on the full time domain waveform of GIS local discharge. and initial discharge position As the initial state, represents the exploration cycle, ; Constructing the action space corresponding to the actual discharge location of the PD source in the simulator : Where: Indicates The action of the moment Greedy strategy is the standard choice of action ; The state transition model corresponding to the process of the reconstructed ant colony algorithm exploring the actual discharge position of the local discharge source with the initial discharge position of the local discharge source as the center is: Where: Indicates in action The discharge node To the discharge node The probability of transfer, Indicates that in the simulator The local discharge source node is explored at The corresponding state, Represented by the node Transfer to Node The corresponding pheromone concentration, Represented by the node Transfer to Node The corresponding heuristic function is, Represented by the node Transfer to the included Nodes in The corresponding pheromone concentration, Represented by the node Transfer to the included Nodes in The corresponding heuristic function is, and Represent the pheromone concentration and the weight of the heuristic function, Represents the set of non-uniform grid local discharge source nodes that are allowed to be selected at the next moment.

23. The GIS field-electricity fusion real-time status perception and early warning system according to claim 22, characterized in that: By Node To Node The reward function for the transfer is , Indicates The measured discharge intensity is obtained by processing the local discharge signal collected by the sensor at each moment. Indicates The partial discharge intensity obtained by simulation at each moment.

24. The GIS field-electricity fusion real-time status perception and early warning system according to claim 21, characterized in that: The heuristic space parameterization module is used to transform the graph neural network The layer is The node and Edge features connecting nodes Mapping to heuristic metrics ; The state transition model is transformed into a state transition model that is influenced by the heuristic metric and requires The location exploration model of step graph traversal is: Where: represents the location exploration model, Indicates that in the simulator The local discharge source node is explored at The corresponding status, Indicates that in the simulator The local discharge source node is explored at The corresponding status, represents the exploration cycle, .

25. The GIS field-electricity fusion real-time status perception and early warning system according to claim 21, characterized in that: The location searcher also includes a training module for: The gradient strategy is used to train the heuristic learner based on the graph neural network. The objective function used in the training process is for: Where: represents the objective function corresponding to the actual discharge location of the PD source using the neural-guided perturbation interleaved local search algorithm, Indicates balance and Parameters, In the heuristic metric The expected value of the actual discharge position of the local discharge source under the influence of represents the state space corresponding to the exploration of the actual discharge position of the PD source, represents the objective function; Among them, the objective function The gradient of , the formula is: Where: represents the average target value for directly exploring the actual discharge location of the PD source, represents the average target value of the actual discharge location of the PD source explored using the local search algorithm with neural-guided perturbation interleaving, In the heuristic metric Explore the gradient of the actual discharge location of the partial discharge source under the influence; When the maximum number of iterations is reached Or the maximum number of iterations was not reached But the objective function The training ends when Indicates the minimum threshold corresponding to the objective function.

26. The GIS field-electricity fusion real-time status perception and early warning system according to claim 21, characterized in that: The iterative search module is used to: A local search unit, used to iteratively search the position exploration model using a local search algorithm to obtain a local optimal solution; A perturbation unit is used to perform neural-guided perturbation on the local optimal solution obtained by the current iterative search, and obtain the optimal exploration scheme of the actual discharge position of the local discharge source of the current iterative search through the local search of neural-guided perturbation interleaving; A pheromone concentration updating unit, used for updating the pheromone concentrations between the local discharge source nodes after all the local discharge source nodes are selected; The exploration unit is used to determine the optimal exploration scheme of the actual discharge position of the partial discharge source when the iterative search reaches the iterative convergence condition, and realize the matching of the actual discharge position of the partial discharge source based on the optimal exploration scheme.

27. The GIS field-electricity fusion real-time status perception and early warning system according to claim 26, characterized in that: The process formula of the disturbance unit exploring the optimal exploration scheme for the actual discharge position of the current iterative partial discharge source is expressed as: Where: represents the optimal exploration scheme of the actual discharge position of the local discharge source in the current iteration obtained by performing neural-guided perturbation on the local optimal solution, is the local optimal solution; represents the number of perturbation moves, Represents a heuristic metric.

28. The GIS field-electricity fusion real-time status perception and early warning system according to claim 1, characterized in that: The early warning device is equipped with a fault classification module, which includes: A deep feature extraction network is used to obtain deep features corresponding to the UHF signal, wherein the deep feature extraction network includes a feature extraction network and an output network, the feature extraction network is formed by stacking a plurality of feature extraction layers, and each of the feature extraction layers includes a channel attention module and a spatial attention module connected in sequence; A reinforcement learning unit, used to construct a data set using the deep features, and train a GIS partial discharge feature matching Markov model using a deep reinforcement learning algorithm to obtain an optimal GIS partial discharge feature matching result; The early warning unit generates early warning information based on the optimal GIS local discharge feature matching result and the actual discharge position of the local discharge source to perform fault early warning.

29. A GIS field-electricity fusion real-time state perception and early warning method, characterized in that: The method for realizing GIS fault warning by using the GIS field-electricity fusion real-time status perception and warning system according to any one of claims 1 to 28 comprises: When the electromagnetic field signal inside the GIS is monitored according to the strong correlation characteristics obtained in advance, including the partial discharge signal, a full time domain waveform diagram including the time domain waveform diagram and the frequency domain waveform diagram of the partial discharge signal is obtained; Based on the full time domain waveform of GIS partial discharge, the partial discharge source is roughly located, and the simulation enhancement calculation is performed using the rough positioning result as the initial injection point to obtain simulation enhancement data of the partial discharge source; Iteratively search the measured discharge intensity calculated according to the full time domain waveform diagram and the simulation enhancement result to obtain the actual discharge position of the partial discharge source; An early warning is provided based on the full time domain waveform diagram and the early warning information including the actual discharge position of the partial discharge source.

Images