Method and device for solving surface current of three-dimensional metal object under electromagnetic wave irradiation

Abstract

The invention relates to a method and device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation, and the method comprises the steps: building a target metal subdivision model corresponding to a target three-dimensional metal object and the graph structure data expression of the surface current; constructing a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object based on the graph structure data expression, generating training data of the graph neural network solver based on a preset moment method strategy, and training the graph neural network solver through the training data; and inputting the to-be-solved metal subdivision model and a plurality of preset plane wave excitation parameters into the trained graph neural network solver so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver. Therefore, the problem that an existing electromagnetic solver based on a neural network cannot process a non-uniform grid form is solved.

Description

Technical Field

[0001] The present application relates to the technical field of solving the surface current of three-dimensional metal objects, and in particular to a method and device for solving the surface current of three-dimensional metal objects under electromagnetic wave irradiation. Background Art

[0002] With the development of electronic equipment, communication systems, automobiles, aerospace and other industries, the contribution of electromagnetic computing technology in design optimization, performance improvement, cost reduction, etc. has gradually emerged, becoming the core force driving the development of many high-tech industries.

[0003] In antenna design, electronic devices, electromagnetic compatibility testing, sensor design, etc., accurate electromagnetic calculations can verify and optimize product designs in a virtual environment, reduce the number of physical prototypes, and reduce R&D costs and time. Metal objects are the main targets for electromagnetic wave reflection, so it is very important to efficiently and accurately solve their surface currents in electromagnetic calculations. Currently, the calculation of metal surface currents requires the moment method to discretize the joint integral equations of the metal target surface, obtain the coefficients of the basis function by solving the matrix equation, and use interpolation to obtain the surface current. However, when dealing with large-scale problems, the filling of the impedance matrix and the solution of the matrix equation greatly reduce the efficiency of the calculation. With the application of deep learning in solving physical equations, neural networks have been introduced into the field of electromagnetic computing.

[0004] However, although neural networks have powerful nonlinear modeling capabilities and good generalization performance, existing neural network-based electromagnetic solvers cannot handle non-uniform grid forms, which needs to be solved urgently. Summary of the invention

[0005] The present application provides a method and device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation, so as to solve the problem that the existing electromagnetic solver based on neural network cannot handle the non-uniform grid form.

[0006] The first aspect of the present application provides a method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation, comprising the following steps: establishing a target metal decomposition model and graph structure data of the surface current corresponding to the target three-dimensional metal object; based on the graph structure data, constructing a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object, and based on a preset method of moments strategy, generating training data for the graph neural network solver, and training the graph neural network solver through the training data; inputting the metal decomposition model to be solved and a plurality of preset plane wave excitation parameters into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver.

[0007] Optionally, in one embodiment of the present application, the establishment of the target metal segmentation model and the graph structure data of the surface current corresponding to the target three-dimensional metal object includes: establishing a joint integral equation between the equivalent current on the surface of the target three-dimensional metal object and the target plane wave excitation, and calculating the induced current of the target plane wave on the surface of the target three-dimensional metal object according to the joint integral equation and preset boundary conditions; obtaining multiple triangular facet metadata of the segmentation of the target three-dimensional metal object, and converting the spatial geometric structure corresponding to the target three-dimensional metal object into graph structure data according to the multiple triangular facet metadata; converting the induced current and the equivalent current at the center of the triangle corresponding to each triangular facet metadata in the multiple triangular facet metadata into real scalars, so as to determine the eigenvector of each node in the graph structure data according to the real scalars; and determining the graph structure data of the surface current based on the eigenvector.

[0008] Optionally, in one embodiment of the present application, based on the graph structure data, a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object is constructed, and based on a preset method of moments strategy, training data for the graph neural network solver is generated, and the graph neural network solver is trained through the training data, including: based on the feature vector and the induced current, an upsampling graph neural network module, a convolutional graph neural network module and a downsampling graph neural network module are constructed, and the graph neural network solver is constructed through the upsampling graph neural network module, the convolutional graph neural network module and the downsampling graph neural network module; multiple metal models corresponding to the target three-dimensional metal object are constructed, and the multiple metal models are segmented to obtain the segmentation corresponding to each of the multiple metal models. data; using the segmented data to establish the graph structure data corresponding to each metal model, and converting the induced current into a feature vector of each graph structure data node input into the up-sampled graph neural network module; solving the equivalent current at the center of each triangle by a preset moment method, and setting the equivalent current as a label, and constructing a feature vector of the graph structure data of the label according to the label; inputting the graph structure data into the graph neural network solver to iteratively train the graph neural network solver, and outputting the graph structure data of the network predicted equivalent current in each iteration process, and calculating the mean square error between the feature vector of each node corresponding to the graph structure data of the network predicted equivalent current and the label feature vector, until a graph neural network solver whose mean square error meets the preset accuracy requirements is obtained.

[0009] Optionally, in one embodiment of the present application, the metal segmentation model to be solved and the preset multiple plane wave excitation parameters are input into the trained graph neural network solver to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver, including: obtaining the analysis data corresponding to the metal segmentation model to be solved, and establishing the graph structure data corresponding to the metal segmentation model to be solved according to the analysis data; converting the multiple plane wave excitation parameters into the feature vectors of the corresponding graph structure data, and inputting the graph structure data into the trained graph neural network solver to output the solved graph structure data, and obtaining the surface current corresponding to the target three-dimensional metal object according to the feature vector of the solved graph structure data.

[0010] The second aspect of the present application provides a device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation, including: a modeling module, used to establish a target metal decomposition model and graph structure data of the surface current corresponding to the target three-dimensional metal object; a training module, used to construct a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object based on the graph structure data, and based on a preset moment method strategy, generate training data for the graph neural network solver, and train the graph neural network solver through the training data; a solving module, used to input the metal decomposition model to be solved and a plurality of preset plane wave excitation parameters into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver.

[0011] Optionally, in one embodiment of the present application, the modeling module includes: a first calculation unit, used to establish a joint integral equation between the equivalent current on the surface of the target three-dimensional metal object and the target plane wave excitation, and calculate the induced current of the target plane wave on the surface of the target three-dimensional metal object according to the joint integral equation and preset boundary conditions; a conversion unit, used to obtain multiple triangular face metadata of the target three-dimensional metal object, and convert the spatial geometric structure corresponding to the target three-dimensional metal object into graph structure data according to the multiple triangular face metadata; a first determination unit, used to convert the induced current and the equivalent current at the center of the triangle corresponding to each triangular face metadata in the multiple triangular face metadata into a real scalar, so as to determine the eigenvector of each node in the graph structure data according to the real scalar; a second determination unit, used to determine the graph structure data of the surface current based on the eigenvector.

[0012] Optionally, in one embodiment of the present application, the training module includes: a construction unit, which is used to construct an upsampling graph neural network module, a convolutional graph neural network module and a downsampling graph neural network module based on the feature vector and the induced current, and construct the graph neural network solver through the upsampling graph neural network module, the convolutional graph neural network module and the downsampling graph neural network module; a segmentation unit, which is used to construct multiple metal models corresponding to the target three-dimensional metal object, and segment the multiple metal models to obtain segmentation data corresponding to each metal model in the multiple metal models; an establishment unit, which is used to establish the graph structure data corresponding to each metal model using the segmentation data, and convert the induced current into The upsampling graph neural network module inputs the feature vector of each graph structure data node; a setting unit is used to solve the equivalent current of the center of each triangle by a preset moment method, and set the equivalent current as a label, and construct a feature vector of the graph structure data of the label according to the label; an iteration unit is used to input the graph structure data into the graph neural network solver to iteratively train the graph neural network solver, and output the graph structure data of the network predicted equivalent current in each iteration process, and calculate the mean square error between the feature vector of each node corresponding to the graph structure data of the network predicted equivalent current and the label feature vector, until a graph neural network solver whose mean square error meets the preset accuracy requirements is obtained.

[0013] Optionally, in one embodiment of the present application, the solution module includes: an acquisition unit, used to acquire analysis data corresponding to the metal segmentation model to be solved, and establish graph structure data corresponding to the metal segmentation model to be solved based on the analysis data; a second calculation unit, used to convert the multiple plane wave excitation parameters into feature vectors of the corresponding graph structure data, and input the graph structure data into the trained graph neural network solver to output the solved graph structure data, and obtain the surface current corresponding to the target three-dimensional metal object based on the feature vector of the solved graph structure data.

[0014] The third aspect of the present application provides an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement a method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation as described in the above embodiment.

[0015] The fourth aspect of the present application provides a computer-readable storage medium, which stores a computer program. When the program is executed by a processor, it implements the above method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation.

[0016] The fifth aspect of the present application provides a computer program product, including a computer program, which is executed to implement the above-mentioned method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation.

[0017] Therefore, the embodiments of the present application have the following beneficial effects:

[0018] The embodiments of the present application can be implemented by establishing a target metal partition model and graph structure data of the surface current corresponding to the target three-dimensional metal object; based on the graph structure data, constructing a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object, and based on a preset method of moments strategy, generating training data for the graph neural network solver, and training the graph neural network solver through the training data; inputting the metal partition model to be solved and a plurality of preset plane wave excitation parameters into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver. The present application utilizes a graph neural network solution strategy for the surface current of a three-dimensional metal object under plane wave excitation, so that neural networks can be widely used in electromagnetic calculations. Thus, the problem that the existing electromagnetic solvers based on neural networks cannot handle non-uniform grid forms is solved.

[0019] Additional aspects and advantages of the present application will be given in part in the description below, and in part will become apparent from the description below, or will be learned through the practice of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0020] The above and / or additional aspects and advantages of the present application will become apparent and easily understood from the following description of the embodiments in conjunction with the accompanying drawings, in which:

[0021] Figure 1 A flow chart of a method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation provided according to an embodiment of the present application;

[0022] Figure 2 A schematic diagram of a process of converting a non-uniform mesh model of a three-dimensional metal object into graph structure data provided by an embodiment of the present application;

[0023] Figure 3 A graph neural network structure and a schematic diagram of network parameters provided for an embodiment of the present application;

[0024] Figure 4 A schematic diagram of the execution logic of a method for solving the current on the surface of a three-dimensional metal object under electromagnetic wave irradiation provided by one embodiment of the present application;

[0025] Figure 5A schematic diagram for comparing x, y, and z components of equivalent current on the surface of a metal object solved by a method of moment and a graph neural network solver provided in one embodiment of the present application;

[0026] Figure 6 A schematic diagram of a comparison of the current on the surface of a conducting sphere solved by a method of moments and a graph neural network solver provided in one embodiment of the present application;

[0027] Figure 7 This is an example diagram of a device for solving the current on the surface of a three-dimensional metal object under electromagnetic wave irradiation according to an embodiment of the present application;

[0028] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application.

[0029] Among them, 10 is a device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation; 100 is a modeling module, 200 is a training module, 300 is a solving module; 801 is a memory, 802 is a processor, and 803 is a communication interface. DETAILED DESCRIPTION

[0030] Embodiments of the present application are described in detail below, and examples of the embodiments are shown in the accompanying drawings, wherein the same or similar reference numerals throughout represent the same or similar elements or elements having the same or similar functions. The embodiments described below with reference to the accompanying drawings are exemplary and are intended to be used to explain the present application, and should not be construed as limiting the present application.

[0031] The following describes the method and device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation of an embodiment of the present application with reference to the accompanying drawings. In view of the problems mentioned in the above background technology, the present application provides a method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation. In the method, a target metal segmentation model corresponding to the target three-dimensional metal object and a graph structure data of the surface current are established; based on the graph structure data, a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object is constructed, and based on the preset moment method strategy, training data for the graph neural network solver is generated, and the graph neural network solver is trained by the training data; the metal segmentation model to be solved and the preset multiple plane wave excitation parameters are input into the trained graph neural network solver to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver. The present application utilizes the graph neural network solution strategy for the surface current of a three-dimensional metal object under plane wave excitation, so that neural networks can be widely used in electromagnetic calculations. As a result, the problem that the existing electromagnetic solver based on neural network cannot handle non-uniform grid forms is solved.

[0032] Specifically, Figure 1A flow chart of a method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation provided in an embodiment of the present application.

[0033] like Figure 1 As shown, the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation includes the following steps:

[0034] In step S101, a target metal segmentation model and graph structure data of surface current corresponding to a target three-dimensional metal object are established.

[0035] The embodiments of the present application can first construct a target metal segmentation model corresponding to the target three-dimensional metal object, and establish a conversion relationship between the variables in the three-dimensional metal object joint integral equation and the computer graphic structure data, thereby providing reliable data support for the subsequent solution of the surface current of the three-dimensional metal object.

[0036] Optionally, in one embodiment of the present application, a target metal segmentation model and graph structure data of the surface current corresponding to the target three-dimensional metal object are established, including: establishing a joint integral equation between the equivalent current on the surface of the target three-dimensional metal object and the target plane wave excitation, and calculating the induced current of the target plane wave on the surface of the target three-dimensional metal object based on the joint integral equation and preset boundary conditions; obtaining multiple triangular facet metadata of the target three-dimensional metal object segmentation, and converting the spatial geometric structure corresponding to the target three-dimensional metal object into graph structure data based on the multiple triangular facet metadata; converting the induced current and the equivalent current at the center of the triangle corresponding to each triangular facet metadata in the multiple triangular facet metadata into real scalars, so as to determine the eigenvector of each node in the graph structure data based on the real scalars; and determining the graph structure data of the surface current based on the eigenvectors.

[0037] In the embodiment of the present application, the conversion process of the metal object segmentation model to the graph structure data is as follows: Figure 2 Specifically, the specific process of establishing the conversion relationship between the variables in the three-dimensional metal object joint integral equation and the computer graphic structure data in the embodiment of the present application is as follows:

[0038] S1. Construct the physical relationship between equivalent current and induced current form:

[0039] Establish the equivalent current J on the surface of a three-dimensional metal object S (r) with plane wave excitation E inc (r),H inc The joint integral equation between (r) is shown as follows:

[0040]

[0041] The boundary conditions are used to calculate the induced current of the plane wave on the metal surface. As shown below:

[0042]

[0043] S2. Build graph structure data of metal segmentation model:

[0044] Input the triangular face data of the metal object, and convert the spatial geometric structure into graph structure data G = (V, E), where V = {v i |i=1,…,M} represents all M nodes, each node corresponds to a triangular face; E={e ij |e ij =(v i ,v j )∈V 2 ,v i ≠v j} are all the edges, which represent the relationship between the corresponding nodes of the triangles with common edges;

[0045] S3. Convert the equivalent current into the characteristic vector of the graph node:

[0046] The induced current at the center of each triangle and the equivalent current J S (r) is converted into a real scalar and used as the graph structure data for each node v i The eigenvector F(v i ), each eigenvector contains the real and imaginary parts of the current in each direction as well as the spatial coordinates, as shown in the following formula:

[0047] F(v i )=[J x,r ,J x,i ,J y,r ,J y,i ,J z,r ,J z,i ,x,y,z]

[0048] Wherein, the subscripts r and i represent the real and imaginary parts of the current in each direction respectively; x, y, and z represent the spatial coordinates of the triangular surface element.

[0049] Therefore, the embodiments of the present application construct the physical relationship between the equivalent current and the induced current form and the graph structure data of the metal dissection model, and convert the equivalent current into the characteristic vector of the graph node, thereby characterizing the surface current of the three-dimensional metal object through the graph structure data.

[0050] In step S102, based on the graph structure data, a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object is constructed, and based on the preset method of moments strategy, training data for the graph neural network solver is generated, and the graph neural network solver is trained through the training data.

[0051] Furthermore, the embodiments of the present application also need to construct multiple graph neural network modules based on the variable format of the graph, and combine multiple graph neural network modules to establish a graph neural network solver, so as to simulate the physical process defined by the joint integral equation; secondly, the embodiments of the present application can use the moment method to calculate the surface current of multiple models under multiple plane wave excitations as the true value, and adopt supervised learning to optimize the graph neural network parameters, so as to effectively ensure the robustness and accuracy of the graph neural network.

[0052] Optionally, in one embodiment of the present application, based on the graph structure data, a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object is constructed, and based on the preset method of moments strategy, training data for the graph neural network solver is generated, and the graph neural network solver is trained through the training data, including: based on the feature vector and the induced current, an upsampling graph neural network module, a convolutional graph neural network module and a downsampling graph neural network module are constructed, and a graph neural network solver is constructed through the upsampling graph neural network module, the convolutional graph neural network module and the downsampling graph neural network module; multiple metal models corresponding to the target three-dimensional metal object are constructed, and the multiple metal models are segmented to obtain the segmentation data corresponding to each of the multiple metal models. ; Use the segmented data to establish the graph structure data corresponding to each metal model, and convert the induced current into the feature vector of each graph structure data node input into the upsampling graph neural network module; solve the equivalent current at the center of each triangle through the preset moment method, set the equivalent current as the label, and construct the feature vector of the graph structure data of the label according to the label; input the graph structure data into the graph neural network solver to iteratively train the graph neural network solver, and output the graph structure data of the network predicted equivalent current in each iteration process, and calculate the mean square error between the feature vector of each node corresponding to the graph structure data of the network predicted equivalent current and the label feature vector, until a graph neural network solver whose mean square error meets the preset accuracy requirements is obtained.

[0053] It should be noted that if Figure 3 As shown, the embodiment of the present application constructs graph neural network modules of upsampling, convolution and downsampling based on the variable format of the graph, and combines various modules to establish a graph neural network solver to simulate the physical process defined by the joint integral equation. The specific steps are as follows:

[0054] S1. Establish an upsampling graph neural network module:

[0055] In an embodiment of the present application, the upsampling fully connected graph neural network module (i.e., the upsampling graph neural network module) G up =(V,E,F up )=Φ up (G in ,Θup ), where G in =(V,E,F in ) represents the input graph structure data of the network; the feature vector F in The induced current formed for excitation G up is the output of the upsampled fully connected graph neural network; F up is the upsampled feature vector of each node; Φ up represents the fully connected network of the upsampled graph; Θ up represents a network parameter; thus, the embodiment of the present application improves the data dimension through an upsampling process, thereby ensuring the accuracy of the solver;

[0056] S2. Establish a convolutional graph neural network module:

[0057] The convolutional graph neural network module G in the embodiment of the present application gc =(V,E,F gc )=Φ gc (G up ,Θ gc ), where G gc Represents the output graph structure data of the graph convolutional neural network; F gc Represents the feature vector of each node after convolution; Φ gc represents graph convolutional neural network; Θ gc Represents the network parameters; the calculation process of graph convolution in the network is shown as follows:

[0058]

[0059] in, and Respectively represent the pth and p+1th layers of the graph convolutional network; the feature vector corresponding to the i-th node; is a trainable kernel function; is the set of adjacent nodes of the i-th node; is the number of adjacent nodes; W is the trainable weight matrix; Represents the position vector of the edge between the graph structure data nodes i and j; thus, the embodiment of the present application simulates the physical process of the joint integral equation through a multi-layer graph convolutional neural network to deeply fit the upsampled graph structure data;

[0060] S3. Establish a downsampling fully connected graph neural network module:

[0061] In the embodiment of the present application, the downsampling fully connected graph neural network module (ie, the downsampling graph neural network module) G x\y\z,r\i =(V,E,F x\y\z,r\i )=Φ x\y\z,r\i (Ggc ,Θ x\y\z,r\i ), where G x\y\z,r\i is the graph structure data output after downsampling, and its feature vector F x\y\z,r\i The directional components of the surface equivalent current calculated by the neural network solver, including the real part, the imaginary part and the nine scalars of the position coordinates F out (v i )=[J S,x,r ,J S,x,i ,J S,y,r ,J S,y,i ,J S,z,r ,J S,z,i ,x,y,z]; Φ x\y\z,r\i is a downsampled fully connected graph neural network; Θ x\y\z,r\i are its network parameters.

[0062] Afterwards, the embodiment of the present application can use the moment method to generate labels to perform supervised learning to optimize network parameters. The specific process is as follows:

[0063] S1. Establish multiple metal models and perform segmentation, setting different plane wave incident frequencies f inc , the incident angle (θ inc ,φ inc ), and the polarization angle And using the RWG basis function ρ n The equivalent current is discretized as shown in the following formula:

[0064]

[0065] Among them, J n is the coefficient of the basis function; the joint integral equation is discretized by the moment method to obtain the coefficient J corresponding to each basis function n , and the equivalent current J of each triangular face element is obtained by interpolation MoM,S (r);

[0066] S2. Use the segmented data to build the graph structure data G of each model train =(V,E), the induced current generated by plane wave excitation at the center of each triangle Convert to upsampled fully connected graph neural network input feature vector of each node Then, the embodiment of the present application can solve the equivalent current J of each triangle center by the moment method. MoM,S (r) as a label and convert the label feature vector of the graph structure data

[0067]

[0068] S3, G train=(V,E,F in (v i )) Input the graph neural network solver and output the network predicted equivalent current graph structure data G train =(V,E,F out (v i )) Calculate the feature vector F of each node of the output data out (v i ) and the label feature vector F MoM (v i ), as shown below:

[0069]

[0070] Among them, N tot is the number of nodes in a single model; ||·|| F is the Frobenius norm.

[0071] As an achievable method, an embodiment of the present application may use the mean square error as the loss function of the graph neural network for training, so as to obtain a graph neural network solver that meets the accuracy requirements through iteration.

[0072] Therefore, the embodiments of the present application construct a graph neural network module and form a graph neural network solver for the joint integral equation, and generate training data through the moment method and train and optimize the neural network, so as to obtain a graph neural network solver with excellent performance, which effectively ensures the reliability of subsequent surface current calculations.

[0073] In step S103, the metal segmentation model to be solved and multiple preset plane wave excitation parameters are input into the trained graph neural network solver to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver.

[0074] Furthermore, the embodiments of the present application also need to use a graph neural network to solve the surface current corresponding to the three-dimensional metal object based on the given metal segmentation model to be solved and the plane wave excitation parameters.

[0075] Optionally, in one embodiment of the present application, the metal segmentation model to be solved and a plurality of preset plane wave excitation parameters are input into a trained graph neural network solver to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver, including: obtaining analysis data corresponding to the metal segmentation model to be solved, and establishing graph structure data corresponding to the metal segmentation model to be solved based on the analysis data; converting a plurality of plane wave excitation parameters into feature vectors of the corresponding graph structure data, and inputting the graph structure data into the trained graph neural network solver to output the solved graph structure data, and obtaining the surface current corresponding to the target three-dimensional metal object based on the feature vectors of the solved graph structure data.

[0076] Specifically, the steps of solving the surface current using the graph neural network in the embodiment of the present application are as follows:

[0077] S1. Use the segmentation file of the metal segmentation model to be solved to establish the graph structure data G target =(V,E), setting the incident frequency f of the plane wave excitation inc , angle of incidence (θ inc ,φ inc ) and polarization angle Secondly, the plane wave is converted into an induced current on the triangular surface element And convert the induced current into the characteristic vector of each node of the network input graph structure data

[0078] S2, the graph structure data G target =(V,E,F target (v i )) Input the graph neural network solver to output the graph structure data G target,out =(V,E,F target,out (v i ), where F target,out (v i ) is the equivalent current feature vector of the three-dimensional metal structure predicted by the graph neural network; thus, the embodiment of the present application obtains the equivalent current J of the metal object surface through the correspondence between the graph structure data and the segmentation data S (r).

[0079] It can be understood that the embodiments of the present application characterize the surface current of a three-dimensional metal object by adopting a graph-type data structure, constructing a graph neural network module and forming a graph neural network solver for a joint integral equation, and generating training data through the moment method and training and optimizing the neural network; thereafter, the embodiments of the present application input the target model and excitation parameters in a graph structure data format, thereby quickly calculating the surface current through the graph neural network solver.

[0080] The following describes and analyzes the execution logic and solution results of the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation of the present application in combination with the accompanying drawings.

[0081] Figure 4 The following is a logic diagram for the method for solving the current on the surface of a three-dimensional metal object under electromagnetic wave irradiation of the present application. Figure 4 As shown, the execution process of the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation of the present application is as follows:

[0082] S401: Establishing metal segmentation model and graphical data expression of surface current;

[0083] S402: Build a graph neural network solver and use the moment method to generate labels for supervised learning to optimize network parameters;

[0084] S403: Convert the preset plane electromagnetic wave excitation parameters into graph structure data and input it into the graph neural network solver to predict the metal surface current through the graph neural network.

[0085] Figure 5 The comparison of the equivalent current x, y, and z components of the metal object surface solved by the MoM and the GNN solver is shown, where MoM, GNN, and AE represent the MoM, GNN, and absolute error, respectively. Figure 5 It can be seen that the surface current solved by the graph neural network is in good agreement with the result obtained by the method of moments, with a low error level.

[0086] Figure 6 The comparison of the surface current of a conducting sphere solved by the method of moments and the graph neural network solver is shown, and Figure 5 The x, y, and z components of the equivalent current on the metal object surface correspond to Figure 6 Three-dimensional surface current in .

[0087] Therefore, the embodiment of the present application characterizes the surface current of a three-dimensional metal object by adopting a graph-type data structure, and uses three graph neural network modules of upsampling, convolution and downsampling to form a graph neural network solver for the joint integral equation of the metal object, and generates training data through the method of moments and trains and optimizes the neural network; thereafter, the embodiment of the present application can convert the segmentation model and excitation parameters of the target to be solved into the form required by the graph network, and input them into the neural network to obtain the predicted surface current, thereby effectively reducing the error level of the surface current obtained by the graph neural network solver of the surface current of three-dimensional metal objects under plane wave excitation, and improving the calculation efficiency and accuracy of the surface current.

[0088] According to the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation proposed in the embodiment of the present application, a target metal segmentation model corresponding to the target three-dimensional metal object and graph structure data of the surface current are established; based on the graph structure data, a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object is constructed, and based on a preset method of moments strategy, training data for the graph neural network solver is generated, and the graph neural network solver is trained with the training data; the metal segmentation model to be solved and a plurality of preset plane wave excitation parameters are input into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver. The present application utilizes the graph neural network solution strategy for the surface current of three-dimensional metal objects under plane wave excitation, so that neural networks can be widely used in electromagnetic computing.

[0089] Secondly, a device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation proposed in an embodiment of the present application is described with reference to the accompanying drawings.

[0090] Figure 7 It is a block diagram of a device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation according to an embodiment of the present application.

[0091] like Figure 7 As shown, the device 10 for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation includes: a modeling module 100, a training module 200 and a solving module 300.

[0092] The modeling module 100 is used to establish a target metal segmentation model and surface current graph structure data corresponding to the target three-dimensional metal object.

[0093] The training module 200 is used to construct a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object based on the graph structure data, and to generate training data for the graph neural network solver based on a preset method of moments strategy, and to train the graph neural network solver through the training data.

[0094] The solution module 300 is used to input the metal segmentation model to be solved and multiple preset plane wave excitation parameters into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver.

[0095] Optionally, in one embodiment of the present application, the modeling module 100 includes: a first calculation unit, a conversion unit, a conversion unit, a first determination unit and a second determination unit.

[0096] Among them, the first calculation unit is used to establish a joint integral equation between the equivalent current on the surface of the target three-dimensional metal object and the target plane wave excitation, and calculate the induced current of the target plane wave on the surface of the target three-dimensional metal object based on the joint integral equation and preset boundary conditions.

[0097] The conversion unit is used to obtain multiple triangular face metadata of the target three-dimensional metal object, and convert the spatial geometric structure corresponding to the target three-dimensional metal object into graph structure data according to the multiple triangular face metadata.

[0098] The first determination unit is used to convert the induced current and the equivalent current at the center of the triangle corresponding to each triangle face data in the multiple triangle face data into a real scalar, so as to determine the characteristic vector of each node in the graph structure data according to the real scalar.

[0099] The second determining unit is used to determine the graph structure data of the surface current based on the characteristic vector.

[0100] Optionally, in one embodiment of the present application, the training module 200 includes: a construction unit, a segmentation unit, a creation unit, a setting unit and an iteration unit.

[0101] Among them, the construction unit is used to construct an upsampling graph neural network module, a convolutional graph neural network module and a downsampling graph neural network module based on the feature vector and the induced current, and to construct a graph neural network solver through the upsampling graph neural network module, the convolutional graph neural network module and the downsampling graph neural network module.

[0102] The segmentation unit is used to construct multiple metal models corresponding to the target three-dimensional metal object, and segment the multiple metal models to obtain segmentation data corresponding to each of the multiple metal models.

[0103] A unit is established for establishing graph structure data corresponding to each metal model using the segmented data, and converting the induced current into a feature vector of each graph structure data node input into an up-sampling graph neural network module.

[0104] A setting unit is used to solve the equivalent current at the center of each triangle by a preset moment method, set the equivalent current as a label, and construct a feature vector of the graph structure data of the label according to the label.

[0105] An iteration unit is used to input graph structure data into a graph neural network solver to iteratively train the graph neural network solver, and output the graph structure data of the network predicted equivalent current in each iteration process, and calculate the mean square error between the feature vector of each node corresponding to the graph structure data of the network predicted equivalent current and the label feature vector, until a graph neural network solver whose mean square error meets the preset accuracy requirements is obtained.

[0106] Optionally, in one embodiment of the present application, the solution module 300 includes: an acquisition unit and a second calculation unit.

[0107] The acquisition unit is used to acquire the analysis data corresponding to the metal dissection model to be solved, and to establish the graph structure data corresponding to the metal dissection model to be solved according to the analysis data.

[0108] The second calculation unit is used to convert multiple plane wave excitation parameters into characteristic vectors of corresponding graph structure data, and input the graph structure data into the trained graph neural network solver to output the solved graph structure data, and obtain the surface current corresponding to the target three-dimensional metal object according to the characteristic vector of the solved graph structure data.

[0109] It should be noted that the aforementioned explanation of the embodiment of the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation is also applicable to the device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation of this embodiment, and will not be repeated here.

[0110] According to the device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation proposed in the embodiment of the present application, it includes a modeling module for establishing a target metal segmentation model corresponding to the target three-dimensional metal object and graph structure data of the surface current; a training module for constructing a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object based on the graph structure data, and generating training data for the graph neural network solver based on a preset method of moments strategy, and training the graph neural network solver through the training data; a solving module for inputting the metal segmentation model to be solved and a plurality of preset plane wave excitation parameters into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver. The present application utilizes the graph neural network solving strategy for the surface current of three-dimensional metal objects under plane wave excitation, so that neural networks can be widely used in electromagnetic computing.

[0111] Figure 8 A schematic diagram of the structure of an electronic device provided in an embodiment of the present application. The electronic device may include:

[0112] A memory 801 , a processor 802 , and a computer program stored in the memory 801 and executable on the processor 802 .

[0113] When the processor 802 executes the program, the method for solving the surface current of a three-dimensional metal object under the irradiation of electromagnetic waves provided in the above embodiment is implemented.

[0114] Furthermore, the electronic device further comprises:

[0115] The communication interface 803 is used for communication between the memory 801 and the processor 802 .

[0116] The memory 801 is used to store computer programs that can be executed on the processor 802 .

[0117] The memory 801 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0118] If the memory 801, the processor 802 and the communication interface 803 are implemented independently, the communication interface 803, the memory 801 and the processor 802 can be connected to each other through a bus and communicate with each other. The bus can be an Industry Standard Architecture (ISA) bus, a Peripheral Component (PCI) bus or an Extended Industry Standard Architecture (EISA) bus. The bus can be divided into an address bus, a data bus, a control bus, etc. For ease of representation, Figure 8 Only one thick line is used in the diagram, but this does not mean that there is only one bus or only one type of bus.

[0119] Optionally, in a specific implementation, if the memory 801, the processor 802 and the communication interface 803 are integrated on a chip, the memory 801, the processor 802 and the communication interface 803 can communicate with each other through an internal interface.

[0120] The processor 802 may be a central processing unit (CPU), or an application specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present application.

[0121] An embodiment of the present application also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the above-mentioned method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation.

[0122] An embodiment of the present application also provides a computer program product, including a computer program, which, when executed, is used to implement the above-mentioned method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation.

[0123] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "example", "specific example", or "some examples" etc. 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 application. In this specification, the schematic representations of the above terms do 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 N embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine the different embodiments or examples described in this specification and the features of the different embodiments or examples, without contradiction.

[0124] 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 technical features indicated. Therefore, a feature defined as "first" or "second" may explicitly or implicitly include at least one of the features. In the description of this application, "N" means at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0125] Any process or method description in a flowchart or otherwise described herein may be understood to represent a module, fragment or portion of code comprising one or N executable instructions for implementing the steps of a custom logical function or process, and the scope of the preferred embodiments of the present application includes alternative implementations in which functions may not be performed in the order shown or discussed, including performing functions in a substantially simultaneous manner or in reverse order depending on the functions involved, which should be understood by technicians in the technical field to which the embodiments of the present application belong.

[0126] The logic and / or steps represented in the flowchart or otherwise described herein, for example, can be considered as an ordered list of executable instructions for implementing logical functions, and can be embodied in any computer-readable medium for use by an instruction execution system, device or apparatus (such as a computer-based system, a system including a processor, or other system that can fetch instructions from an instruction execution system, device or apparatus and execute instructions), or in combination with these instruction execution systems, devices or apparatuses. For the purpose of this specification, "computer-readable medium" can be any device that can contain, store, communicate, propagate or transmit a program for use by an instruction execution system, device or apparatus, or in combination with these instruction execution systems, devices or apparatuses. More specific examples of computer-readable media (a non-exhaustive list) include the following: an electrical connection with one or N wirings (electronic devices), a portable computer disk box (magnetic device), a random access memory (RAM), a read-only memory (ROM), an erasable and programmable read-only memory (EPROM or flash memory), a fiber optic device, and a portable compact disk read-only memory (CDROM). In addition, the computer-readable medium may even be paper or other suitable medium on which the program is printed, since the program may be obtained electronically by optically scanning the paper or other medium and then editing, interpreting or processing in other suitable ways as necessary and then storing it in a computer memory.

[0127] It should be understood that the various parts of the present application can be implemented by hardware, software, firmware or a combination thereof. In the above embodiment, the N steps or methods can be implemented by software or firmware stored in a memory and executed by a suitable instruction execution system. If implemented by hardware, as in another embodiment, it can be implemented by any one of the following technologies known in the art or their combination: a discrete logic circuit having a logic gate circuit for implementing a logic function for a data signal, a dedicated integrated circuit having a suitable combination of logic gate circuits, a programmable gate array (PGA), a field programmable gate array (FPGA), etc.

[0128] A person skilled in the art may understand that all or part of the steps in the method for implementing the above-mentioned embodiment may be completed by instructing related hardware through a program, and the program may be stored in a computer-readable storage medium, which, when executed, includes one or a combination of the steps of the method embodiment.

[0129] In addition, each functional unit in each embodiment of the present application may be integrated into a processing module, or each unit may exist physically separately, or two or more units may be integrated into one module. The above-mentioned integrated module may be implemented in the form of hardware or in the form of a software functional module. If the integrated module is implemented in the form of a software functional module and sold or used as an independent product, it may also be stored in a computer-readable storage medium.

[0130] The storage medium mentioned above may be a read-only memory, a magnetic disk or an optical disk, etc. Although the embodiments of the present application have been shown and described above, it can be understood that the above embodiments are exemplary and cannot be understood as limiting the present application. A person of ordinary skill in the art may change, modify, replace and modify the above embodiments within the scope of the present application.

Claims

1. A method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation, characterized in that: The following steps are involved: Establishing a target metal segmentation model and surface current graph structure data corresponding to the target three-dimensional metal object; Based on the graph structure data, a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object is constructed, and based on a preset method of moments strategy, training data for the graph neural network solver is generated, and the graph neural network solver is trained by the training data; The metal segmentation model to be solved and multiple preset plane wave excitation parameters are input into the trained graph neural network solver to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver.

2. The method according to claim 1, characterized in that The step of establishing the target metal segmentation model and the graph structure data of the surface current corresponding to the target three-dimensional metal object includes: Establishing a joint integral equation between the equivalent current on the surface of the target three-dimensional metal object and the target plane wave excitation, and calculating the induced current on the surface of the target three-dimensional metal object by the target plane wave according to the joint integral equation and preset boundary conditions; Acquire multiple triangular facet metadata of the target three-dimensional metal object, and convert the spatial geometric structure corresponding to the target three-dimensional metal object into graph structure data according to the multiple triangular facet metadata; Converting the induced current and the equivalent current at the center of the triangle corresponding to each triangle face data in the plurality of triangle face data into a real scalar, so as to determine the characteristic vector of each node in the graph structure data according to the real scalar; Based on the eigenvector, graph structure data of the surface current is determined.

3. The method according to claim 2, characterized in that The method of constructing a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object based on the graph structure data, generating training data for the graph neural network solver based on a preset method of moments strategy, and training the graph neural network solver through the training data includes: Based on the feature vector and the induced current, construct an upsampling graph neural network module, a convolutional graph neural network module and a downsampling graph neural network module, and construct the graph neural network solver through the upsampling graph neural network module, the convolutional graph neural network module and the downsampling graph neural network module; Constructing a plurality of metal models corresponding to the target three-dimensional metal object, and segmenting the plurality of metal models to obtain segmentation data corresponding to each of the plurality of metal models; Using the segmented data to establish graph structure data corresponding to each metal model, and converting the induced current into a feature vector of each graph structure data node input into the up-sampling graph neural network module; Solving the equivalent current at the center of each triangle by a preset moment method, setting the equivalent current as a label, and constructing a feature vector of the graph structure data of the label according to the label; The graph structure data is input into the graph neural network solver to iteratively train the graph neural network solver, and the graph structure data of the network predicted equivalent current in each iteration process is output, and the mean square error between the feature vector of each node corresponding to the graph structure data of the network predicted equivalent current and the label feature vector is calculated, until a graph neural network solver whose mean square error meets the preset accuracy requirements is obtained.

4. The method according to claim 3, characterized in that The step of inputting the metal segmentation model to be solved and a plurality of preset plane wave excitation parameters into the trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver, comprises: Acquire the analysis data corresponding to the metal dissection model to be solved, and establish the graph structure data corresponding to the metal dissection model to be solved according to the analysis data; The multiple plane wave excitation parameters are converted into feature vectors of corresponding graph structure data, and the graph structure data is input into the trained graph neural network solver to output the solved graph structure data, and the surface current corresponding to the target three-dimensional metal object is obtained according to the feature vector of the solved graph structure data.

5. A device for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation, characterized in that: include: A modeling module, used to establish a target metal segmentation model and surface current graph structure data corresponding to a target three-dimensional metal object; A training module, used to construct a graph neural network solver corresponding to the joint integral equation of the target three-dimensional metal object based on the graph structure data, generate training data for the graph neural network solver based on a preset method of moments strategy, and train the graph neural network solver through the training data; A solution module is used to input the metal segmentation model to be solved and multiple preset plane wave excitation parameters into a trained graph neural network solver, so as to calculate the surface current corresponding to the target three-dimensional metal object through the trained graph neural network solver.

6. The device according to claim 5, characterized in that The modeling module includes: A first calculation unit is used to establish a joint integral equation between an equivalent current on the surface of the target three-dimensional metal object and a target plane wave excitation, and calculate an induced current on the surface of the target three-dimensional metal object by the target plane wave according to the joint integral equation and preset boundary conditions; A conversion unit, used to obtain a plurality of triangular face data of the target three-dimensional metal object, and convert a spatial geometric structure corresponding to the target three-dimensional metal object into graph structure data according to the plurality of triangular face data; A first determining unit is used to convert the induced current and the equivalent current at the center of the triangle corresponding to each triangle face data in the plurality of triangle face data into a real scalar, so as to determine the feature vector of each node in the graph structure data according to the real scalar; The second determining unit is used to determine the graph structure data of the surface current based on the characteristic vector.

7. The device according to claim 6, characterized in that The training module includes: A construction unit, configured to construct an upsampling graph neural network module, a convolutional graph neural network module and a downsampling graph neural network module based on the feature vector and the induced current, and to construct the graph neural network solver through the upsampling graph neural network module, the convolutional graph neural network module and the downsampling graph neural network module; A segmentation unit, used for constructing a plurality of metal models corresponding to the target three-dimensional metal object, and segmenting the plurality of metal models to obtain segmentation data corresponding to each of the plurality of metal models; An establishing unit, used to establish graph structure data corresponding to each metal model using the segmented data, and convert the induced current into a feature vector of each graph structure data node input into the up-sampling graph neural network module; A setting unit, configured to solve the equivalent current at the center of each triangle by a preset moment method, set the equivalent current as a label, and construct a feature vector of the graph structure data of the label according to the label; An iteration unit is used to input the graph structure data into the graph neural network solver to iteratively train the graph neural network solver, and output the graph structure data of the network predicted equivalent current in each iteration process, and calculate the mean square error between the feature vector of each node corresponding to the graph structure data of the network predicted equivalent current and the label feature vector, until a graph neural network solver whose mean square error meets the preset accuracy requirement is obtained.

8. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the program to implement the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation as described in any one of claims 1 to 4.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: The program is executed by a processor to implement the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation as described in any one of claims 1 to 4.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed to implement the method for solving the surface current of a three-dimensional metal object under electromagnetic wave irradiation as described in any one of claims 1 to 4.

Images