Prediction method for unit scattered field of variable-curvature frequency selective surface

Abstract

The invention discloses a method for predicting a scattered field of a variable-curvature frequency selective surface unit, which comprises the following steps of: performing logarithmic sampling on the curvature of the frequency selective surface unit, generating a curvature sampling point sample set, establishing a frequency selective surface unit model library, and performing simulation solution respectively to obtain near-field data of the frequency selective surface unit with different curvatures; solving the infinitesimal dipole position vector matrix and dipole moments in the infinitesimal dipole group to obtain dipole moment data sets of different units; building a long-short-term memory deep neural network model, adjusting hyper-parameters of the neural network model, and training the neural network model; and predicting the dipole moment, and solving the scattered field of the frequency selective surface unit in combination with the infinitesimal dipole group position vector matrix to complete prediction of the scattered field of the frequency selective surface unit. According to the method, the frequency selective surface unit scattering field with any curvature characteristic can be accurately calculated, the design efficiency of the frequency selective surface is improved, and the method has important engineering application value for large-scale frequency selective surface scattering calculation.

Description

Technical Field

[0001] The present invention belongs to the technical field of radomes, and particularly relates to a method for predicting the scattering field of a variable-curvature frequency selective surface unit, which can be used to assist in the calculation and analysis of the scattering field of a large-scale frequency selective surface. Background Art

[0002] A frequency selective surface is a large-scale structure composed of exactly the same frequency selective units arranged in a two-dimensional periodic manner. Due to its spatial filtering characteristics, it is widely used in military and wireless communication systems. However, due to its complex periodic structure and electrically large size, traditional low-frequency algorithms (such as the method of moments) often face challenges of scarce computing resources and high computing costs when dealing with such objects. Regarding the calculation problem of the scattering characteristics of a frequency selective surface, some scholars have proposed that the scattering field can be calculated by taking the frequency selective unit as the basic unit for array calculation, and the effect is remarkable. In this method, the calculation problem of the scattering field of the frequency selective unit becomes very important.

[0003] For a structure with a mixed metal-dielectric property such as a frequency selective unit, there are two methods in the related research field for calculating its scattering field: numerical methods and equivalent methods. Numerical calculation methods such as: the method of moments, finite difference time domain method, and finite element method, etc. These algorithms often have too many mesh division numbers when calculating the above objects, resulting in too high a calculation cost. Different from the traditional grid-based calculation methods, the infinitesimal dipole model equivalent method (IDM equivalent method) uses a group of infinitesimal dipoles to equivalent the current distribution on the model surface. This method can ignore the shape characteristics of the object and has great advantages in solving the scattering calculation problem of objects such as frequency selective units. Some foreign research results have shown the superiority of this algorithm. Mikki introduced the Quantum Particle Swarm Optimization (QPSO) algorithm in the equivalent process. In a large number of attempts of the equivalent model, the QPSO optimization algorithm reduced the degree of freedom of each dipole from ten in the genetic algorithm to seven. Yang optimized the seven parameters to two parameters by restricting the position and direction of a group of IDMs, and optimized the parameters with the near-field data as the target, and finally successfully predicted the scattering field results.

[0004] In practical applications, the frequency selective unit is initially designed in a planar state. However, when the frequency selective unit is applied to a large-scale frequency selective surface (such as a radome), the shape of the frequency selective unit changes from a plane to an irregular shape, which brings a lot of trouble to the designers. The repeated modeling and simulation analysis greatly reduces the work efficiency.

[0005] Therefore, it is of great significance to propose a method for predicting the scattering field of a variable-curvature frequency selective surface unit in the engineering application of large-scale frequency selective surface scattering calculation. Summary of the Invention

[0006] In view of the above problems, the present invention proposes a method for predicting the scattering field of a frequency selective surface (FSS) unit with variable curvature. By building a neural network model, this method can accurately calculate the scattering field of FSS units with arbitrary curvature characteristics, which is of great significance for the scattering calculation of large-scale FSSs.

[0007] The present invention is realized through the following technical solutions.

[0008] A method for predicting the scattering field of a frequency selective surface unit with variable curvature provided by an embodiment of the present invention includes the following steps:

[0009] Perform logarithmic sampling on the curvature of the frequency selective surface unit and convert it into the original space curvature to generate a curvature sampling point sample set;

[0010] Establish a frequency selective surface unit model library according to the curvature sampling point sample set;

[0011] Perform simulation solutions on the established frequency selective surface unit model library respectively to obtain the near-field data of FSS units with different curvatures;

[0012] Solve the infinitesimal dipole vector matrix according to the frequency selective surface unit model library;

[0013] Solve the dipole moment in the infinitesimal dipole group according to the near-field data of FSS units with different curvatures and the infinitesimal dipole vector matrix to obtain the dipole moment data set of different units;

[0014] Build a long short-term memory (LSTM) deep neural network model according to the dipole moment data set;

[0015] Adjust the hyperparameters of the neural network model and train the neural network model according to the LSTM deep neural network model;

[0016] Predict the dipole moment according to the trained neural network model and solve the scattering field of the frequency selective surface unit by combining with the infinitesimal dipole group vector matrix to complete the prediction of the scattering field of the frequency selective surface unit.

[0017] Preferably, perform logarithmic sampling on the curvature of the frequency selective surface unit. For circular arcs, sample on the logarithmic scale and restore the curvature on the logarithmic scale to the original space curvature sampling points to generate a sample set.

[0018] Preferably, establishing a frequency selective surface unit model library includes:

[0019] Let two orthogonal planes, the XOZ plane and the YOZ plane, serve as the main arc surface and the secondary arc surface respectively. The geometric shape within the main arc surface consists of two arcs and two line segments connected to form a closed surface; the secondary arc surface consists of one arc. The swept cross-section within the main arc surface scans along the arc trajectory within the secondary arc surface to generate a geometric body.

[0020] Preferably, perform simulation solutions on the established frequency selective surface element model library respectively, including:

[0021] Set observation points, plane wave forms, and solution options for the established frequency selective surface element solid model in a full-wave simulation software to solve the near-field data of frequency selective surface elements with different curvatures.

[0022] Preferably, solve the infinitesimal dipole position vector matrix according to the frequency selective surface element model library, including:

[0023] Discretize the arc within the YOZ plane into nodes to obtain the corresponding position information of the dipoles;

[0024] Rotate the coordinates of the dipoles on the arc within the YOZ plane along the X-axis to obtain the position coordinates of the dipoles on the frequency selective unit surface;

[0025] Place three orthogonally arranged dipoles at each position, and organize the desired information and position information into a position vector matrix.

[0026] Preferably, solve the dipole moments in the infinitesimal dipole group to obtain the dipole moment data sets of different units, including:

[0027] Combine with the free space dyadic Green's function, based on the expression representing the scattered near-field of the infinitesimal dipole model, extract the common factor in the scattered field expression of the infinitesimal dipole model, and rewrite it in the form of the product of the dipole moment of the infinitesimal dipole and the position vector matrix;

[0028] Perform singular value decomposition on the position vector matrix;

[0029] Solve the dipole moments of the infinitesimal dipole group so that the field calculation results of the infinitesimal dipole group are consistent with the full-wave simulation results, and organize the dipole moments of different units in the sample set into a dipole moment data set.

[0030] Preferably, build a long short-term memory deep neural network model, including:

[0031] The long short-term memory deep neural network model includes an input layer and a hidden layer. The input layer contains 2 time steps, the dimension of the input features for each time step is 1, and the number of samples participating in each training is 50; the hidden layer has 2 layers, the number of neurons in each layer is 256, and the Dropout ratio between the LSTM layers is 0.3 to prevent overfitting. The number of neurons in the output layer is 150, which is the same as the dimension of the final regression target.

[0032] Preferably, adjusting the hyperparameters of the neural network model and training the neural network model includes:

[0033] First, normalize the data in the dataset. Since the data in the dataset is complex, it is divided into a real part dataset and an imaginary part dataset for separate training, and then divided into a training set, a validation set, and a test set with proportions of 7:2:1 respectively; label the dataset with the curvature features of each sample as labels.

[0034] Train the neural network model. Set the learning rate during the training process to 0.0001, and decay the learning rate to 0.85 of the original every 65 steps; use early stopping logic to stop training if the loss value does not decrease within 50 steps.

[0035] Preferably, solving the scattering field of the frequency selective surface unit includes:

[0036] Use the trained neural network model as a database, denormalize the predicted generated data, and then combine it with the infinitesimal dipole vector matrix to form a predicted dipole group.

[0037] Substitute the predicted dipole group into the calculation formula of the radiation pattern of the frequency selective unit to obtain the scattering field of the frequency selective surface unit.

[0038] Due to the above technical solutions adopted by the technology of the present invention, the following beneficial effects are achieved:

[0039] 1. The present invention adopts the infinitesimal dipole model method, which can calculate the scattering field results of frequency selective units, improve the equivalent accuracy, and solve the problems of long equivalent time and high computational resource requirements in the traditional infinitesimal dipole model equivalent method.

[0040] 2. The present invention proposes a method for reconstructing the frequency selective surface unit model, and parametrically generates the required model through the curve sweeping method, overcoming the disadvantage of difficult acquisition of array elements in the variable curvature frequency selective surface, and expanding the application of the infinitesimal dipole model method in the scattering calculation of variable curvature frequency selective surfaces.

[0041] 3. The present invention establishes a long short-term memory deep neural network and an end-to-end network model between the curvature of the frequency selective surface unit and the dipole moment, avoiding the repetitive modeling and equivalent work in the infinitesimal dipole model method and improving the design efficiency of the frequency selective surface. BRIEF DESCRIPTION OF THE DRAWINGS

[0042] Figure 1 is a flowchart of a prediction method for the scattering field of a variable curvature frequency selective surface unit of the present invention;

[0043] Figure 2 is the modeling process of the intermediate frequency selection unit in the present invention;

[0044] Figure 3 is the equivalent idea of the equivalent method of the infinitesimal dipole model in the present invention;

[0045] Figure 4 The basic information of the frequency selection unit case used in the present invention;

[0046] Figure 5 is the comparison result of the real part prediction of the dipole moment of the dipole group of frequency selection unit 1;

[0047] Figure 6 is the comparison result of the real part prediction of the dipole moment of the dipole group of frequency selection unit 2;

[0048] Figure 7 is the comparison result of the imaginary part prediction of the dipole moment of the dipole group of frequency selection unit 1;

[0049] Figure 8 is the comparison result of the imaginary part prediction of the dipole moment of the dipole group of frequency selection unit 2;

[0050] Figure 9 is the comparison of the scattering pattern of the simulation results of the present invention and the commercial software FEKO for frequency selection unit 1;

[0051] Figure 10 is the comparison of the scattering pattern of the simulation results of the present invention and the commercial software FEKO for frequency selection unit 2. Specific implementation manner

[0052] The present invention will be further described in detail below with reference to the drawings and embodiments, but it shall not be used as a basis for any limitation to the present invention.

[0053] Refer to Figure 1 , the present invention is a flowchart of a method for predicting the scattering field of a variable curvature frequency selective surface unit, and the specific steps are as follows:

[0054] Step 1, generate a sample set according to the logarithmic sampling principle

[0055] According to the logarithmic sampling principle, logarithmically sample the curvature of the frequency selective surface unit and convert it into the original space curvature to generate a curvature sampling point sample set, including the following steps:

[0056] (1a) Sample the curvature of the frequency selective surface unit on the logarithmic scale, and the curvature z in the logarithmic scale is

[0057]

[0058] where x min and x maxare the lower and upper limits of the curvature sampling interval respectively, z is the curvature sampling point in the logarithmic scale, N is the number of sampling points, and i is the sampling sequence number.

[0059] (1b) Restore the curvature in the logarithmic scale to the original space curvature sampling point. The curvature x in the original space is

[0060] x = exp(z)

[0061] where x is the curvature sampling point obtained by mapping the curvature in the logarithmic scale back to the original space.

[0062] Step 2: Establish a variable curvature frequency selective unit model library

[0063] Establish arbitrary-shaped frequency selective unit models according to different curvatures, including the following steps:

[0064] Let the two orthogonal planes XOZ and YOZ be the main circular arc surface and the secondary circular arc surface respectively. The geometric shape in the main circular arc surface is composed of two circular arcs and two line segments connecting to form a closed surface. The radii of curvature of the circular arcs are R x and R x -h; the secondary circular arc surface is composed of one circular arc with a radius of curvature of R y . The swept cross-section in the main circular arc surface scans along the circular arc trajectory in the secondary circular arc surface to generate a geometric body.

[0065] Step 3: Conduct near-field simulation analysis on the unit

[0066] Set the observation points, plane wave form, and solution options in the full-wave simulation software for the established frequency selective surface unit solid model library, and perform simulation to solve the near-field data of frequency selective surface units with different curvatures respectively.

[0067] Step 4: Generate the dipole position vector matrix

[0068] Solve the dipole moment in the infinitesimal dipole group according to the frequency selective unit model library established in Step 2, including the following steps:

[0069] (4a) Discretize the circular arc in the YOZ plane into nodes to obtain the corresponding position information of the dipoles; the coordinates of the discrete points on the circular arc are as follows:

[0070]

[0071] where are the coordinates of the discrete points on the circular arc; r is the curvature of the circular arc; θ i is the discrete angle of the circular arc.

[0072] (4b) Rotate the dipole coordinates on the circular arc in the YOZ plane along the X-axis to obtain the dipole position coordinates on the frequency selective unit surface

[0073]

[0074] Among them, are the dipole coordinates on the surface of the frequency-selective unit; is the rotation matrix along the X-axis; φ j is the discrete angle of the inner arc of the secondary arc.

[0075] (4c) Place three dipoles at each position, and the three dipoles are placed orthogonally. Organize the pointing information and position information into the position vector matrix C.

[0076] C = [α, β, x, y, z]

[0077] Among them, α is the angle between the dipole and the X-axis, and the dipoles pointing to the X, Y, and Z axes are set to 0°, 90°, and 90° respectively; β is the angle between the dipole and the Y-axis, and the dipoles pointing to the X, Y, and Z axes are set to 90°, 0°, and 90° respectively; x, y, z are the position coordinates of the dipole in free space.

[0078] Step 5, solve the dipole moment in the infinitesimal dipole group

[0079] According to the near-field data results of the surface unit selected by different curvature frequencies obtained in Step 3 and the dipole position vector matrix obtained in Step 4, solve the dipole moment in the infinitesimal dipole group, including the following steps:

[0080] (5a) Derive the expression of the free-space dyadic Green's function as:

[0081]

[0082] Among them, is the identity matrix; k is the propagation coefficient; is the scalar Green's function; r and r' are the position vectors of the near-field observation point and the infinitesimal dipole respectively.

[0083] (5b) Combine the free-space dyadic Green's function and represent the expression of the scattered field based on the infinitesimal dipole model (IDM):

[0084]

[0085] Among them, j is the imaginary unit; ω is the angular frequency; μ is the magnetic permeability of free space; α i , β i , γ i are the angles between the i-th infinitesimal dipole and the X, Y, and Z axes respectively; x, y, z are the unit vectors in the X, Y, and Z axis directions; M i is the dipole moment of the infinitesimal dipole; i is the dipole number, N d is the total number of dipoles, and E is the scattered near-field result solved using the dipoles; is the dyadic Green's function.

[0086] (5c) Factor out the common factor from the scattering field expression of the infinitesimal dipole model and rewrite it in the form of the product of the dipole moment and the position vector matrix

[0087]

[0088] where E mask is the near-field scattering result in the full-wave simulation software; C i is the position vector matrix of the i-th dipole.

[0089] (5d) Perform singular value decomposition on the position vector matrix

[0090] C = UΣV T

[0091] where U is an m×n orthogonal matrix; V is an m×n orthogonal matrix; Σ is a diagonal matrix.

[0092] (5f) Solve for the dipole moment of the dipole group in the dataset

[0093] M = VΣ + U T E mask

[0094] where Σ + is the pseudo-inverse matrix of matrix Σ in step (5d).

[0095] Make the calculation results of the infinitesimal dipole group coincide with the simulation results, and organize the dipole moments of different units in the sample set into a dipole moment dataset.

[0096] Step 6, dataset processing

[0097] First, normalize the data in the dataset. The data in the dataset are complex numbers. Divide them into a real part dataset and an imaginary part dataset for separate training, and then divide them into a training set, a validation set, and a test set with proportions of 7:2:1; label the dataset with the curvature features of each sample as labels.

[0098] Step 7, construction of the long short-term memory deep neural network model

[0099] According to the dataset, construct a long short-term memory deep neural network model, including the following steps:

[0100] In the established LSTM network model, the input layer contains 2 time steps, the dimension of the input features at each time step is 1, and the number of samples participating in each training is 50; the hidden layer has 2 layers, with 256 neurons in each layer. The Dropout ratio between LSTM layers is 0.3 to prevent overfitting. The number of neurons in the output layer is 150, which is the same as the dimension of the final regression target.

[0101] Step 8: Train and validate the network model

[0102] (8a) The learning rate during the training process is set to 0.0001. In order to learn the features of the data, the learning rate decays to 0.85 of the original value every 65 steps; the loss function is MSE, and early stopping logic is used, that is, if the loss value does not decrease within 50 steps, the training is stopped to prevent the model from overfitting. The loss function MSE is as follows:

[0103]

[0104] where S represents the numerical result predicted by the neural network, represents the true numerical result.

[0105] (8b) After the model training is completed, the input features in the validation set are input into the neural network to verify the predicted data with the true data.

[0106] Step 9: Use the neural network model to predict the scattering field of the frequency selective unit

[0107] (9a) Take the trained neural network model as a database, denormalize the predicted data, and then combine it with the infinitesimal dipole vector matrix C to form a predicted dipole group.

[0108] (9b) Substitute the dipole group into the calculation formula of the scattering field pattern of the frequency selective unit to obtain the scattering field of the frequency selective surface unit.

[0109] The calculation formula of the scattering field pattern of the frequency selective unit is as follows:

[0110]

[0111] where E is the electric field strength at the scattering field observation point; f(θ, φ) represents the scattering field pattern of the frequency selective unit; (θ, φ) is the observation angle; I is the excitation current; R is the scattering field distance.

[0112] The advantages of the present invention can be further illustrated by the following simulation cases.

[0113] 1. Simulation parameters

[0114] The information of the frequency selective unit is in Figure 2As shown, the operating frequency is 8.56 GHz, and the curvature information of the main and secondary surfaces is given in the form of pairs of numbers. Two sets of results are selected for case analysis, and the curvatures of these units are (108, 145) and (600, 90) respectively.

[0115] 2. Simulation Contents and Results

[0116] Figure 2 It shows how the frequency selection unit in the present invention is built. Figure 3 It shows a schematic diagram of the infinitesimal dipole model equivalent method in the present invention. Figure 4 Two frequency selection units are given to verify the accuracy of the present invention, and the shape information is shown in the figure. Figure 5 and Figure 6 respectively show the comparison between the predicted real part values and the true values of Unit 1 and Unit 2; Figure 7 and Figure 8 respectively show the comparison between the predicted imaginary part values and the true values of Unit 1 and Unit 2. Table 1 shows the errors between the predictions and the true results of the two units, proving that the neural network established in the present invention has accurate prediction ability.

[0117] Table 1 Errors between the predictions and the true results of two frequency selective surface units

[0118]

[0119] Figure 9 and Figure 10 show the comparison between the scattering field predicted by using the network model for numerical calculation and the scattering field simulated by commercial software. It is consistent with the analysis result of the commercial software FEKO, verifying the accuracy of the method of the present invention.

[0120] The present invention is not limited to the above-mentioned embodiments. Based on the technical solutions disclosed in the present invention, those skilled in the art can make some substitutions and deformations to some of the technical features without creative labor according to the disclosed technical content, and these substitutions and deformations are all within the protection scope of the present invention.

Claims

1. A method for predicting the scattered field of a variable curvature frequency selective surface unit, characterized in that: The steps include: Perform logarithmic sampling on the frequency selective surface unit curvature and convert it into the original space curvature to generate a curvature sampling point sample set; Establishing a frequency selective surface unit model library according to the curvature sampling point sample set; The established frequency selective surface unit model library is simulated and solved respectively to obtain the near-field data of frequency selective surface units with different curvatures; Based on the frequency selective surface unit model library, the infinitesimal dipole position vector matrix is ​​solved; According to the near-field data of the surface unit and the infinitesimal dipole position-vector matrix of different curvature frequencies, the dipole moment in the infinitesimal dipole group is solved to obtain the dipole moment data sets of different units; Building a long short-term memory deep neural network model based on the dipole moment data set; According to the long short-term memory deep neural network model, adjusting the neural network model hyperparameters and training the neural network model; According to the trained neural network model, the dipole moment is predicted and combined with the infinitesimal dipole group position vector matrix to solve the scattering field of the frequency selective surface unit, thus completing the prediction of the scattering field of the frequency selective surface unit.

2. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: The frequency selective surface unit curvature is logarithmically sampled, the arc is sampled in a logarithmic scale, the curvature in the logarithmic scale is restored to the original space curvature sampling point, and a sample set is generated.

3. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: Establish a frequency selective surface unit model library, including: Let the two orthogonal planes XOZ and YOZ planes be the main arc surface and the secondary arc surface respectively. The geometric shape inside the main arc surface is a closed surface composed of two arc segments and two line segments; the secondary arc surface is composed of an arc segment, and the swept section inside the main arc surface is scanned along the arc trajectory inside the secondary arc surface to generate a geometric body.

4. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: The established frequency selective surface unit model library is simulated and solved separately, including: The constructed frequency selective surface unit entity model is used to set observation points, plane wave forms and solution options in the full-wave simulation software to solve the near-field data of frequency selective surface units with different curvatures.

5. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: Based on the frequency selective surface unit model library, solve the infinitesimal dipole position vector matrix, including: Discretize the arcs in the YOZ plane into nodes to obtain the corresponding position information of the dipoles; Rotate the dipole coordinates on the arc in the YOZ plane along the X axis to obtain the dipole position coordinates on the unit surface of the frequency selective surface; Three orthogonally arranged dipoles are placed at each position to organize the directional information and position information into a position vector matrix.

6. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: Solve for the dipole moment in a set of infinitesimal dipoles and obtain dipole moment data sets for different units, including: Combined with the free space and the Green's function, based on the infinitesimal dipole model to express the scattered near field, the common factors in the scattered field expression of the infinitesimal dipole model are extracted and rewritten as the multiplication of the dipole moment of the infinitesimal dipole and the position vector matrix; Perform singular value decomposition on the position matrix; The dipole moment of the infinitesimal dipole group is solved so that the field calculation results of the infinitesimal dipole group are consistent with the full-wave simulation results, and the dipole moments of different units in the sample set are organized into a dipole moment data set.

7. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 6, characterized in that: The scattered field expression of the infinitesimal dipole model is as follows: Where j is a complex unit; ω is the angular frequency; μ is the free space permeability; α i , β i , γ i are the angles between the ith infinitesimal dipole and the X, Y, and Z axes respectively; x, y, and z are the unit vectors in the X, Y, and Z directions respectively; M i is the dipole moment of the infinitesimal dipole; i is the number of the dipole, N d is the total number of dipoles, E is the scattered near-field result solved using dipoles; is the dyadic Green's function; Extract the common factors of the scattered field expression of the infinitesimal dipole model and rewrite it into the form of multiplication of the dipole moment and the position vector matrix: Among them, E mask is the scattered near-field result in the full-wave simulation software; C i is the position matrix of the ith dipole; Solve for the dipole moment of an infinitesimal set of dipoles: M=VΣ + The T And mask Among them, U is an m×n orthogonal matrix; V is an m×n orthogonal matrix; Σ is a diagonal matrix, Σ + is the pseudo-inverse of the diagonal matrix Σ.

8. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: Build a long short-term memory deep neural network model, including: The long short-term memory deep neural network model includes an input layer and a hidden layer. The input layer contains 2 time steps. The dimension of the input feature of each time step is 1, and the number of samples involved in each training is 50; there are 2 hidden layers, and the number of neurons in each layer is 256. The Dropout ratio between LSTM layers is 0.3 to prevent overfitting. The number of neurons in the output layer is 150, which is the same as the dimension of the final regression target.

9. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: Adjusting the neural network model hyperparameters and training the neural network model includes: The data in the dataset are first normalized. The data in the dataset are complex numbers, which are divided into real and imaginary data sets, and then divided into training, validation and test sets, with a ratio of 7:2:1 respectively. The curvature features of each sample are used as labels to annotate the dataset. The neural network model was trained with the learning rate set to 0.0001 during training, and the learning rate decayed to the original 0.85 every 65 steps. The early stopping logic was used and the training was stopped if the loss value did not decrease within 50 steps.

10. The method for predicting the scattered field of a variable curvature frequency selective surface unit according to claim 1, characterized in that: Solve the scattered field of the frequency selective surface unit, including: The trained neural network model is used as a database, the predicted data is denormalized, and then combined with the infinitesimal dipole position vector matrix to form a predicted dipole group; Substitute the predicted dipole group into the calculation formula of the frequency selective unit scattering field pattern to obtain the frequency selective surface unit scattering field; The calculation formula of the scattered field pattern of the frequency-selective unit is as follows: Where E is the electric field intensity at the observation point of the unit scattering field; f(θ, φ) represents the frequency-selective unit scattering field pattern; (θ, φ) is the observation angle; I is the excitation current; j is the complex unit; k is the propagation coefficient; and R is the scattering field distance.

Images