Metamaterial wave absorber construction method and device based on deep learning and proxy optimization

Abstract

The invention discloses a metamaterial wave absorber construction method and device based on deep learning and proxy optimization. The method comprises the following steps: constructing a plurality of metamaterial structures; calling a CST-MATLAB-API interface, and controlling CST software to perform electromagnetic field simulation on the plurality of metamaterial structures to obtain a three-dimensional structure diagram of each metamaterial and S parameters under different frequencies; performing inversion on the S parameters under different frequencies to obtain the absorptivity of each metamaterial under different frequencies, converting the design layer structure of the three-dimensional structure diagram into a two-dimensional grey-scale map, taking the absorptivity and the two-dimensional grey-scale map as a group of data, and repeating the steps to obtain N groups of data; constructing a deep learning model by using the N groups of data, taking the two-dimensional grey-scale map as input and taking the corresponding group of absorptivity as output; and based on a deep learning model, taking the longest interval length continuously greater than a threshold value in a group of absorptivity as a fitness function, and utilizing a proxy optimization method to optimize the two-dimensional grey-scale map to obtain an optimal structure of the metamaterial.

Description

Technical Field

[0001] The present application relates to the technical field of electromagnetic wave absorption material design, and particularly to a method and device for constructing a metamaterial absorber based on deep learning and surrogate optimization. Background Art

[0002] With the increasing dependence of modern society on electronic devices, the level of electromagnetic radiation has risen significantly. Exploring the high efficiency of metamaterial absorbers is of great significance for facing the increasingly complex electromagnetic application environment. Due to the unique characteristics of terahertz, terahertz technology has been widely studied and developed. Among them, terahertz metamaterial absorbers are popular functional devices in recent years and have broad application prospects in the fields of terahertz imaging, sensors, and stealth. However, most terahertz absorbers have single functions, low design efficiency, and often poor design effects. Therefore, researchers have extensive attention to the study of efficient and high-performance absorbers.

[0003] When designing metamaterials, the traditional method of manually designing metamaterial absorbers often has the disadvantages of lack of flexibility, low efficiency, and difficulty in dealing with complex structures. In recent years, artificial intelligence technology has achieved remarkable research results in the field of terahertz metamaterial absorber devices. People have proposed artificial intelligence methods such as artificial intelligence networks and deep neural networks to assist in the design of metamaterials. Machine learning, a data-driven method, has the advantages of efficient optimization, rapid design, breaking through complexity limitations, generating diverse structures, and reducing empirical dependence in the design of metamaterials. Therefore, it is of great significance to design metamaterials based on machine learning.

[0004] Patent [202210840401.2] discloses a method for constructing a metamaterial absorber, which uses a multi-objective optimization algorithm to optimize the design to achieve the trade-off between the absorber bandwidth and the material thickness, so as to improve the performance of the metamaterial absorber. Disadvantages: First, this method does not clearly explain the selection basis of the threshold and setting value, making this setting may lack theoretical support, resulting in insufficient adaptability of the optimization result in practical applications; second, in the process of constructing the metamaterial absorber, this method relies on an equivalent homogeneous medium model to approximate the electromagnetic characteristics of the metamaterial. This simplified treatment ignores the inhomogeneity and anisotropy characteristics of the internal structure of the metamaterial, which is likely to affect the accuracy and reliability of the design result. Finally, this method performs multi-objective optimization on the material thickness and absorption bandwidth, resulting in only a suboptimal solution after the trade-off between the material thickness and absorption bandwidth, and it is impossible to optimize and obtain the optimal material thickness under the specified absorption bandwidth.

[0005] Patent [202410021869.8] discloses a broadband all-dielectric three-dimensional electromagnetic metamaterial, a design method and an application. Disadvantages: First, in the optimization using the genetic algorithm, the CST needs to be called for simulation in each iteration to obtain the fitness function value of each generation, resulting in low optimization efficiency and difficulty in optimizing and solving complex structures; Second, encoding the three-dimensional structure into a one-dimensional sequence as the input of the neural network cannot extract the spatial information between the blocks of the electromagnetic metamaterial, and serialization may introduce pseudo-patterns irrelevant to the electromagnetic performance (such as accidental correlations in a specific arrangement order), interfering with the model's modeling of the real physical mechanism, and the model may be unstable for unseen three-dimensional variants (such as rotated and mirrored structures); Finally, this method uses a single performance index (such as the frequency range with reflectivity less than -10 dB) as the fitness function, overly relying on algorithm generation, and may ignore the multi-performance coupling characteristics of the metamaterial in a complex electromagnetic environment, resulting in insufficient comprehensive performance of the optimization result.

[0006] Patent [201910501055.3] discloses a design method for a metamaterial absorber structure based on a neural network, using the neural network to replace the full-wave simulation software. Disadvantages: First, when collecting data, CST is directly used, and parameters are set manually one by one and simulated, resulting in low design efficiency; Second, using a single-layer neural network, for complex problems such as metamaterial absorbers, it may not be able to generalize well and can only be applicable to the parameter range of the given data set; Finally, using the particle swarm optimization algorithm, this method is only applicable to the case of few input parameters. When the parameters are increased, the optimization efficiency may decrease significantly and it is easy to fall into local optima, with poor generalization. Summary of the Invention

[0007] The purpose of the embodiments of this application is to provide a method and device for constructing a metamaterial absorber based on deep learning and surrogate optimization to solve the technical problems of low design efficiency, reliance on experience, lack of data support, and time-consuming optimization existing in the related technologies.

[0008] According to the first aspect of the embodiments of this application, a method for constructing a metamaterial absorber based on deep learning and surrogate optimization is provided, including: S1: Implement a parametric randomization generation strategy in MATLAB to construct multiple metamaterial structures; S2: Control the CST software to perform electromagnetic field simulation on the multiple metamaterial structures through the MATLAB-CST-MATLAB-API interface to obtain the three-dimensional structure diagram of each metamaterial and the S parameters at different frequencies; S3: Use MATLAB to invert the S-parameters at different frequencies to obtain the absorption rate of each metamaterial at different frequencies. Convert the design layer structure of the three-dimensional structure diagram into a two-dimensional grayscale image. Use the absorption rate and the two-dimensional grayscale image as a set of data, and repeat S1 - S2 to obtain N sets of data; S4: Use the N sets of data, with the two-dimensional grayscale image as the input and its corresponding set of absorption rates as the output, to build and train a deep learning model; S5: Based on the deep learning model, use the longest interval length in a set of absorption rates that is continuously greater than the threshold as the fitness function, and use the surrogate optimization method to optimize the two-dimensional grayscale image to obtain the optimal structure of the metamaterial.

[0009] According to the second aspect of the embodiments of the present application, there is provided a device for constructing a metamaterial absorber based on deep learning and surrogate optimization, including: A construction module, configured to implement a parameterized randomization generation strategy in MATLAB to construct multiple metamaterial structures; A simulation module, configured to control the CST software to perform electromagnetic field simulation on the multiple metamaterial structures by calling the CST-MATLAB-API interface through MATLAB, and obtain the three-dimensional structure diagram and S-parameters at different frequencies of each metamaterial; A data generation module, configured to use MATLAB to invert the S-parameters at different frequencies to obtain the absorption rate of each metamaterial at different frequencies, convert the design layer structure of the three-dimensional structure diagram into a two-dimensional grayscale image, and use the absorption rate and the two-dimensional grayscale image as a set of data. Repeat the construction module - simulation module to obtain N sets of data; A model building module, configured to use the N sets of data, with the two-dimensional grayscale image as the input and its corresponding set of absorption rates as the output, to build and train a deep learning model; An optimization module, configured to based on the deep learning model, use the longest interval length in a set of absorption rates that is continuously greater than the threshold as the fitness function, and use the surrogate optimization method to optimize the two-dimensional grayscale image to obtain the optimal structure of the metamaterial.

[0010] According to the third aspect of the embodiments of the present application, there is provided an electronic device, including: One or more processors; A memory, configured to store one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method as described in the first aspect.

[0011] According to a third aspect of the embodiments of the present application, a computer-readable storage medium is provided, on which computer instructions are stored, and when the instructions are executed by a processor, the steps of the method described in the first aspect are implemented.

[0012] The technical solutions provided by the embodiments of the present application may include the following beneficial effects: As can be seen from the above embodiments, the present invention adopts deep learning and surrogate optimization technical means, overcoming technical problems such as the traditional metamaterial design relying on prior knowledge, low efficiency, insufficient optimization ability, and lack of large-scale data support. Through the CST-MATLAB-API interface, the collection of the data set is completed, solving the problem of low efficiency in the traditional artificial collection of the data set. Through the deep learning model, it can efficiently process complex structural design spaces, breaking through the limitations of existing methods relying on trial-and-error methods or single optimization algorithms, avoiding structural homogenization, and meeting the optimization requirements of multiple performance indicators. At the same time, using the surrogate optimization method significantly reduces the computational cost and time consumption in the optimization process, solves the efficiency bottleneck of existing optimization methods in high-dimensional discrete problems, improves the automation degree of the design process, reduces manual parameter adjustment, and enhances the design effect. In addition, the deep learning model is trained based on large-scale data, establishing an effective design model, reducing the dependence on expert experience and prior knowledge, and further improving the design efficiency and optimization ability.

[0013] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0014] The drawings here are incorporated into the specification and constitute a part of this specification, showing embodiments consistent with the present application, and are used together with the specification to explain the principles of the present application.

[0015] Figure 1 is a flowchart of a method for constructing a metamaterial absorber based on deep learning and surrogate optimization shown according to an exemplary embodiment.

[0016] Figure 2 is an overall example diagram of a three-dimensional structure shown according to an exemplary embodiment.

[0017] Figure 3 is an example of a two-dimensional grayscale image of the input data set of deep learning shown according to an exemplary embodiment.

[0018] Figure 4 is a structural diagram of a deep learning model shown according to an exemplary embodiment.

[0019] Figure 5 is an example of the fitting effect diagram of a deep learning model shown according to an exemplary embodiment.

[0020] Figure 6 It is an agent optimization iteration diagram shown according to an exemplary embodiment.

[0021] Figure 7 It is a reverse design result diagram shown according to an exemplary embodiment.

[0022] Figure 8 It is a schematic structural diagram of a device for constructing a metamaterial absorber based on deep learning and agent optimization shown according to an exemplary embodiment. Detailed implementation manners

[0023] Here, the exemplary embodiments will be described in detail, and the examples are shown in the drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementation manners described in the following exemplary embodiments do not represent all implementation manners consistent with the present application. On the contrary, they are merely examples of devices and methods consistent with some aspects of the present application as detailed in the appended claims.

[0024] The terms used in the present application are only for the purpose of describing specific embodiments and are not intended to limit the present application. The singular forms "a", "the", and "said" used in the present application and the appended claims are also intended to include the plural forms unless the context clearly indicates otherwise. It should also be understood that the term "and / or" used herein refers to and includes any or all possible combinations of one or more of the associated listed items.

[0025] Figure 1 It is a flowchart of a method for constructing a metamaterial absorber based on deep learning and agent optimization shown according to an exemplary embodiment. As Figure 1 shown, the method may include the following steps: S1: Implement a parameterized randomization generation strategy in MATLAB to construct multiple metamaterial structures; S2: Call the CST-MATLAB-API interface through MATLAB to control the CST software to perform electromagnetic field simulation on the multiple metamaterial structures, and obtain the three-dimensional structure diagram of each metamaterial and the S parameters at different frequencies; S3: Use MATLAB to invert the S parameters at different frequencies to obtain the absorption rate of each metamaterial at different frequencies, convert the design layer structure of the three-dimensional structure diagram into a two-dimensional grayscale diagram, and use the absorption rate and the two-dimensional grayscale diagram as a set of data. Repeat S1 - S2 to obtain N sets of data; S4: Use the N sets of data, with the two-dimensional grayscale diagram as the input and the corresponding set of absorption rates as the output, to build and train a deep learning model; S5: Based on the deep learning model, using the length of the longest continuous interval greater than the threshold in a set of absorption rates as the fitness function, the two-dimensional grayscale image is optimized by using a surrogate optimization method to obtain the optimal structure of the metamaterial.

[0026] As can be seen from the above embodiments, the present application effectively solves multiple technical bottlenecks in traditional metamaterial design through deep learning and surrogate optimization technical means, combined with the CST-MATLAB-API interface. First, the CST-MATLAB-API interface is used to realize the automatic collection of data, avoiding the problem of low efficiency in manually collecting data in traditional methods. At the same time, combined with the deep learning model, it can efficiently process complex structure design spaces, breaking through the limitations of existing methods that rely on trial-and-error methods or single optimization algorithms. In addition, the deep learning model trained based on large-scale data reduces the dependence on expert experience and prior knowledge and establishes a more effective design model. The application of the surrogate optimization method significantly reduces the computational cost and time consumption in the optimization process and solves the efficiency bottleneck of existing optimization methods in high-dimensional discrete problems. Through the above technical solutions, the present invention not only improves the efficiency and accuracy of metamaterial design, but also reduces the design cost, enhances the generalization ability and adaptability of the design model, and provides strong technical support for the wide application of metamaterials.

[0027] The following elaborates on each step in detail.

[0028] S1: Implement a parameterized randomization generation strategy in MATLAB to construct multiple metamaterial structure parameters; this step includes the following sub-steps: S11: Set the simulation configuration of the metamaterial structure in MATLAB, and the simulation configuration includes operating frequency, background material, dielectric material, boundary conditions, global unit, and solver; Specifically, in MATLAB, the operating frequency of the designed metamaterial is configured in detail to be 12 - 15 THz, the boundary conditions are set such that the x-axis direction and the y-axis direction are set as unit cells, the z-axis is set as open (add space), the global unit is μm, and the solver is set as a frequency solver and other key simulation parameters.

[0029] The simulation configuration should set the boundary conditions first and then other parameters, and such setting helps to ensure the consistency of model parameters and simulation conditions and improve simulation efficiency.

[0030] S12: Under the above simulation configuration, build a basic model of a three-dimensional structure. Based on the basic model, randomly create n*n required structural slices, that is, n*n square units, on the two-dimensional surface to be designed using a MATLAB program. Each unit has two cases of being filled or not filled. Repeat the random creation of n*n required structural slices to complete the creation of multiple metamaterial structures.

[0031] Specifically, the CST software is programmatically called through a MATLAB script to complete the random design of metamaterials and perform accurate electromagnetic field simulation analysis on the electromagnetic metamaterials. The basic model is designed with three layers. As Figure 2 shown, the bottom layer is a metal plate with a thickness of 0.2 μm and a period of 36 μm, and the material is Copper (pure); the second layer is a dielectric layer with a thickness of 18 μm, and the relative dielectric constant of this material is 4.4; the third layer is the structure layer to be designed. In this structure layer, 1 - 36 metal sheets are randomly created using a MATLAB program. The size of the metal sheets is 6×6 μm, and the filling rule is a two-dimensional three-dimensional structure composed of 6×6×1 square units, arranged in an array. There are a total of 36 square units, and each unit has two cases: filled or unfilled. The code "0" is used to represent no square unit, and the code "1" is used to represent a square unit. They are arranged one by one from left to right and from front to back. These models are represented by codes for convenient co-simulation processing with MATLAB. The sequence example is as follows:

[0032] It should be noted that in the present invention, for the convenience of understanding, the basic model used is relatively simple, and other different basic models can also be realized.

[0033] S2: Call the CST - MATLAB - API interface through MATLAB to control the CST software to perform electromagnetic field simulation on the multiple metamaterial structure parameters, and obtain the three-dimensional structure diagram of each metamaterial and the S parameters at different frequencies; Specifically, use CST Studio Suite 2020 for simulation, set 251 frequency points between 12 - 15 THz, use CST to draw the electromagnetic field simulation result diagram, and import the three-dimensional structure diagram of the metamaterial and the S11 and S21 parameters in the simulation results back to MATLAB.

[0034] S3: Use MATLAB to invert the S parameters at different frequencies to obtain the absorption rate of each metamaterial at different frequencies, convert the designed layer structure of the three-dimensional structure diagram into a two-dimensional grayscale diagram, and use the absorption rate and the two-dimensional grayscale diagram as a set of data, and repeat S1 - S2 to obtain N sets of data; Specifically, use MATLAB to invert the S parameters at different frequencies according to the S parameter inversion formula to obtain the absorption rate of each metamaterial at different frequencies. The formula is as follows: 2 Among them, is the absorption rate, is the reflectivity, is the frequency.

[0035] Encode the layer structure of the three-dimensional structure diagram. Use "0" to represent no filling and "1" to represent filling. Arrange them one by one from left to right and from front to back to obtain a 6×6 matrix composed of "0" and "1", and convert the 6×6 matrix into a 6×6×1 two-dimensional grayscale image.

[0036] Take the absorption rate and the two-dimensional grayscale image as a set of data, and repeat S1 - S2 to obtain 15,000 sets of data.

[0037] S4: Use the N sets of data. Take the two-dimensional grayscale image as the input and the corresponding set of absorption rates as the output to build and train a deep learning model; S41: Take the two-dimensional grayscale image as the input and the corresponding set of absorption rates as the output response to build a convolutional neural network; Specifically, in the deep learning database, the input set is to take every 6 of the 6×6 0 / 1 sequences as a row to form a 6×6 two-dimensional array, and convert it into a 6×6×1 grayscale image. The example is as Figure 3 shown. The output set is the absorption rate of a set of 251 consecutive frequency points corresponding to this structure. As Figure 4 shown, use two convolutional layers and a fully connected layer as the hidden layer. Select ReLu as the activation function. The first convolutional layer contains two steps of convolution. The number of convolutional kernels is 60, the filter size is 3×3, and the stride is 1×1. After the convolutional layer, connect a max pooling layer. The pooling size is 2×2 and the stride is 1×1. Then connect the second convolutional layer, which contains two steps of convolution. The number of convolutional kernels is 60, the filter size is 1×1, and the stride is 1×1. Then connect the max pooling layer again. The pooling size is 2×2 and the stride is 1×1. After that, the fully connected layer connects to the output layer. Use sigmoid as the function of the output layer, and establish a deep learning model with the absorption rates at the 251 different frequencies as the output.

[0038] Selecting a 1×1 size filter for the second convolutional layer can perform weighted summation on all input channels without changing the number of channels, generate new channels, and use the ReLu activation function to further enhance the non-linear fitting ability of the network; since the output response absorption rates are all between 0 and 1, using sigmoid as the function of the output layer can enhance the output response characteristics and avoid the neuron death problem caused by multiple ReLus at the same time.

[0039] S42: Set the training parameters, import the N sets of data into the convolutional neural network, and train the convolutional neural network to obtain a deep learning model.

[0040] Specifically, the dataset is divided into a training set and a test set in a ratio of 7:3. The two-dimensional grayscale image of the metamaterial's upper surface structure is used as the input (1 for white and 0 for black), as Figure 4 shown. During model establishment, by continuously adjusting the training parameter settings, the optimal results are finally obtained with Solver being sgdm, MiniBatchSize being 60, MaxEpochs being 500, ValidationFrequency being 50, and LearnRate being 0.01. For the designed model, the training set loss is approximately 0.0188, RMSE is approximately 0.261, the test set loss is approximately 0.0312, and RMSE is approximately 0.268. Among them, the input 6×6 matrices are [1 1 1 0 1 1; 1 1 1 0 0 1; 1 1 1 1 0 1; 0 1 0 0 0 0; 0 0 0 0 0 0; 0 0 0 0 0 0] and [0 1 0 0 0 0; 0 0 0 1 0 0; 1 0 1 1 0 0; 0 0 0 1 0 1; 0 0 0 0 0 0; 0 0 0 0 0 0], and the corresponding fitting effect diagrams of a set of absorption rates are as Figure 5 shown in (a) and (b) of

[0041] S5: Based on the deep learning model, using the length of the longest interval in a set of absorption rates that is continuously greater than the threshold as the fitness function, and using the surrogate optimization method to optimize the two-dimensional grayscale image to obtain the optimal structure of the metamaterial, including the following sub-steps: S51: In MATLAB, use the numerical values of each pixel point of the two-dimensional grayscale image as the optimization variables, and use the length of the longest interval in a set of absorption rates that is continuously greater than the threshold as the fitness function; Specifically, based on the established deep learning model, optimize with the goal of maximizing the number of absorption rate values that are continuously greater than 0.65 at 251 frequency points.

[0042] S52: Set the surrogate optimization parameters, optimize the two-dimensional grayscale image, and obtain the optimal structure of the metamaterial through reverse optimization.

[0043] Specifically, the model is optimized using the surrogate optimization method. The parameters of the surrogate optimization method are set as follows: the number of iterations is set to 300, and the rest are kept as the default settings in MATLAB. Finally, the 6×6 matrix corresponding to the optimal structure has an input of [0 0 0 0 0 0; 0 0 0 0 0 0; 0 0 0 0 0 0; 0 0 1 0 0 0; 0 0 0 0 0 0; 0 0 0 0 0 0]. The optimal structure can achieve an absorption rate of 0.65 within the frequency range of 13.788 - 13.920 THz, and the highest absorption rate is 0.6882 at 13.848 THz, meeting the reverse design conditions, indicating that the model fitting and prediction effects are good. The iterative process is as Figure 6 shown. The function objective value converges after 180 iterations. The reverse design diagram is as Figure 7 , and it is compared with the true absorption rate curve under this model, and the two almost completely overlap.

[0044] Corresponding to the foregoing embodiments of the method for constructing a metamaterial absorber based on deep learning and surrogate optimization, the present application also provides an embodiment of a device for constructing a metamaterial absorber based on deep learning and surrogate optimization.

[0045] Figure 8 FIG. is a block diagram of a device for constructing a metamaterial absorber based on deep learning and surrogate optimization shown according to an exemplary embodiment. Referring to Figure 8 , the device includes: A construction module 1, configured to implement a parameterized randomization generation strategy in MATLAB to construct multiple metamaterial structures; A simulation module 2, configured to control the CST software to perform electromagnetic field simulation on the multiple metamaterial structures through the MATLAB call to the CST - MATLAB - API interface, and obtain the three - dimensional structure diagram of each metamaterial and the S parameters at different frequencies; A data generation module 3, configured to use MATLAB to invert the S parameters at different frequencies to obtain the absorption rate of each metamaterial at different frequencies, convert the design layer structure of the three - dimensional structure diagram into a two - dimensional grayscale image, and use the absorption rate and the two - dimensional grayscale image as a set of data, and repeat the construction module - simulation module to obtain N sets of data; A model building module 4, configured to use the N sets of data, with the two - dimensional grayscale image as the input and the corresponding set of absorption rates as the output, to build and train a deep learning model; An optimization module 5, configured to, based on the deep learning model, use the longest interval length of consecutive absorption rates greater than a threshold in a set of absorption rates as a fitness function, and use the surrogate optimization method to optimize the two - dimensional grayscale image to obtain the optimal structure of the metamaterial.

[0046] Regarding the device in the above embodiments, the specific manner in which each module performs operations has been described in detail in the embodiments related to the method, and will not be elaborated here.

[0047] For the device embodiments, since they basically correspond to the method embodiments, the relevant parts can refer to the descriptions in the method embodiments. The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed to multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this application. Those of ordinary skill in the art can understand and implement it without creative efforts.

[0048] Correspondingly, this application also provides an electronic device, including: one or more processors; a memory for storing one or more programs; when the one or more programs are executed by the one or more processors, the one or more processors implement the method for constructing a metamaterial absorber based on deep learning and surrogate optimization as described above.

[0049] Correspondingly, this application also provides a computer-readable storage medium, on which computer instructions are stored, and when the instructions are executed by a processor, the method for constructing a metamaterial absorber based on deep learning and surrogate optimization as described above is implemented.

[0050] Those skilled in the art will readily think of other implementation schemes of this application after considering the specification and practicing the content disclosed here. This application aims to cover any variations, uses, or adaptive changes of this application, which follow the general principles of this application and include the common general knowledge or conventional technical means in the technical field not disclosed in this application. The specification and embodiments are only regarded as exemplary, and the true scope and spirit of this application are pointed out by the claims.

[0051] It should be understood that this application is not limited to the exact structures described above and shown in the drawings, and various modifications and changes can be made without departing from its scope. The scope of this application is only limited by the appended claims.

Claims

1. A method for constructing a metamaterial absorber based on deep learning and agent optimization, characterized in that: include: S1: Implementing a parametric randomization generation strategy in MATLAB to construct multiple metamaterial structures; S2: Calling the CST-MATLAB-API interface through MATLAB to control the CST software to perform electromagnetic field simulation on the multiple metamaterial structures to obtain the three-dimensional structure diagram of each metamaterial and the S parameters at different frequencies; S3: Invert the S parameters at different frequencies using MATLAB to obtain the absorption rate of each metamaterial at different frequencies, convert the designed layer structure of the three-dimensional structure diagram into a two-dimensional grayscale image, use the absorption rate and the two-dimensional grayscale image as a set of data, repeat S1-S2 to obtain N sets of data; S4: Using the N groups of data, taking the two-dimensional grayscale image as input and a corresponding set of absorption rates as output, a deep learning model is built and trained; S5: Based on the deep learning model, taking the longest interval length of a set of absorption rates that are continuously greater than a threshold as the fitness function, the two-dimensional grayscale image is optimized using a proxy optimization method to obtain the optimal structure of the metamaterial.

2. The method according to claim 1, characterized in that The parametric randomization generation strategy was implemented in MATLAB to construct multiple metamaterial structures, including: Setting a metamaterial structure simulation configuration in MATLAB, wherein the simulation configuration includes an operating frequency, a background material, a dielectric material, a boundary condition, a global unit, and a solver; Under the simulation configuration, a base model of a three-dimensional structure is built. Based on the base model, n*n required structural pieces, i.e., n*n square units, are randomly created on the two-dimensional surface of the desired design using a MATLAB program. Each unit has two conditions: filled or unfilled. The random creation of n*n required structural pieces is repeated to complete the creation of multiple metamaterial structures.

3. The method according to claim 1, characterized in that The S parameters at different frequencies are inverted using MATLAB to obtain the absorption rate of each metamaterial at different frequencies, including: The S parameters at different frequencies are inverted using MATLAB according to the S parameter inversion formula to obtain the absorption rate of each metamaterial at different frequencies. The formula is as follows: 2 in, is the absorption rate, is the reflectivity, is the frequency.

4. The method according to claim 1, characterized in that Converting the design layer structure of the three-dimensional structure diagram into a two-dimensional grayscale diagram, including: Encoding the design layer structure of the three-dimensional structure diagram, using "0" to represent no filling and "1" to represent filling, arranging them one by one from left to right and from front to back to obtain an n*n matrix composed of "0" and "1"; The n*n matrix is ​​converted into a two-dimensional grayscale image of n*n*1.

5. The method according to claim 1, characterized in that: Using the N groups of data, taking the two-dimensional grayscale image as input and a corresponding set of absorption rates as output, a deep learning model is built and trained, including: The two-dimensional grayscale image is used as input, and a corresponding set of absorption rates is output to build a convolutional neural network; The training parameters are set, the N groups of data are imported into the convolutional neural network, and the convolutional neural network is trained to obtain a deep learning model.

6. The method according to claim 1, characterized in that Based on the deep learning model, the longest interval length of a set of absorption rates continuously greater than a threshold is used as a fitness function, and the two-dimensional grayscale image is optimized by using a proxy optimization method to obtain the optimal structure of the metamaterial, including: In MATLAB, the values ​​of each pixel point of the two-dimensional grayscale image are used as optimization variables, and the length of the longest interval of the set of absorption rates that are continuously greater than the threshold is used as the fitness function; The proxy optimization parameters are set, the two-dimensional grayscale image is optimized, and the optimal structure of the metamaterial absorber is obtained through reverse optimization.

7. A metamaterial absorber construction device based on deep learning and agent optimization, characterized in that: include: Building blocks for implementing parameterized randomized generation strategies in MATLAB to construct multiple metamaterial structures; A simulation module, used to call the CST-MATLAB-API interface through MATLAB to control the CST software to perform electromagnetic field simulation on the multiple metamaterial structures to obtain a three-dimensional structure diagram of each metamaterial and S parameters at different frequencies; A data generation module is used to invert the S parameters at different frequencies using MATLAB to obtain the absorptivity of each metamaterial at different frequencies, convert the design layer structure of the three-dimensional structure diagram into a two-dimensional grayscale diagram, use the absorptivity and the two-dimensional grayscale diagram as a set of data, and repeat the construction module-simulation module to obtain N sets of data; A model building module, for building and training a deep learning model using the N groups of data, taking the two-dimensional grayscale image as input and a corresponding set of absorption rates as output; The optimization module is used to optimize the two-dimensional grayscale image based on the deep learning model, taking the longest interval length of a set of absorption rates continuously greater than a threshold as a fitness function, and using a proxy optimization method to obtain the optimal structure of the metamaterial.

8. An electronic device, characterized in that: include: one or more processors; A memory for storing one or more programs; When the one or more programs are executed by the one or more processors, the one or more processors implement the method according to any one of claims 1 to 6.

9. A computer-readable storage medium having computer instructions stored thereon, characterized in that: When the instruction is executed by a processor, the steps of the method according to any one of claims 1 to 6 are implemented.

Images