Omnidirectional dual-frequency wave-absorbing metasurface intelligent preparation method based on deep learning

Abstract

The invention discloses an intelligent preparation method of an omnidirectional dual-frequency wave-absorbing metasurface based on deep learning, and belongs to the technical field of metamaterial and metasurface design. Comprising the steps that automatic modeling and simulation of different design configurations are carried out, corresponding electromagnetic responses are obtained, and a data set is formed; training a VAE network model based on the data set, and evaluating the accuracy of the VAE network model; when the VAE network model meets a preset condition, inputting a target absorption spectrum into a trained VAE model, and predicting a 0 / 1 coding matrix of the super-surface configuration; and generating the metasurface configuration and the absorption spectrum thereof according to the predicted 0 / 1 coding matrix of the metasurface configuration, and outputting a result. Compared with a traditional design method, the design period can be remarkably shortened, the wave absorbing performance is improved, and meanwhile, the method does not need to depend on complex physical modeling or manual angle adjustment.

Description

Technical Field

[0001] The present invention relates to the technical field of metamaterials and metasurface design, and more specifically, to an intelligent preparation method for an omnidirectional dual-frequency absorbing metasurface based on deep learning. Background Art

[0002] With the rapid development of wireless communication, radar detection, and stealth technology, the demand for absorbing materials is increasing day by day. Most traditional absorbing materials rely on thick layered structures, suffering from problems such as large volume, heavy weight, and limited absorbing efficiency. In recent years, absorbing materials based on metasurface technology have gradually attracted attention. A metasurface is a two-dimensional material that can regulate electromagnetic waves through the design of microstructural units, and can achieve effective absorption of electromagnetic waves with a relatively thin thickness.

[0003] However, traditional absorbing metasurfaces mostly have absorbing characteristics in a single direction or angle, lacking the ability to efficiently absorb omnidirectional electromagnetic waves. In practical applications, the incident angle and polarization state of electromagnetic waves are uncertain.

[0004] Therefore, how to provide a preparation method for metasurfaces with omnidirectional absorbing characteristics is an urgent problem to be solved by those skilled in the art. Summary of the Invention

[0005] In view of this, the present invention provides an intelligent preparation method for an omnidirectional dual-frequency absorbing metasurface based on deep learning, aiming to break through the limitations of traditional design methods in angle adjustment by combining deep learning technology, so as to provide new ideas and methods for the application of absorbing materials and metasurface technology. This method improves the performance and adaptability of the absorbing metasurface by optimizing the design process, and can achieve omnidirectional and dual-frequency electromagnetic wave absorption effects.

[0006] To achieve the above object, the present invention provides the following technical solutions:

[0007] An intelligent preparation method for an omnidirectional dual-frequency absorbing metasurface based on deep learning, comprising:

[0008] Performing automatic modeling and simulation of different design configurations to obtain corresponding electromagnetic responses, and constituting a data set;

[0009] Training a VAE network model based on the data set, and evaluating the accuracy of the VAE network model;

[0010] After the VAE network model meets the preset conditions, inputting a target absorption spectrum into the trained VAE model to predict a 0 / 1 coding matrix of the metasurface configuration;

[0011] Generating a metasurface configuration and its absorption spectrum according to the predicted 0 / 1 coding matrix of the metasurface configuration, and outputting the result.

[0012] Furthermore, the automatic modeling and simulation of different design configurations are performed to obtain corresponding electromagnetic responses, constituting a dataset, including:

[0013] Pre-coding design is carried out using a dual-frequency metamaterial unit based on a five-layer structure, and genetic algorithms are utilized to achieve automatic and simulation, obtaining corresponding electromagnetic responses to constitute a dataset.

[0014] Furthermore, the pre-coding design using a dual-frequency metamaterial unit based on a five-layer structure includes:

[0015] A 0 / 1 coding matrix is used to represent the configuration of the three-layer metal layer;

[0016] A five-layer structure with alternating metal and dielectric is adopted, with a centrosymmetric layout to ensure omnidirectional wave absorption performance;

[0017] On the basis of the five-layer structure, a basic structure with good wave absorption characteristics is set, and 0 / 1 coding optimization is carried out on the basic structure.

[0018] Furthermore, using genetic algorithms to achieve automatic and simulation, obtaining corresponding electromagnetic responses to constitute a dataset, including:

[0019] The frequency range corresponding to when the electromagnetic absorption rate is greater than 80% is selected as the fitness function;

[0020] By simulating the process of natural selection and inheritance, matrices with higher fitness are continuously selected for crossover and mutation operations to generate new coding matrices;

[0021] Each newly generated 0 / 1 coding matrix is fed into a preset automatic modeling and simulation program for electromagnetic response simulation. The response of the metasurface in a specific electromagnetic environment is simulated by numerical methods to generate electromagnetic response data for each configuration, constituting a dataset.

[0022] Furthermore, the calculation formula for the electromagnetic absorption rate is:

[0023] A = 1 - R - T = 1 - |S 11 | 2 - |S 21 | 2 ;

[0024] In the formula, A is the electromagnetic absorption rate of the metamaterial absorber, R is the reflectivity, T is the transmittance, S 11 is the reflection coefficient, and S 21 is the transmission coefficient.

[0025] Furthermore, the VAE network model includes an encoder and a decoder. Among them, the electromagnetic response data in the input data set is mapped into the latent space through the encoder, and a new structural configuration is generated through the decoder.

[0026] Based on the data set, the VAE network model is iteratively trained multiple times so that the VAE network model outputs a 0 / 1 coding matrix of the corresponding metasurface configuration, and a metasurface configuration matching the target electromagnetic absorption performance is obtained by combining with simulation software.

[0027] As can be seen from the above technical solutions, compared with the prior art, the present invention discloses an intelligent preparation method for an omnidirectional dual-band absorbing metasurface based on deep learning, and proposes a new solution to the design problem of an efficient omnidirectional dual-band absorbing metasurface. This method uses a 0 / 1 coding matrix to uniformly characterize the metamaterial configuration, and generates a 0 / 1 matrix based on the genetic algorithm. The automatic modeling and combined simulation program quickly collects a large number of and structurally diverse metamaterial absorber configurations and their corresponding electromagnetic response data to construct a data set for subsequent deep learning. Thereafter, the absorbing metasurface is intelligently optimized through a deep learning algorithm, and the omnidirectional dual-band absorbing effect can be achieved without relying on traditional angle adjustment, thereby improving the performance and adaptability of the metasurface. The key lies in combining deep learning technology with metasurface design, and using its powerful self-learning and optimization capabilities to automatically adjust the shape and layout of metasurface units to achieve the best absorbing effect.

[0028] Specifically, the present invention constructs a deep neural network model, inputs performance targets, and the network automatically generates an optimal design scheme by learning a large number of simulation results. The intelligent design method of the angle-free absorbing metasurface based on deep learning can automatically search for a metasurface structure with excellent omnidirectional absorbing performance in a complex design space by using the powerful non-linear mapping ability of the deep learning model. This method not only greatly improves the design efficiency of absorbing materials, but also provides a new idea for realizing high-performance and lightweight omnidirectional absorbing metasurfaces. Compared with traditional design methods, this method can significantly reduce the design cycle, improve the absorbing performance, and at the same time does not need to rely on complex physical modeling or manual angle adjustment. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] In order to more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the drawings in the following description are only the embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained according to the provided drawings without creative efforts.

[0030] Figure 1 It is the overall flowchart of the intelligent preparation of the angle-free (omnidirectional) absorbing metasurface of the present invention;

[0031] Figure 2 Schematic diagram of the preparation structure of the microwave absorption metasurface of the present invention;

[0032] Figure 3 Schematic diagram of the data set collected by iterative genetic algorithm in the embodiment of the present invention. Detailed implementation manners

[0033] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0034] An embodiment of the present invention discloses an intelligent preparation method for an omnidirectional dual-band microwave absorption metasurface based on deep learning, including:

[0035] Performing automatic modeling and simulation of different design configurations to obtain corresponding electromagnetic responses, and constituting a data set;

[0036] Training a VAE network model based on the data set, and evaluating the accuracy of the VAE network model;

[0037] After the VAE network model meets the preset conditions, input the target absorption spectrum into the trained VAE model to predict the 0 / 1 coding matrix of the metasurface configuration;

[0038] Generating a metasurface configuration and its absorption spectrum according to the predicted 0 / 1 coding matrix of the metasurface configuration, and outputting the result.

[0039] In a specific embodiment, referring to Figure 1 as shown, an intelligent preparation method for an omnidirectional dual-band microwave absorption metasurface based on deep learning specifically includes the following processes:

[0040] Start: The process starts from the "Start" node;

[0041] Initializing the population: In the first step of the process, it is necessary to initialize a population. Specifically, a set of initial solutions or individuals are created, and these individuals will be used in the subsequent optimization process;

[0042] Calculating the fitness function: Next, calculate the fitness value for each individual. The specific fitness function is a standard used to measure how well an individual performs in solving a specific problem;

[0043] Selection, Crossover, Mutation: Based on the calculated fitness values, perform selection operations to determine which individuals will be retained to participate in the generation of the next generation. Then, generate new offspring individuals through crossover and mutation operations;

[0044] Automatic Modeling and Simulation Based on 0 / 1 Encoding Matrix: Use the 0 / 1 encoding matrix constructed by the new generation of individuals generated in the previous step to perform automatic modeling and simulation;

[0045] Collect the Encoding Matrix and Its Corresponding S 11 Construct a Dataset: In this step, collect the 0 / 1 encoding matrices used by all individuals who participated in the modeling and simulation, as well as their corresponding reflection coefficients S 11 to form the dataset;

[0046] Test Set: Use a portion of the data as the test set. This portion of the data does not participate in the training process but is used to evaluate the generalization ability of the VAE network model;

[0047] Training Set: Use another portion of the data as the training set to train the VAE network model;

[0048] Test Network Accuracy: Use the test set to test the accuracy of the VAE network model;

[0049] VAE Network Model Training: If the test results are not satisfactory, re - perform operations such as selection, crossover, and mutation; otherwise:

[0050] Output Evaluation Results: When certain conditions are met (such as reaching the preset maximum number of iterations or minimum error threshold), output the final evaluation results;

[0051] Input the Target Absorption Spectrum into the Trained VAE Model: Input the target absorption spectrum into the trained Variational Autoencoder (VAE) model;

[0052] The Model Predicts the 0 / 1 Encoding Matrix of the Metasurface Configuration: The VAE model will predict the 0 / 1 encoding matrix of the corresponding metasurface configuration based on the input target absorption spectrum;

[0053] Combine with Simulation Software to Obtain the Metasurface Configuration and Absorption Spectrum: Finally, combine with the simulation software to generate the specific metasurface configuration and its corresponding absorption spectrum;

[0054] End: The entire process ends here.

[0055] In a specific embodiment, the pre - coding of the dual - frequency metamaterial unit structure based on a five - layer structure includes:

[0056] To facilitate the mathematical characterization of the design of the metasurface absorber, a 0 / 1 coding matrix is used to represent the three-layer metal layer configuration of the metamaterial absorber. During the design process, the 0 / 1 coding matrix can effectively represent different structural parameters and simplifies the optimization of the metamaterial configuration.

[0057] The wave-absorbing metasurface absorber designed in the present invention adopts a five-layer structure of metal-dielectric-metal-dielectric-metal, as Figure 2 shown. To achieve the omnidirectional wave absorption characteristics of the metasurface, it is necessary to be insensitive to the angles of incident waves at different angles. Therefore, the configuration of the metal layer is centrosymmetric, and the configurations of the upper and lower metal layers are the same. The middle metal layer is generated by a single central symmetry of a certain specific pattern, and the top and bottom metal layers are generated by two central symmetries of another specific pattern. This design can not only improve the stability of the absorber, make it easier to achieve dual-band performance, but also optimize the absorption effect of electromagnetic waves at different angles and frequency bands.

[0058] Since the metasurface design involves complex geometries and structural regulations, a completely random coding matrix will result in a structure without wave absorption ability. Therefore, some optimization measures are taken to solve this problem. First, during the design process, based on engineering experience, simple basic structures with wave absorption characteristics are preset, and 0 / 1 coding within the metamaterial unit region is further established on these basic configurations to achieve more detailed and targeted optimization. This can increase the flexibility and optimization space of the design without losing the wave absorption performance of the basic structure, avoid the problem of excessive simulation time consumption that may occur when directly using a completely random matrix in large-scale design, and at the same time ensure that the metasurface has effective wave absorption ability during the optimization process.

[0059] Specifically, the five-layer structure design refers to the specific physical architecture, that is, how the metal and dielectric layers are arranged and their symmetry design, which is to ensure the wave absorption performance of the absorber at different angles.

[0060] The basic structure refers to some preliminary structural patterns (such as metal patterns with certain specific shapes or layouts) selected based on existing experience and knowledge after determining the five-layer structure. These basic structures have been verified to have good wave absorption performance. On this basis, further in-depth optimization is carried out by finely adjusting the 0 / 1 coding matrix, aiming to explore more possible improvement spaces while maintaining the basic wave absorption performance. That is to say, in this embodiment, the overall framework and key design principles are first defined, and then the process of refinement and optimization is carried out within this framework.

[0061] Specifically, through the precoding of the dual - frequency metamaterial unit structure with a five - layer structure, the interaction between the metal layer and the dielectric layer can effectively regulate the propagation characteristics of electromagnetic waves, reduce reflection, and enhance absorption performance. Due to the central symmetry of the metal layer configuration, the absorbing metasurface can maintain a good absorbing effect at different incident angles, achieving the goal of omnidirectional absorption.

[0062] In a specific embodiment, a dataset is constructed based on the automatic modeling and joint simulation of the genetic algorithm, including:

[0063] To further improve the design efficiency and accuracy, the present invention adopts a method of combining the genetic algorithm with an automatic modeling and joint simulation program for design optimization. The specific implementation steps are as follows:

[0064] Define a suitable fitness function to ensure that the design meets the specific requirements of electromagnetic wave absorption, such as the absorption frequency band range and the absorption rate numerical standard. The design of the fitness function considers expanding the absorption bandwidth as much as possible within the design frequency band to enhance the absorption ability of the metasurface to electromagnetic waves of different frequencies.

[0065] To optimize the design of the 0 / 1 coding matrix, a binary genetic algorithm is adopted. The genetic algorithm simulates the processes of natural selection and heredity, continuously selects matrix structures with higher fitness for operations such as crossover and mutation, thereby generating new coding matrices. Each 0 / 1 matrix represents a possible metasurface configuration. Then, these generated matrices are fed into the automatic modeling and simulation program for simulation to obtain the electromagnetic response data corresponding to each configuration.

[0066] The electromagnetic absorption rate (A) of the metamaterial absorber is determined by the reflectivity (R) and the transmittance (T). Therefore, during the design optimization process, relevant data on reflectivity and transmittance need to be concerned. During the simulation process, the system will collect each 0 / 1 coding matrix and the corresponding electromagnetic response parameters (reflection coefficient S11 and transmission coefficient S21). These data will be used to construct a dataset and for subsequent training and testing. The specific steps of the genetic algorithm module design are as follows:

[0067] Fitness function design

[0068] In the genetic algorithm, the fitness function is the criterion for evaluating the quality of an individual. In the design of the metamaterial absorber, the goal of this embodiment is to improve the absorption efficiency as much as possible, broaden the absorption bandwidth, and meet specific electromagnetic wave absorption requirements. For this reason, this embodiment defines a suitable fitness function considering the following aspects:

[0069] Absorption frequency band range: This function needs to ensure that the design meets the required absorption frequency band range. This embodiment hopes that the metamaterial can have a strong absorption ability for electromagnetic waves within a specific frequency range.

[0070] Absorption rate: The absorption rate represents the absorption efficiency of materials for electromagnetic waves and can usually be indirectly measured by the reflectivity (R). This function will consider the requirement of the absorption rate, that is, to obtain a relatively high absorption rate within the designed frequency band as much as possible.

[0071] Absorption bandwidth: To ensure that the metamaterial can provide good absorption effects for electromagnetic waves of different frequencies, this embodiment requires designing a relatively wide absorption bandwidth. The fitness function needs to comprehensively consider the reflectivity data within the frequency band to optimize the bandwidth.

[0072] The fitness function selected in this design is the frequency range corresponding to when the electromagnetic absorption rate is greater than 80%, where the calculation formula for the electromagnetic absorption rate is:

[0073] A = 1 - R - T = 1 - |S 11 | 2 -|S 21 | 2 (1)

[0074] Binary coding and genetic operations

[0075] In the genetic algorithm, the individual solutions are represented by coding. For the metamaterial design problem, a 0 / 1 coding matrix method is adopted for design, and each element (0 or 1) in the matrix represents. Through different 0 / 1 coding matrices, different metamaterial structures can be represented, where "1" represents a metal sheet and "0" represents an air sheet. The genetic operations include the following three aspects:

[0076] Selection operation: In each generation, the genetic algorithm will evaluate the quality of each matrix according to the fitness function and select the matrices with higher fitness as the "parent generations". These parent generation matrices generate the next generation of matrices through crossover and mutation operations.

[0077] Crossover operation: Through the crossover operation, the genetic algorithm combines some genes of two parent generation matrices into a new matrix. The purpose of the crossover operation is to combine the characteristics of two excellent individuals to produce better offspring.

[0078] Mutation operation: The mutation operation randomly changes some elements in the matrix to introduce new gene combinations. The mutation operation increases the diversity and helps to prevent the algorithm from falling into a local optimal solution.

[0079] These operations can continuously generate new design solutions and finally form a matrix structure with higher fitness.

[0080] Automatic modeling and simulation program

[0081] Each generated 0 / 1 coding matrix is fed into an automatic modeling and simulation program for electromagnetic response simulation. This program simulates the response of the metasurface in a specific electromagnetic environment through numerical methods, generating electromagnetic response data for each configuration. To ensure the accuracy of the simulation results, the automatic modeling program can automatically generate geometric structures and perform electromagnetic field calculations to obtain S11 and S21. The key task in the design optimization process is to minimize the reflectivity and transmittance, achieve the wave absorption goal of no reflection and no transmission, and at the same time broaden the absorption bandwidth and optimize the absorption performance of the design.

[0082] Collection of Absorption Rate and Construction of Dataset

[0083] During the simulation process, the system collects each 0 / 1 coding matrix and its corresponding S11 and S21 data, and then calculates the corresponding absorption rate. These data are organized into a dataset and used as the basis for subsequent training and testing. The dataset includes different design configurations and their corresponding electromagnetic responses, which can provide training samples for subsequent deep learning models.

[0084] Specifically, see Figure 3 As shown, it is an example of the dataset collected by the genetic algorithm in this embodiment.

[0085] In a specific embodiment, the generation of the metasurface wave absorption unit configuration based on VAE includes:

[0086] After obtaining a large number of 0 / 1 coding matrices and their corresponding electromagnetic response data, the present invention designs a variational autoencoder (VAE) generation model to achieve the ability to generate corresponding metasurface absorber configurations according to target electromagnetic absorption data. VAE is a deep learning model that can generate new structural designs through input target data (such as electromagnetic absorption parameters).

[0087] The design focus of the VAE generation model is that it can learn the latent space of the electromagnetic absorber from a large-scale dataset and generate new designs with similar performance. Specifically, the VAE model maps the input electromagnetic response data to the latent space through the encoder, and then generates new structural configurations through the decoder. After multiple iterations and training, the VAE model can generate metasurface configurations that match the target electromagnetic absorption performance.

[0088] The VAE model contains two main parts: an encoder and a decoder, where:

[0089] (1) Encoder: The role of the encoder is to map the input data (such as electromagnetic absorption rate) to the latent space to obtain the latent variable z. The latent space is a low-dimensional representation of the data, containing the main features of the data. Through training, the encoder can learn the relationship between the reflectivity and the metasurface structure, and convert the input data into a vector representation in the latent space.

[0090] (2) Decoder: The role of the decoder is to generate a new design structure based on the latent variable z. The decoder samples the latent variable from the latent space and decodes it into a new metasurface design. After multiple trainings, the decoder can generate new design configurations that match the target electromagnetic absorption performance.

[0091] The VAE optimizes the loss function so that the generated metasurface design not only has good electromagnetic absorption performance but also has characteristics similar to the input data. The loss function includes reconstruction loss and KL divergence loss. The reconstruction loss ensures that the generated structure can restore the input data, while the KL divergence loss constrains the distribution of the latent space.

[0092] During the training process, the VAE learns its latent space by inputting a large amount of electromagnetic response data (i.e., electromagnetic absorption rate and other related parameters). As the training progresses, the VAE model can generate new designs that match the target absorption performance. In each iteration, the model gradually optimizes the parameters until the generated design can meet the predetermined electromagnetic absorption requirements.

[0093] After the VAE training is completed, it can generate new metamaterial structures according to the target electromagnetic absorption performance. These new designs will serve as candidate structures for future metamaterial absorbers and further enter the experimental verification stage. Through the VAE generation model, it is possible to automatically generate new designs with similar electromagnetic responses starting from the target performance requirements, avoiding the limitations of manual intervention and empirical knowledge in traditional design methods.

[0094] In a specific embodiment, first, aiming at the geometric complexity of the metasurface design structure, the ineffectiveness and time-consuming nature of completely random coding, in the design process, the basic structure designed by artificial simulation is combined with a random 0 / 1 coding matrix, increasing the design flexibility and optimization space without losing the wave absorption performance. Second, the application of the VAE generation model in the present invention helps to accelerate the design process of the metamaterial absorber. Automatically generating an efficient wave-absorbing metasurface structure through the deep learning model greatly reduces the manual intervention and repeated simulation processes required by traditional design methods. At the same time, the VAE model can continuously improve the design accuracy and efficiency in multiple iterations, making the finally generated metasurface have better wave absorption characteristics.

[0095] The various embodiments in this specification are described in a progressive manner. Each embodiment focuses on the differences from other embodiments. For the same or similar parts between the embodiments, reference can be made to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0096] The foregoing description of the disclosed embodiments enables those skilled in the art to practice or use the present invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Thus, the present invention is not intended to be limited to the embodiments shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. A method for intelligent preparation of omnidirectional dual-frequency absorbing metasurface based on deep learning, characterized in that: include: Automatically model and simulate different design configurations to obtain corresponding electromagnetic responses and form data sets; Training a VAE network model based on the data set and evaluating the accuracy of the VAE network model; When the VAE network model meets the preset conditions, the target absorption spectrum is input into the trained VAE model to predict the 0 / 1 coding matrix of the metasurface configuration; The metasurface configuration and its absorption spectrum are generated according to the predicted 0 / 1 coding matrix of the metasurface configuration, and the results are output.

2. According to the deep learning-based omnidirectional dual-frequency absorbing metasurface intelligent preparation method of claim 1, it is characterized in that: The automatic modeling and simulation of different design configurations are performed to obtain corresponding electromagnetic responses and form a data set, including: A dual-frequency metamaterial unit based on a five-layer structure is used for precoding design, and a genetic algorithm is used to realize automation and simulation to obtain the corresponding electromagnetic response and form a data set.

3. The method for intelligently preparing an omnidirectional dual-frequency absorbing metasurface based on deep learning according to claim 2, characterized in that: The precoding design using a dual-frequency metamaterial unit based on a five-layer structure includes: A 0 / 1 coding matrix is ​​used to represent the three-layer metal layer configuration; Adopting a five-layer structure of alternating metal and dielectric, with a central symmetrical layout, to ensure omnidirectional wave absorption performance; A basic structure with good wave absorption characteristics is set on the basis of the five-layer structure, and 0 / 1 coding optimization is performed on the basic structure.

4. The method for preparing an omnidirectional dual-frequency absorbing metasurface intelligently based on deep learning according to claim 2, characterized in that: Genetic algorithms are used to achieve automation and simulation, and the corresponding electromagnetic responses are obtained to form a data set, including: Select the frequency range corresponding to when the electromagnetic absorption rate is greater than 80% as the fitness function; By simulating the process of natural selection and inheritance, the matrix structure with higher fitness is continuously selected for crossover and mutation operations to generate new encoding matrices; Each new 0 / 1 coding matrix generated is sent to the preset automatic modeling and simulation program for electromagnetic response simulation. The response of the metasurface in a specific electromagnetic environment is simulated by numerical methods, and the electromagnetic response data of each configuration is generated to form a data set.

5. The method for intelligently preparing an omnidirectional dual-frequency absorbing metasurface based on deep learning according to claim 4, characterized in that: The calculation formula of the electromagnetic absorption rate is: A=1-R-T=1-|S 11 | 2 -|S 21 | 2 ; Where A is the electromagnetic absorption rate of the metamaterial absorber, R is the reflectivity, T is the transmittance, S 11 is the reflection coefficient, S 21 is the transmission coefficient.

6. The method for intelligently preparing an omnidirectional dual-frequency absorbing metasurface based on deep learning according to claim 2, characterized in that: The VAE network model includes an encoder and a decoder, wherein the electromagnetic response data in the input data set is mapped into the latent space by the encoder, and a new structural configuration is generated by the decoder; Based on the data set, after multiple iterations and training of the VAE network model, the VAE network model outputs the corresponding 0 / 1 coding matrix of the metasurface configuration, and the joint simulation software obtains the metasurface configuration that matches the target electromagnetic absorption performance.

Images