Method, device, medium and equipment for obtaining training data applied to electromagnetic device design

Abstract

The application discloses a training data obtaining method and device applied to electromagnetic device design, a medium and equipment, and relates to the technical field of electromagnetic device design. The application utilizes the inverse mapping relationship between the electromagnetic response curve and the topological distribution of the frequency selective surface structure, firstly performs topological modeling and full-wave simulation, and utilizes a binary matrix to express the topological structure. In the iteration of the target function, the topological binary matrix and the electromagnetic response reserved in each iteration are taken as a training data set, more useful design information can be reserved, the electromagnetic response has more similar characteristics with the design target, misleading information is less, and the quality of the training data in the electromagnetic device design is improved.

Description

Technical Field

[0001] This application relates to the field of electromagnetic device design technology, specifically to a method, apparatus, medium, and device for obtaining training data applied to electromagnetic device design. Background Technology

[0002] Deep learning neural networks (DNNs), with their powerful data fitting capabilities, are increasingly being applied to solve various problems in the electromagnetic field, such as microwave filter design, antenna design, beamform synthesis, and metasurface design. To grant greater freedom in electromagnetic device design, topology modeling is attracting increasing attention. Topology modeling largely incorporates more modeling variables to construct the topological structure of the modeling domain, thus providing more design freedom. However, due to the large number of binary variable combinations and the uncertainty of the initial structure, existing methods based on random collection and experimental design for collecting training samples are not suitable. Therefore, how to construct a high-quality training dataset within a limited full-wave simulation time—that is, to obtain more qualified samples to provide useful information for design—is a very meaningful research problem. Summary of the Invention

[0003] The main purpose of this application is to provide a method, apparatus, medium and equipment for obtaining training data for electromagnetic device design, aiming to solve the problem of poor dirt rendering effect caused by the influence of lighting conditions in the prior art.

[0004] To achieve the above objectives, the technical solutions adopted in the embodiments of this application are as follows:

[0005] In a first aspect, embodiments of this application provide a method for obtaining training data for electromagnetic device design, comprising the following steps:

[0006] Initialize structure parameters;

[0007] Topological modeling and full-wave simulation of frequency-selective surface structures are performed based on structural parameters to obtain the topological binary matrix and electromagnetic response curves of the frequency-selective surface structures.

[0008] Calculate the objective function based on the electromagnetic response curve; the objective function is used to characterize the optimization direction.

[0009] Determine if the objective function has reached the maximum number of iterations;

[0010] If the result of the judgment is negative, the structural parameters are updated according to the update mechanism to obtain the topological binary matrix of the frequency selective surface structure and the corresponding electromagnetic response curve. Then, the steps of topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters are returned to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure, until the result of the judgment is positive.

[0011] A training dataset is constructed by selecting the topological binary matrix of the surface structure and the corresponding electromagnetic response curve based on the frequency.

[0012] In one embodiment of the first aspect, before calculating the objective function based on the electromagnetic response curve, the method further includes:

[0013] The electromagnetic response curve is processed into a square wave at each sampling frequency point to obtain the target electromagnetic response curve.

[0014] Based on the electromagnetic response curve, calculate the objective function, including:

[0015] Calculate the objective function based on the target electromagnetic response curve.

[0016] In one embodiment of the first aspect, the electromagnetic response curve is processed into a square wave at each sampling frequency point to obtain the target electromagnetic response curve, including:

[0017] The electromagnetic response curve is binarized at each sampling frequency point based on a threshold. Values ​​less than the threshold are set as the first value, and those not are set as the second value, thus obtaining the target electromagnetic response curve.

[0018] In one embodiment of the first aspect, topological modeling and full-wave simulation of a frequency-selective surface structure are performed based on structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure, including:

[0019] Topological modeling of frequency-selective surface structures is performed based on structural parameters to obtain the topological structure of frequency-selective surface structures.

[0020] The topological structure is binarized based on the binary representation rule to obtain the topological binary matrix of the frequency-selective surface structure.

[0021] In one embodiment of the first aspect, before performing binary representation of the topology based on binary representation rules to obtain the topological binary matrix of the frequency-selective surface structure, the method further includes:

[0022] Based on the frequency of the selected surface structure samples, the metal distribution is analyzed to obtain a binary characterization rule.

[0023] In one embodiment of the first aspect, a binary characterization rule is obtained by selecting the metal distribution of the surface structure sample based on frequency, including:

[0024] The stacked structure of the surface structure sample is selected based on frequency, the metal layer is identified, and the metal layer is discretized into multiple pixel-level metal sheets;

[0025] Based on the metal distribution in the metal sheet representation region in topological modeling, binary representation rules are obtained.

[0026] In one embodiment of the first aspect, calculating the objective function based on the electromagnetic response curve includes:

[0027] Determine the optimization direction based on the electromagnetic response curve;

[0028] Based on the optimization direction and importance, obtain the sub-functions and their weights;

[0029] Calculate the objective function based on the sub-functions and their weights.

[0030] Secondly, embodiments of this application provide a training data acquisition device for electromagnetic device design, comprising:

[0031] The initialization module is used to initialize structure parameters;

[0032] The simulation module is used for topological modeling and full-wave simulation of frequency-selective surface structures based on structural parameters, and to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure.

[0033] The calculation module is used to calculate the objective function based on the electromagnetic response curve; the objective function is used to characterize the optimization direction.

[0034] The judgment module is used to determine whether the objective function has reached the maximum number of iterations;

[0035] The iterative module is used to update the structural parameters according to the update mechanism when the judgment result is negative, obtain the topological binary matrix of the frequency selective surface structure and the corresponding electromagnetic response curve, and return the topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure, until the judgment result is positive.

[0036] The module is used to select the topological binary matrix of the surface structure and the corresponding electromagnetic response curve based on the frequency, and to build the training dataset.

[0037] Thirdly, embodiments of this application provide a computer-readable storage medium storing a computer program, which, when loaded and executed by a processor, implements the training data acquisition method for electromagnetic device design provided in any of the first aspects above.

[0038] Fourthly, embodiments of this application provide an electronic device, including a processor and a memory, wherein,

[0039] Memory is used to store computer programs;

[0040] The processor is used to load and execute a computer program to cause the electronic device to perform a training data acquisition method for electromagnetic device design as provided in any of the first aspects above.

[0041] Compared with the prior art, the beneficial effects of this application are:

[0042] This application proposes a method, apparatus, medium, and device for obtaining training data for electromagnetic device design. The method includes: initializing structural parameters; performing topological modeling and full-wave simulation of a frequency-selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure; calculating an objective function based on the electromagnetic response curve, wherein the objective function is used to characterize the optimization direction; determining whether the objective function has reached the maximum number of iterations; if the determination result is negative, updating the structural parameters according to an update mechanism to obtain the topological binary matrix and corresponding electromagnetic response curve of the frequency-selective surface structure, and returning to the step of performing topological modeling and full-wave simulation of the frequency-selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure, until the determination result is positive; and constructing a training dataset based on the topological binary matrix and corresponding electromagnetic response curve of the frequency-selective surface structure. This application utilizes the inverse mapping relationship between electromagnetic response curves and the topological distribution of frequency-selective surface structures. First, topological modeling and full-wave simulation are performed, and a binary matrix is ​​used to represent the topological structure. In the iteration of the objective function, the topological binary matrix and electromagnetic response retained in each iteration are used as training datasets. This can retain more useful design information, and the electromagnetic response has more similar characteristics to the design target, with less misleading information, thus improving the quality of training data in electromagnetic device design. Attached Figure Description

[0043] Figure 1 This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of this application;

[0044] Figure 2 A flowchart illustrating a method for obtaining training data for electromagnetic device design, provided in an embodiment of this application;

[0045] Figure 3 A schematic diagram of the stacked structure of the frequency-selective surface structure in the training data acquisition method for electromagnetic device design provided in the embodiments of this application;

[0046] Figure 4 The schematic diagrams of three frequency-selective surface structures in the training data acquisition method for electromagnetic device design provided in the embodiments of this application are structure 1, structure 2 and structure 3, respectively.

[0047] Figure 5 Schematic diagrams of three other frequency-selective surface structures provided in the training data acquisition method for electromagnetic device design according to the embodiments of this application are structure 4, structure 5 and structure 6, respectively.

[0048] Figure 6 A schematic diagram illustrating the discretization method of the metal layer in the training data acquisition method for electromagnetic device design provided in the embodiments of this application;

[0049] Figure 7 A schematic diagram illustrating the effectiveness of the method provided in the embodiments of this application for testing Example 1;

[0050] Figure 8 A schematic diagram illustrating the testing of Example 2 to verify the effectiveness of the method provided in the embodiments of this application;

[0051] Figure 9 A schematic diagram illustrating the testing of Example 3 to verify the effectiveness of the method provided in the embodiments of this application;

[0052] Figure 10 A schematic diagram illustrating the effectiveness of the method provided in the embodiments of this application for testing Example 4;

[0053] Figure 11 A schematic diagram of a training data acquisition device for electromagnetic device design provided in an embodiment of this application;

[0054] The diagram is labeled as follows: 101-Processor, 102-Communication bus, 103-Network interface, 104-User interface, 105-Memory. Detailed Implementation

[0055] It should be understood that the specific embodiments described herein are merely illustrative of this application and are not intended to limit this application.

[0056] See attached document Figure 1 , attached Figure 1This is a schematic diagram of the electronic device structure of the hardware operating environment involved in the embodiments of this application. The electronic device may include: a processor 101, such as a central processing unit (CPU), a communication bus 102, a user interface 104, a network interface 103, and a memory 105. The communication bus 102 is used to realize the connection and communication between these components. The user interface 104 may include a display screen and an input unit such as a keyboard. Optionally, the user interface 104 may also include a standard wired interface and a wireless interface. The network interface 103 may optionally include a standard wired interface and a wireless interface (such as a Wi-Fi interface). The memory 105 may be a storage device independent of the aforementioned processor 101. The memory 105 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as at least one disk storage device. The processor 101 may be a general-purpose processor, including a central processing unit, a network processor, etc., or it may be a digital signal processor, an application-specific integrated circuit, a field-programmable gate array or other programmable logic device, discrete gate or transistor logic device, or discrete hardware component.

[0057] Those skilled in the art will understand that the appendix Figure 1 The structure shown does not constitute a limitation on the electronic device and may include more or fewer components than shown, or combine certain components, or have different component arrangements.

[0058] As attached Figure 1 As shown, the memory 105, which serves as a storage medium, may include an operating system, a network communication module, a user interface module, and a training data acquisition device for electromagnetic device design.

[0059] In the appendix Figure 1 In the electronic device shown, the network interface 103 is mainly used for data communication with the network server; the user interface 104 is mainly used for data interaction with the user; the processor 101 and the memory 105 in this application can be set in the electronic device. The electronic device calls the training data acquisition device for electromagnetic device design stored in the memory 105 through the processor 101 and executes the training data acquisition method for electromagnetic device design provided in the embodiment of this application.

[0060] Deep learning neural networks (DNNs), with their powerful data fitting capabilities, are increasingly being applied to solve various problems in the electromagnetic field, such as microwave filter design, antenna design, beamform synthesis, and metasurface design. To grant greater freedom in electromagnetic device design, topology modeling is attracting increasing attention. However, unlike parametric modeling, a major type of machine learning-based design method, topology modeling largely incorporates more modeling variables to construct the topological structure of the modeling domain, thus providing greater design freedom.

[0061] During training, the training dataset plays a crucial role; its quantity and quality directly determine the training convergence and performance of the DNN model. For parametric modeling, random collection and design of experiment (DoE) methods are typically used to collect training samples. However, for topological modeling, due to the large number of binary variable combinations and the uncertainty of the initial structure, the above methods are not applicable. Therefore, effectively constructing a training dataset remains a challenging task.

[0062] In terms of the number of training samples, a large number (sometimes thousands) can lead to better mapping performance and avoid overfitting to some extent. However, collecting a large training dataset from full-wave electromagnetic simulations is computationally expensive. Therefore, improving the performance of machine learning models by increasing the number of training samples is both costly and inefficient.

[0063] Regarding the quality of the training dataset, high-quality samples can also improve the mapping performance of DNN models. High quality means that the electromagnetic responses of the samples have some important characteristics similar to the design objectives and contain little misleading information. High-quality training datasets are helpful for design; given two training datasets of equal size, training a DNN with a higher-quality dataset is more likely to achieve better performance without requiring additional training samples.

[0064] Therefore, how to construct a high-quality training dataset within a limited full-wave simulation time, i.e., to obtain more qualified samples and provide useful information for design, is a very meaningful research problem. However, in machine learning-based electromagnetic device design, there is currently no research addressing this problem. To this end, this application provides a solution that utilizes the inverse mapping relationship between the electromagnetic response curve and the topological distribution of the frequency-selective surface structure. First, topological modeling and full-wave simulation are performed, and a binary matrix is ​​used to represent the topological structure. In the iteration of the objective function, the topological binary matrix and electromagnetic response retained in each iteration are used as the training dataset. This approach can retain more useful design information, and the electromagnetic response has more similar characteristics to the design target, with less misleading information, thus improving the quality of training data in electromagnetic device design.

[0065] See attached document Figure 2 Based on the hardware device described in the foregoing embodiments, embodiments of this application provide a method for obtaining training data for electromagnetic device design, comprising the following steps:

[0066] S10: Initialize structure parameters.

[0067] S20: Based on structural parameters, perform topological modeling and full-wave simulation of frequency-selective surface structures to obtain the topological binary matrix and electromagnetic response curves of the frequency-selective surface structures.

[0068] In the specific implementation process, since it is an iterative process, the geometric parameters first need to be initialized. Based on the structural parameters, the frequency selective surface (FSS) structure was modeled and full-wave simulated in the MATLAB-CST co-simulation program. The topological model and electromagnetic response curves |S| of the frequency selective surface structure were obtained respectively. 11 Topology modeling uses a binary matrix to represent the structure. Specifically, based on structural parameters, topology modeling and full-wave simulation of frequency-selective surface structures are performed to obtain the topological binary matrix and electromagnetic response curves of the frequency-selective surface structures, including:

[0069] Topological modeling of frequency-selective surface structures is performed based on structural parameters to obtain the topological structure of frequency-selective surface structures.

[0070] The topological structure is binarized based on the binary representation rule to obtain the topological binary matrix of the frequency-selective surface structure.

[0071] In practical implementation, binary matrices can simplify the representation of the topological structure of frequency-selective surface structures. By establishing rules for binary representation in advance, topological modeling can directly output the topological structure in the form of a binary matrix. That is, before obtaining the topological binary matrix of the frequency-selective surface structure by performing binary representation of the topological structure based on the binary representation rules, the method also includes:

[0072] Based on the frequency of the selected surface structure samples, the metal distribution is analyzed to obtain a binary characterization rule.

[0073] The binary distribution is characterized by its metal distribution, thus obtaining binary representation rules. Specifically, the binary representation rules are obtained by selecting surface structure samples based on their metal distribution according to frequency, including:

[0074] The stacked structure of the surface structure sample is selected based on frequency, the metal layer is identified, and the metal layer is discretized into multiple pixel-level metal sheets;

[0075] Based on the metal distribution in the metal sheet representation region in topological modeling, binary representation rules are obtained.

[0076] In the specific implementation process, the stacked structure of the frequency-selective surface structure is as shown in the attached figure. Figure 3 As shown, the basic structure consists of three metal layers and two dielectric layers. The top and bottom metal layers have the same metal patch pattern, and both substrates have the same dimensions. D is the dielectric layer thickness, and P is the width of the frequency-selective surface structure. (See attached diagram.) Figure 4 Appendix Figure 5 Taking the six different types of FSS structures shown as examples, L1-L5 respectively identify some parameter dimensions on the metal layers. The two images in each column represent different metal layers of the same FSS structure. To obtain the topological distribution of the FSS structure, each metal layer of the FSS structure is discretized into q×q square pixel metal sheets, as shown in the attached figure. Figure 6 As shown, using a discrete method with q=160, the topological distribution of the FSS structure can then be represented by a binary matrix, where "1" represents metal and "0" represents non-metal.

[0077] S30: Calculate the objective function based on the electromagnetic response curve; whereby the objective function is used to characterize the optimization direction.

[0078] In practical implementation, the objective function used for electromagnetic response limiting determines the direction of data optimization. The objective function typically consists of several sub-functions, each with a specific weight to determine its importance. That is, based on the electromagnetic response curve, the objective function is calculated, including:

[0079] Determine the optimization direction based on the electromagnetic response curve;

[0080] Based on the optimization direction and importance, obtain the sub-functions and their weights;

[0081] Calculate the objective function based on the sub-functions and their weights.

[0082] Taking the FSS design as an example, the objective function in this embodiment is shown in the following formula:

[0083]

[0084] The objective function consists of two sub-functions with equal weights w1 and w2, both set to 0.5, because the two corresponding objectives are of comparable importance. Sub-function f1(x) optimizes the passband bandwidth, as narrow bandwidth is detrimental to DNN model training; sub-function f2(x) optimizes the number of passbands, because when |S... 11 Drastic changes can also hinder the learning of deep learning models. It's important to note that the objective function is not for strict electromagnetic response constraints, but rather to ensure that the samples contain some important features similar to the design objective.

[0085] In one embodiment, before calculating the objective function based on the electromagnetic response curve, the method further includes:

[0086] The electromagnetic response curve is processed into a square wave at each sampling frequency point to obtain the target electromagnetic response curve.

[0087] In practical implementation, to aid in iterative optimization, square wave processing can be performed before calculating the objective function. An ideal square wave has only two values, high and low, which is essentially a form of binarization. Specifically, the electromagnetic response curve at each sampling frequency point is binarized based on a threshold; values ​​less than the threshold are set as the first value, and others as the second value, thus obtaining the target electromagnetic response curve. For example, to obtain |S 11 The value at each sampling frequency point is processed into two values: 0dB and -10dB. -10dB is set as a threshold; if the value is less than -10dB, it is set to -10dB; otherwise, it is set to 0dB. This square wave processing makes |S 11 The curve makes it easy to identify the number of passbands and bandwidth.

[0088] Based on the aforementioned steps and the electromagnetic response curve, the objective function is calculated, including:

[0089] Calculate the objective function based on the target electromagnetic response curve.

[0090] S40: Determine whether the objective function has reached the maximum number of iterations.

[0091] S50: If the judgment result is negative, update the structural parameters according to the update mechanism, obtain the topological binary matrix of the frequency selective surface structure and the corresponding electromagnetic response curve, and return the topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure, until the judgment result is positive.

[0092] S60: Select the topological binary matrix of the surface structure and the corresponding electromagnetic response curve based on the frequency to construct a training dataset.

[0093] In the specific implementation process, after the objective function calculation is completed, it is necessary to determine whether the maximum number of iterations has been reached. If not, an evolutionary algorithm update mechanism is used to update the structural parameters, and topology modeling and full-wave simulation are performed again. In each iteration, the binary matrix of the FSS structure topology distribution and the electromagnetic response obtained from the simulation are retained. Iteration continues until the maximum number of iterations is reached. At this point, a training dataset can be constructed, which consists of the optimized FSS structure binary matrix and the corresponding electromagnetic response.

[0094] In this embodiment, the inverse mapping relationship between the electromagnetic response curve and the topological distribution of the frequency-selective surface structure is utilized. First, topological modeling and full-wave simulation are performed, and a binary matrix is ​​used to represent the topological structure. In the iteration of the objective function, the topological binary matrix and electromagnetic response retained in each iteration are used as training datasets. This can retain more useful design information, and the electromagnetic response has more similar characteristics to the design target, with less misleading information, thus improving the quality of training data in electromagnetic device design.

[0095] After obtaining the training dataset, a deep learning model is constructed. In this embodiment, the deep learning model MLP-mixer is used as an example. |S 11 The structure topology is mapped to FSS. MLP-Mixer mainly consists of three parts: a fully connected module, a hybrid architecture module, and a structure output module.

[0096] The fully connected module segments the input electromagnetic response parameters and feeds them into the network. Each segment is encoded through fully connected layers with different parameters, resulting in multiple encoded results. These results are then combined and input into the hybrid architecture module. The hybrid architecture module consists of multiple hybrid layers, each using two types of MLP blocks: spatial hybrid MLP and channel hybrid MLP. Spatial hybrid MLP mixes data blocks from different spatial locations, while channel hybrid MLP mixes data blocks from different channels, operating independently on each data block. Spatial hybrid MLP and channel hybrid MLP are interleaved within the hybrid layers. The structure prediction module consists of a global average pooling layer, a fully connected layer, and a binarization layer. The global average pooling layer reduces the dimensionality of the electromagnetic response parameter features extracted by the hybrid architecture module, and then the fully connected layer is used to obtain the initial frequency-selective surface structure.

[0097] Specifically, in this embodiment, when applying the MLP-mixer to the FSS design, the input |S| of size 1×1000 is first connected via a fully connected module. 11 The layer is evenly divided into 250 non-overlapping blocks. Then, by setting the hidden layer dimension to 1024, a two-dimensional table is exported. Spatial mixing MLP and channel mixing MLP will be performed separately. and The mapping is then performed. Finally, the structural parameters of the FSS will be output by the structural prediction layer. For modeling, the discrete values ​​of the binary matrix of the FSS's topological distribution need to be quantized to 0 or 1. Therefore, the output topological distribution parameters are calculated using the following formula:

[0098]

[0099] Where x is the output value of the classifier module, and using the trained MLP-mixer, the input target |S 11 |, and you can get the binary matrix of FSS.

[0100] To fully demonstrate the effectiveness of the training data collection method based on evolutionary algorithms proposed in this invention, the same number of samples were collected using two methods based on CS and PSO-GA, as well as a random training data collection method. Then, the MLP-mixer was trained using these three training sets respectively, and its performance was evaluated.

[0101] In this embodiment, the amount of training data is set to be proportional to the number of geometric variables because more data is needed in a larger sampling space. Therefore, the data size for each FSS structure is set to 100×p, where p is the number of variables for each FSS structure. (See Appendix) Figure 4 - Appendix Figure 5It is known that each FSS structure has [3, 2, 4, 5, 5] variables, and 2400 samples are collected from each of the three data collection methods for training. 240 samples (1 / 10 of the training data) are randomly selected for testing, and all calculations are performed on the same computer.

[0102] During training, the learning rate was set to 5×10. -4 The mean absolute percentage error (MAPE) is used as the loss function, which can be expressed as follows:

[0103] MAPE =

[0104] in, This represents the binary matrix representing the predicted FSS structure of the nth sample. This represents the actual FSS structure binary matrix of the nth sample in the training dataset, where N is the total number of samples in the training dataset.

[0105] For testing, the output is first obtained through MATLAB-CST co-simulation. Inputting it into CST, the derived electromagnetic response is |S 11 The test error is calculated using the absolute percentage error (APE). APE can be expressed as:

[0106] APE = m∈

[0107] in, Let represent the set of passband frequency points obtained from the simulation of the m-th predicted FSS structure. Let M represent the set of passband frequency points obtained from the simulation of the m-th actual FSS structure, where M is the total number of samples in the test dataset.

[0108] To compare performance, MLP-Mixer was trained using the three different datasets mentioned above and tested using the same 240 test samples. The trained MLP-Mixer was evaluated using APE, and the performance comparison is shown in Table 1.

[0109] Table 1 - Performance Comparison Table

[0110]

[0111] For the MLP-Mixer trained using the CS-based training data collection method, 54.17% of the samples in the test dataset had an error of ≤5%, and 62.50% had an error of ≤10%. For the MLP-Mixer trained using the PSO-GA-based training data collection method, 50.00% of the test data samples had an error of ≤5%, and 62.92% had an error of ≤10%. However, for the MLP-Mixer trained using the random training data collection method, only 35.00% of the samples in the test dataset had an error of ≤5%, and only 47.08% had an error of ≤10%. With an error of ≤5%, the MLP-Mixer test accuracy using the CS-based and PSO-GA-based training data collection methods was 25% and 19.17% higher than that using the random training data collection method, respectively. Meanwhile, with an error ≤10%, the test accuracy of the CS-based training data collection method and the PSO-GA-based training data collection method improved by 15.42% and 15.84%, respectively. If the goal is to achieve the same test error level as the proposed training data collection method using a randomized training data collection method, the number of full-wave simulations required for the former will be approximately twice that of the latter.

[0112] To further demonstrate the test performance, see attached Figure 7 - Appendix Figure 10 The effectiveness of the method provided in the embodiments of this application is verified, and attached... Figure 7 - Appendix Figure 8 A CS-based training data collection method is presented, along with... Figure 9 - Appendix Figure 10 Some test examples of the training data collection method based on PSO-GA are presented. The left side of the examples compares the actual and predicted structures, while the right side compares the electromagnetic responses of the actual and predicted structures. The APE values ​​of the four examples are 1.64%, 1.81%, 0.49%, and 1.92%, respectively. Due to the non-uniqueness of the inverse problem, some discrepancies between the predicted and actual structures may be observed in the test dataset. These results demonstrate that the training data acquisition method proposed in this application is effective and can improve the performance of DNN models.

[0113] See attached document Figure 11 Based on the same inventive concept as in the foregoing embodiments, this application also provides a training data acquisition device for electromagnetic device design, comprising:

[0114] The initialization module is used to initialize structure parameters;

[0115] The simulation module is used for topological modeling and full-wave simulation of frequency-selective surface structures based on structural parameters, and to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure.

[0116] The calculation module is used to calculate the objective function based on the electromagnetic response curve; the objective function is used to characterize the optimization direction.

[0117] The judgment module is used to determine whether the objective function has reached the maximum number of iterations;

[0118] The iterative module is used to update the structural parameters according to the update mechanism when the judgment result is negative, obtain the topological binary matrix of the frequency selective surface structure and the corresponding electromagnetic response curve, and return the topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure, until the judgment result is positive.

[0119] The module is used to select the topological binary matrix of the surface structure and the corresponding electromagnetic response curve based on the frequency, and to build the training dataset.

[0120] Those skilled in the art should understand that the division of the various modules in the embodiments is merely a logical functional division. In actual applications, they can be fully or partially integrated into one or more actual carriers. These modules can be implemented entirely in software through processing unit calls, entirely in hardware, or a combination of software and hardware. It should be noted that each module in the training data acquisition device for electromagnetic device design in this embodiment corresponds one-to-one with each step in the training data acquisition method for electromagnetic device design in the aforementioned embodiments. Therefore, the specific implementation of this embodiment can refer to the implementation of the training data acquisition method for electromagnetic device design described above, and will not be repeated here.

[0121] Based on the same inventive concept as in the foregoing embodiments, embodiments of this application also provide a computer-readable storage medium storing a computer program. When the computer program is loaded and executed by a processor, it implements the training data acquisition method for electromagnetic device design provided in the embodiments of this application.

[0122] Based on the same inventive concept as in the foregoing embodiments, embodiments of this application also provide an electronic device, including a processor and a memory, wherein,

[0123] Memory is used to store computer programs;

[0124] The processor is used to load and execute computer programs to cause the electronic device to perform a training data acquisition method for electromagnetic device design, as provided in the embodiments of this application.

[0125] In some embodiments, the computer-readable storage medium may be a memory such as FRAM, ROM, PROM, EPROM, EEPROM, flash memory, magnetic surface memory, optical disk, or CD-ROM; or it may be a device including one or any combination of the above-mentioned memories. The computer may be a variety of computing devices, including smart terminals and servers.

[0126] In some embodiments, executable instructions may take the form of a program, software, software module, script, or code, written in any form of programming language (including compiled or interpreted languages, or declarative or procedural languages), and may be deployed in any form, including as a standalone program or as a module, component, subroutine, or other unit suitable for use in a computing environment.

[0127] As an example, executable instructions may, but do not necessarily, correspond to files in the file system. They may be stored as part of a file that holds other programs or data, for example, in one or more scripts in a Hyper Text Markup Language (HTML) document, in a single file dedicated to the program in question, or in multiple collaborative files (e.g., a file that stores one or more modules, subroutines, or code sections).

[0128] As an example, executable instructions can be deployed to execute on a single computing device, or on multiple computing devices located in one location, or on multiple computing devices distributed across multiple locations and interconnected via a communication network.

[0129] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or system that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or system. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or system that includes that element.

[0130] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0131] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk) and includes several instructions to cause a multimedia terminal device (which may be a mobile phone, computer, television receiver, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0132] In summary, this application provides a method, apparatus, medium, and device for obtaining training data for electromagnetic device design. The method includes: initializing structural parameters; performing topological modeling and full-wave simulation of a frequency-selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure; calculating an objective function based on the electromagnetic response curve, wherein the objective function is used to characterize the optimization direction; determining whether the objective function has reached the maximum number of iterations; if the determination result is negative, updating the structural parameters according to an update mechanism to obtain the topological binary matrix and corresponding electromagnetic response curve of the frequency-selective surface structure, and returning to the step of performing topological modeling and full-wave simulation of the frequency-selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure, until the determination result is positive; and constructing a training dataset based on the topological binary matrix and corresponding electromagnetic response curve of the frequency-selective surface structure. This application utilizes the inverse mapping relationship between electromagnetic response curves and the topological distribution of frequency-selective surface structures. First, topological modeling and full-wave simulation are performed, and a binary matrix is ​​used to represent the topological structure. In the iteration of the objective function, the topological binary matrix and electromagnetic response retained in each iteration are used as training datasets. This can retain more useful design information, and the electromagnetic response has more similar characteristics to the design target, with less misleading information, thus improving the quality of training data in electromagnetic device design.

[0133] The above description is only a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. A method for obtaining training data applied to the design of electromagnetic devices, characterized in that, Includes the following steps: Initialize structure parameters; Based on the structural parameters, topological modeling and full-wave simulation of the frequency selective surface structure are performed to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure. Based on the electromagnetic response curve, the objective function is calculated; wherein, the objective function is used to characterize the optimization direction; Determine whether the objective function has reached the maximum number of iterations; If the result of the judgment is negative, the structural parameters are updated according to the update mechanism to obtain the topological binary matrix of the frequency selective surface structure and the corresponding electromagnetic response curve, and the steps of performing topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure are returned until the result of the judgment is positive. A training dataset is constructed by selecting the topological binary matrix of the surface structure based on the frequency and the corresponding electromagnetic response curve.

2. The method for obtaining training data for electromagnetic device design according to claim 1, characterized in that, Before calculating the objective function based on the electromagnetic response curve, the method further includes: The electromagnetic response curve is processed into a square wave at each sampling frequency point to obtain the target electromagnetic response curve. The step of calculating the objective function based on the electromagnetic response curve includes: Calculate the objective function based on the target electromagnetic response curve.

3. The method for obtaining training data for electromagnetic device design according to claim 2, characterized in that, The step of processing the electromagnetic response curve at each sampling frequency point using square wave processing to obtain the target electromagnetic response curve includes: The electromagnetic response curve is binarized at each sampling frequency point based on a threshold. Values ​​less than the threshold are set as the first value, and otherwise as the second value, to obtain the target electromagnetic response curve.

4. The method for obtaining training data for electromagnetic device design according to claim 1, characterized in that, The process of performing topological modeling and full-wave simulation of the frequency-selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency-selective surface structure includes: Based on the structural parameters, a topological model of the frequency-selective surface structure is performed to obtain the topological structure of the frequency-selective surface structure. The topological structure is binarized based on the binary representation rules to obtain the topological binary matrix of the frequency-selective surface structure.

5. The method for obtaining training data for electromagnetic device design according to claim 4, characterized in that, Before performing binary representation of the topology based on binary representation rules to obtain the topological binary matrix of the frequency-selective surface structure, the method further includes: The binary characterization rule is obtained by selecting the metal distribution of the surface structure sample based on the frequency.

6. The method for obtaining training data for electromagnetic device design according to claim 5, characterized in that, The process of selecting the metal distribution of surface structure samples based on frequency to obtain the binary characterization rule includes: The stacked structure of the surface structure sample is selected based on frequency, the metal layer is identified, and the metal layer is discretized into multiple pixel-level metal sheets; The binary representation rule is obtained based on the metal distribution of the metal sheet representation region described in the topology modeling.

7. The method for obtaining training data for electromagnetic device design according to claim 1, characterized in that, The step of calculating the objective function based on the electromagnetic response curve includes: Based on the electromagnetic response curve, determine the optimization direction; Based on the optimization direction and importance, the sub-functions and their weights are obtained; The objective function is calculated based on the sub-functions and their weights.

8. A training data acquisition device for electromagnetic device design, characterized in that, include: The initialization module is used to initialize structure parameters; The simulation module is used to perform topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters, and to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure. The calculation module is used to calculate the objective function based on the electromagnetic response curve; wherein the objective function is used to characterize the optimization direction. The judgment module is used to determine whether the objective function has reached the maximum number of iterations; The iterative module is used to update the structural parameters according to the update mechanism when the judgment result is negative, obtain the topological binary matrix of the frequency selective surface structure and the corresponding electromagnetic response curve, and return the topological modeling and full-wave simulation of the frequency selective surface structure based on the structural parameters to obtain the topological binary matrix and electromagnetic response curve of the frequency selective surface structure, until the judgment result is positive. A construction module is used to select the topological binary matrix of the surface structure and the corresponding electromagnetic response curve according to the frequency, and construct a training dataset.

9. A computer-readable storage medium storing a computer program, characterized in that, When the computer program is loaded and executed by the processor, it implements the training data acquisition method for electromagnetic device design as described in any one of claims 1-7.

10. An electronic device, characterized in that, Including processor and memory, among which, The memory is used to store computer programs; The processor is used to load and execute the computer program to cause the electronic device to perform the training data acquisition method for electromagnetic device design as described in any one of claims 1-7.

Images