Neural network construction method and microwave absorbing material structure optimization method based on the method

By optimizing the structure of the absorbing material through a continuous learning strategy and differential evolution algorithm, the problem of neural network forgetting under different data distributions was solved, thereby improving the prediction ability and optimizing the absorption performance.

CN123120123APending Publication Date: 2026-11-17SHANGHAI INST OF CERAMIC CHEM & TECH CHINESE ACAD OF SCI +1
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202510625778.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2025-05-15
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

In existing technologies, neural networks tend to forget previously learned information when processing the performance prediction of absorbing materials with different data distributions, which affects their prediction ability.

Method used

A continuous learning strategy is adopted, consisting of a replay sample selection phase, a learning and consolidation phase, and a reselection and learning and consolidation phase. This strategy combines the Latin hypercube method and nearest neighbor search to generate uniform sampling points, and the differential evolution algorithm is used to optimize the structure of the absorbing material.

Benefits of technology

This improves the ability of neural networks to predict the absorption performance of microwave absorbing materials, reduces the forgetting phenomenon, saves time and manpower costs, and optimizes the structural design of microwave absorbing materials.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123120123A_ABST
    Figure CN123120123A_ABST
Patent Text Reader

Abstract

This invention relates to the field of microwave absorbing material preparation, specifically to a neural network construction method and a microwave absorbing material structure optimization method based on this method. The neural network sample construction method includes: a replay sample selection stage: generating uniform sampling points, matching the uniform sampling points with sample points in the old sample to obtain replay sample points; a learning and consolidation stage: combining the replay sample points with the data of the current stage to participate in the training of the neural network; and a reselection and learning and consolidation stage: repeating the steps of the replay sample selection stage and the learning and consolidation stage, merging all sample points into the old sample. This invention can improve the predictive ability of the neural network for the microwave absorbing performance of the microwave absorbing material.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of microwave absorbing material preparation, specifically to a neural network construction method and a microwave absorbing material structure optimization method based on the method. Background Technology

[0002] Microwave-absorbing materials can significantly dissipate the energy of incident electromagnetic waves and are widely used in military, communications, and medical fields. However, traditional methods for designing microwave-absorbing material structures are time-consuming and labor-intensive.

[0003] Neural networks possess strong learning, generalization, and adaptability, providing new solutions for the structural design of microwave absorbing materials. As an important technology for realizing artificial intelligence, neural networks can uncover the mapping relationship between the structural parameters of microwave absorbing materials and their absorption performance, and have broad application prospects in the field of microwave absorbing materials.

[0004] While existing technologies utilize a combination of artificial intelligence and computational electromagnetics to shorten the time and reduce manpower costs in the metamaterial simulation and design phase, the construction of neural networks for predicting microwave absorption performance is generally based on static, identically distributed data. This means that when learning new data with different distributions, the neural network may forget previously learned information, thus affecting its ability to predict microwave absorption performance. Summary of the Invention

[0005] This invention provides a neural network construction method and a method for optimizing the structure of microwave absorbing materials based on the method, which can improve the ability of neural networks to predict the microwave absorption performance of microwave absorbing materials.

[0006] The first aspect of this invention provides a method for constructing neural network samples, comprising: Replay sample selection stage: uniform sampling points are generated by sampling, and the uniform sampling points are matched with the sample points in the old samples to obtain replay sample points; Learning consolidation phase: The replayed sample points are combined with the data from the current phase to participate in the training of the neural network; Then select the learning and consolidation phase: repeat the steps of the sample selection phase and the learning and consolidation phase, and merge all sample points into the old samples.

[0007] As an optional implementation, the replay sample selection stage includes: generating uniform sampling points by sampling using the Latin hypercube method.

[0008] As an optional implementation, the replay sample selection stage also includes: A nearest neighbor search is constructed for the sample points in the old sample to obtain the nearest neighbors of the sample points in the old sample. The uniform sampling points and the nearest neighbor points are matched to obtain the replay sample points.

[0009] A second aspect of the present invention provides a method for optimizing microwave absorbing materials based on the neural network sample construction method described in the first aspect of the present invention, comprising: The trained neural network is used as the evaluation function; With strong wave absorption performance as the target, the structure of the wave absorbing material is continuously optimized using a differential evolution algorithm. The strong wave absorption performance includes at least the wave absorbing material having a reflectivity lower than a preset reflectivity r in a preset frequency band and having the maximum wave absorption performance.

[0010] As an optional implementation, the step of using the trained neural network as an evaluation function, with strong microwave absorption performance as the objective, and continuously optimizing the structure of the microwave absorbing material using a differential evolution algorithm includes: Set sampling points according to preset unit frequency bands within the preset frequency band range; Obtain the structural parameters a of the absorbing material 1~3 ,w,r 1~3 ,d 1~3 ; The reflectivity of the absorbing material at a single sampling point is obtained based on the evaluation function and structural parameters: f(a 1~3 ,w,r 1~3 ,d 1~3 )-+r; Configure the penalty function P; Using the minimum sum of reflectivity of the absorbing material at each sampling point in a preset frequency band as the basic objective function, an objective function with a penalty function is constructed:

[0011] As an optional implementation, the method for optimizing the absorbing material further includes dividing the preset frequency band into low-frequency band, mid-frequency band, and high-frequency band; The method of constructing an objective function with a penalty function, using the minimum sum of reflectivity of the absorbing material at each sampling point in a preset frequency band as the basic objective function, further includes:

[0012] As an optional implementation, the low frequency band is less than 2 GHz, the high frequency band is greater than 18 GHz, and the intermediate frequency band is between the low frequency band and the high frequency band.

[0013] A third aspect of the present invention provides a neural network sample construction system, comprising: The replay sample selection unit is used at least to generate uniform sampling points, and to match the uniform sampling points with the sample points in the old samples to obtain replay sample points. The learning consolidation unit is used at least to jointly train the neural network with the replay sample points and the data of the current stage; the next learning consolidation unit is used at least to repeat the steps of the replay sample selection stage and the learning consolidation stage, and to incorporate all sample points into the old samples.

[0014] A fourth aspect of the present invention provides a microwave absorbing material optimization system, comprising: The replay sample selection unit is used at least to sample and generate uniform sampling points, and to match the uniform sampling points with the sample points in the old samples to obtain replay sample points. The learning consolidation unit is used at least to jointly train the neural network with the replayed sample points and the data of the current stage; the evaluation function determination unit is used at least to use the trained neural network as an evaluation function. The structural optimization unit is used to continuously optimize the structure of the absorbing material using a differential evolution algorithm with the goal of strong absorption performance. The strong absorption performance includes at least the absorbing material having a reflectivity lower than a preset reflectivity r and the maximum absorption performance in a preset frequency band.

[0015] A fifth aspect of the present invention provides a readable storage medium storing a computer program, characterized in that the computer program is executed by a processor as described in the first or second aspect of the present invention. Attached Figure Description

[0016] Figure 1 A flowchart illustrating a method for constructing neural network samples according to an embodiment of the present invention is shown.

[0017] Figure 2 The diagram illustrates the operational principle of a neural network sample construction method according to an embodiment of the present invention.

[0018] Figure 3 The diagram shows a flowchart of an absorbing material optimization method for a neural network sample construction method according to an embodiment of the present invention.

[0019] Figure 4 A diagram of a BP neural network structure is shown in one embodiment of the present invention.

[0020] Figure 5 The diagram shows the performance of a neural network trained using a continuous learning strategy in one embodiment of the present invention on various sample sets.

[0021] Figure 6 An absorption performance spectrum is shown in one embodiment of the present invention.

[0022] Figure 7 The diagram shows another absorption performance spectrum in one embodiment of the present invention.

[0023] Figure 8 A block diagram of a neural network sample construction system according to an embodiment of the present invention is shown.

[0024] Figure 9 A block diagram of an absorbing material optimization system according to an embodiment of the present invention is shown. Detailed Implementation

[0025] The present invention will be further illustrated by the following embodiments. It should be understood that the following embodiments are for illustrative purposes only and are not intended to limit the present invention.

[0026] like Figure 1 and Figure 2 As shown, the first aspect of this invention provides a method for constructing neural network samples. This method redesigns the training method of the neural network based on the replay method to achieve continuous learning of the neural network on sample sets with different data distributions. When the neural network trains on the current sample set, the proposed continuous learning strategy can be divided into three training stages: replay sample selection stage, learning consolidation stage, and reselection learning consolidation stage. These three stages specifically correspond to the following three steps.

[0027] Step S100: Replay Sample Selection Stage: Generate uniform sampling points, and match the uniform sampling points with the sample points in the old samples to obtain replay sample points.

[0028] Specifically, in the replay sample selection stage, the strategy for selecting replay samples is often crucial. If the selected replay samples are concentrated in a specific data distribution, the neural network may suffer from forgetting due to the failure to consolidate other old samples in a timely manner. Furthermore, the forgetting rate of information learned in the earliest stages is usually higher than that of information learned in the most recent stages. Considering the above, this invention divides the number of replay samples into multiple parts according to different stages, thereby consolidating the knowledge at each stage. Simultaneously, the number of replay samples in each part decreases exponentially with increasing stages, with a decay rate of λ. A larger number of replay samples can increase the coverage of samples from the earliest stages, better consolidate the information learned in the earliest stages, and minimize forgetting. However, the spatial distribution of old samples in each stage is irregular. The extraction of replay samples needs to consider the spatial relationship of each instance, ensuring that the replay samples are as spatially uniform as possible.

[0029] Therefore, in one embodiment of the present invention, uniform sampling points are generated by sampling using the Latin hypercube method.

[0030] Latin Hypercube Sampling (LHS) is a stratified sampling strategy widely used in many fields such as engineering and computer science, and can effectively reduce the number of computationally intensive evaluations.

[0031] Furthermore, in practice, it is unrealistic to expect the generated uniform sampling points to perfectly match the old samples. To minimize disruption to the distribution characteristics of the points generated by the Latin hypercube method, one embodiment of the present invention further includes the following in the replay sample selection stage: A nearest neighbor search is constructed for the sample points in the old sample to obtain the nearest neighbors of the sample points in the old sample. The uniform sampling points and the nearest neighbor points are matched to obtain the replay sample points.

[0032] Specifically, this invention achieves nearest neighbor search by constructing a KD tree for old samples. By selecting the nearest neighbor of the generated point in the old samples, the spatial distribution uniformity of the replay sample set can be preserved to a certain extent.

[0033] Step S200: Learning consolidation phase: The replay sample points are combined with the data of the current phase to participate in the training of the neural network.

[0034] Specifically, in the learning and consolidation phase, by jointly training the replay sample points selected in step S100 with the data of the current phase, the neural network can be guaranteed to learn the current information while mitigating the forgetting of previous information to the greatest extent.

[0035] Step S300: Select the learning and consolidation phase again: Repeat the steps of the sample selection phase and the learning and consolidation phase to merge all sample points into the old samples.

[0036] Specifically, in the reselection learning consolidation phase, by repeating the first and second phases, the neural network can access a wider sample space and strengthen its memory of old information through multiple selections. After the current phase of continuous learning is completed, the samples from that phase are incorporated into the old samples.

[0037] Here, by selecting replay samples from old samples according to a certain strategy, the neural network can continuously learn and improve its ability to predict the absorption performance of absorbing materials.

[0038] like Figure 3 As shown, the second aspect of the present invention provides a method for optimizing absorbing materials based on the neural network sample construction method described in any of the above embodiments. The method uses the trained neural network as an evaluation function, takes strong absorbing performance as the target, and uses a differential evolution algorithm to continuously optimize the structure of the absorbing material. The absorbing material in the present invention is a square ring shape, and specifically includes the following steps.

[0039] Step S400: Use the trained neural network as the evaluation function;

[0040] Step S500: Using strong absorption performance as the target, the structure of the absorbing material is continuously optimized using a differential evolution algorithm. The strong absorption performance includes at least the absorbing material having a reflectivity lower than a preset reflectivity r within a preset frequency band and maximizing its absorption performance. Excellent absorption performance typically aims for an overall reflection of less than -10dB in the 1–40GHz frequency band. Those skilled in the art can select absorption performance data based on actual technical requirements.

[0041] In one application scenario of this invention, when the reflectivity of the absorbing material is less than -10dB in a certain frequency band, it means that more than 90% of the incident microwave energy is absorbed, and only such an absorbing material can meet the actual needs.

[0042] Therefore, with the design goal of achieving a reflectivity of less than -10dB and maximizing absorption performance in the 2–18GHz frequency band, optimizing the structural parameters of the absorbing material can be categorized as a constrained optimization problem. The preset frequency band range can be set according to actual conditions; for example, 2–18GHz and 1–10GHz are preset frequency band ranges, while -10dB and -15dB are preset feedback rates. The highest absorption performance is achieved when the sum of reflection attenuation at all frequency points is minimized.

[0043] In one embodiment of the present invention, the step of using the trained neural network as an evaluation function, with strong microwave absorption performance as the objective, and continuously optimizing the structure of the microwave absorbing material using a differential evolution algorithm includes: Set sampling points within a preset frequency band range according to a preset unit frequency band, for example, the preset unit frequency band is 1GHz; Obtain the structural parameters a of the absorbing material 1~3 ,w,r 1~3 ,d 1~3 Where a represents the outer dimension of the square ring, w represents the ring width of the square ring, r represents the surface resistance value of the square ring, and d represents the thickness value of the dielectric; specifically, a1, a2, and a3 represent the outer dimensions of the first, second, and third layer square rings, respectively; d1, d2, and d3 represent the thickness values ​​of the first, second, and third layer dielectrics, respectively; and r1, r2, and r3 represent the surface resistance values ​​of the first, second, and third layer square rings, respectively. The reflectivity of the absorbing material at a single sampling point is obtained based on the evaluation function and structural parameters: f(a 1~3 ,w,r 1~3 ,d 1~3 )-+r; Configure the penalty function P; Using the minimum sum of reflectivity of the absorbing material at each sampling point in a preset frequency band as the basic objective function, an objective function with a penalty function is constructed:

[0044] Specifically, this invention selects the Differential Evolutionary Algorithm (DEA) as the optimization algorithm. DEA is an optimization algorithm based on swarm intelligence theory, which completes the global search through cooperation and competition among individuals in the swarm. It exhibits excellent performance in large-space, nonlinear, and non-differentiable continuous problems and is suitable for handling high-dimensional problems. This invention uses the above formula as the objective function of the optimization algorithm. The microwave absorbing material evaluation function f(x) in the objective function uses the aforementioned BP neural network to achieve a fast approximate evaluation of the microwave absorbing material's performance, thereby realizing continuous optimization of the microwave absorbing material's structural parameters.

[0045] Specifically, this invention implements constrained optimization by using a penalty function method. By adding a penalty function to the objective function, the constrained optimization problem is transformed into an unconstrained optimization problem. The penalty function will impose a penalty on solutions that violate the constraints, making the optimization algorithm tend to find solutions that satisfy the constraints.

[0046] For example, to facilitate quantization, sampling is performed at equal intervals of 1 GHz within the 2–18 GHz frequency band, with 17 sampling points. Assume f(a 1~3 ,w,r 1~3 ,d 1~3 ) is a structure parameter of a 1~3 ,w,r 1~3 ,d 1~3 The reflectivity of the absorbing material at a sampling point in a certain frequency band is then expressed by the expression f(a). 1~3 ,w,r 1~3 ,d 1~3 The degree of violation of the constraint of less than -10dB by the absorbing material in this frequency band is evaluated by 10, and the violation of the constraint is punished by a penalty coefficient P. The minimum sum of the reflectivity of the absorbing material at each sampling point in the frequency band from 2 to 18 GHz is the basic objective function.

[0047] The microwave absorption performance of absorbing materials depends on their electromagnetic parameters, including relative permittivity and relative permeability, which vary with frequency. At low frequencies (e.g., 1 GHz) and high frequencies (e.g., 19 GHz), the electromagnetic parameters may deviate from the optimal matching state, leading to a decrease in absorption performance. However, at other frequency bands, the electromagnetic parameters of the material change less, allowing for better matching of electromagnetic waves and resulting in stronger absorption performance.

[0048] In one embodiment of the present invention, the preset frequency band range is divided into a low-frequency band, a mid-frequency band, and a high-frequency band; wherein the low-frequency band is less than 2 GHz, the high-frequency band is greater than 18 GHz, and the mid-frequency band is between the low-frequency band and the high-frequency band.

[0049] In one embodiment of the present invention, the construction of an objective function with a penalty function, using the minimum sum of reflectivity of the absorbing material at each sampling point in a preset frequency band as the basic objective function, further includes:

[0050] For example, if designing absorbing materials with reflectivity below -10dB across the 2–18 GHz frequency band requires greater attention to their absorption performance in the low and high frequency bands. Therefore, if the reflectivity of the absorbing material violates the constraints in the low and high frequency bands, a larger penalty coefficient P' is applied. Thus, this invention, by designing penalty functions based on different frequency bands, facilitates the optimization of absorbing materials operating in different frequency bands.

[0051] Specifically, the absorbing material sample set used in this invention includes the structural parameters of the absorbing materials and their corresponding reflectivities. In one application scenario, the structural parameters of the absorbing materials were randomly generated under 14 different experimental settings, and the reflectivity of each absorbing material in the 2–18 GHz frequency band was calculated through simulation. The sample set contains a total of 73,889 data sets. The w value is the same in each stage of the samples, but the w value differs between different stages. As can be seen from the description, the data distribution is different in different stages. For example, this invention divides the training set and test set in an 8:2 ratio for each stage. The neural network used in this invention is specifically a BP neural network. The BP neural network continuously learns on the training set of each stage, and the test set of each stage is used to verify the performance of the BP neural network at each stage.

[0052] The neural network structure used in this invention is as follows: Figure 4 As shown in Table 1, the sample data input to the neural network is first preprocessed using min-max normalization to ensure that the dimensions of different features input into the neural network remain consistent, thus avoiding affecting the convergence of the neural network. The neural network uses mean squared error (MSE) as the loss function and the Adam optimizer as the optimization algorithm. The hyperparameter learning rate of the Adam optimizer is 0.0001, and the training batch size is set to 50. The specific implementation of the continuous learning strategy proposed in this invention is shown in Table 1 and... Figure 5 As shown, the neural network is trained using sample sets with different data distributions based on this continuous learning strategy. The LHS sample set refers to the sample set generated by the Latin hypercube sampling method, with a decay rate λ of 0.7.

[0053] Table 1

[0054] In one embodiment of the present invention, a trained neural network is used as an evaluation function, and the structural parameters of the absorbing material are continuously optimized using a differential evolution algorithm. The population size of the differential evolution algorithm is 50, the scaling factor is 0.8, the crossover rate is 0.7, and the maximum number of iterations is 100.

[0055] This invention, through a Latin hypercube sampling method combined with a nearest-neighbor replay sample selection strategy, ensures the spatial uniformity of replay samples within the old samples, maximizing the representation of the data distribution of each sample set. The design of exponentially decreasing replay sample numbers with each stage achieves deep consolidation of information from earlier stages, facilitating the avoidance of forgetting. Furthermore, the continuous optimization of the absorbing material structure parameters based on this neural network can quickly find the approximately optimal combination of absorbing material parameters that meets the target requirements, saving time and manpower costs.

[0056] To measure the effectiveness of the continuous learning strategy, this invention further employs average accuracy, let a k,j Let represent the evaluation accuracy of the model on the test set corresponding to stage j after learning through stages 1 to k. Then, the average accuracy can be defined as... The average accuracy is represented by the coefficient of determination. When judging performance, a coefficient of determination close to 1 indicates that the model performance is good, while a coefficient of determination less than 0 indicates that the model performance is very poor. To facilitate overall evaluation, it is limited to 0. k and j both represent the number of learning stages. j≥k+1 means that after completing k rounds of learning, validation is performed in j rounds.

[0057] The performance of the neural network trained based on the continuous learning strategy designed in this invention on the sample set is as follows: Figure 5 As shown, the results indicate that the neural network performs the worst on the earliest sample set, but still achieves an accuracy of 84.27%, which meets the needs of practical applications.

[0058] The present invention is further illustrated by providing specific examples. It should also be understood that the following examples are for illustrative purposes only and should not be construed as limiting the scope of protection of the present invention. Any non-essential improvements and adjustments made by those skilled in the art based on the above description of the present invention are within the scope of protection of the present invention.

[0059] To design materials with specific optical properties, the optimal values ​​of structural variables are explored under the constraints of target reflectivity and other structural variables within a specific frequency band.

[0060] by Using a neural network as the objective function and a differential evolution algorithm as the evaluation function, the structural parameters of the absorbing material are continuously optimized. The final optimization results are verified by simulation, and the absorption performance spectrum is shown below. Figure 6 As shown, it can be found that the reflectivity of the absorbing material is less than -10dB in the 2-18GHz frequency band, and the electromagnetic wave absorption rate reaches more than 90%, which meets the design requirements.

[0061] In one embodiment of the present invention, the thickness of the absorbing material is considered. For example, the thickness of the absorbing material is set to 20 mm as a further limiting condition. Under this limiting condition, the minimum reflectivity of less than -10 dB in the 2-18 GHz frequency band is studied. The value range of the thickness d is [0.1, 8]. Under the condition that d1+d2+d3=20, d3 is eliminated by using the property of equality constraint, that is, d3=20-d1-d2. Then d1 should satisfy d1∈[4,8], and d2 should satisfy d2∈[max(0.1,12-d1),min(8,19.9-d1)], unit: mm; where 20 mm is the maximum thickness of the total thickness in the design of the absorbing material, 0.1 mm is the minimum thickness of the medium of the absorbing material, 19.9 mm is the difference between the two thicknesses, and 12 mm and 8 mm are obtained from the maximum medium thickness of 8 mm set above, 12 mm=20 mm-8 mm.

[0062] Under the above conditions, the structural parameters of the absorbing material are optimized using the differential evolution algorithm, and similarly... The final optimization results, using the objective function as d1 = 6.2, d2 = 7.7, and d3 = 6.1, are as follows: Simulation of the final optimization results yields the corresponding absorption performance spectra. Figure 7 As shown in the figure. The results show that the material achieves an absorption rate of over 90% in the 2–18 GHz frequency band while maintaining a total thickness of 20 mm.

[0063] In summary, the continuous learning strategy proposed in this invention can effectively alleviate the forgetting phenomenon that occurs when neural networks learn from sample sets with different data distributions, and effectively improve the prediction performance of neural networks on various sample sets. Based on the design requirements and the characteristics of the absorbing material, an objective function is designed. The neural network trained through continuous learning is used as the evaluation function for the absorbing material in the objective function. The differential evolution algorithm is used to continuously optimize the structural parameters of the absorbing material, resulting in an absorbing material that meets practical requirements. Within a specified frequency band, this absorbing material can absorb a significant amount of incident microwave energy, thus satisfying the absorption requirements.

[0064] like Figure 8 As shown, a third aspect of the present invention provides a neural network sample construction system, comprising: The replay sample selection unit is used at least to generate uniform sampling points, and to match the uniform sampling points with the sample points in the old samples to obtain replay sample points. The learning consolidation unit is used at least to jointly train the neural network with the replay sample points and the data of the current stage; the next learning consolidation unit is used at least to repeat the steps of the replay sample selection stage and the learning consolidation stage, and to incorporate all sample points into the old samples.

[0065] like Figure 9 As shown, a fourth aspect of the present invention provides a microwave absorbing material optimization system, comprising: The replay sample selection unit is used at least to sample and generate uniform sampling points, and to match the uniform sampling points with the sample points in the old samples to obtain replay sample points. The learning consolidation unit is used at least to jointly train the neural network with the replayed sample points and the data of the current stage; the evaluation function determination unit is used at least to use the trained neural network as an evaluation function. The structural optimization unit is used to continuously optimize the structure of the absorbing material using a differential evolution algorithm with the goal of strong absorption performance. The strong absorption performance includes at least the absorbing material having a reflectivity lower than a preset reflectivity r and the maximum absorption performance in a preset frequency band.

[0066] A fifth aspect of the present invention provides a readable storage medium storing a computer program that is executed by a processor as described in any of the above embodiments.

[0067] It is understood that computer-readable storage media can include: any entity or device capable of carrying computer programs, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media, etc. Computer programs include computer program code. Computer program code can be in the form of source code, object code, executable files, or certain intermediate forms, etc. Computer-readable storage media can include: any entity or device capable of carrying computer program code, recording media, USB flash drives, portable hard drives, magnetic disks, optical disks, computer memory, read-only memory (ROM), random access memory (RAM), and software distribution media, etc.

[0068] In some embodiments of the present invention, the electronic device may include a controller or a processor. The controller is a microcontroller chip that integrates a processor, memory, communication module, etc. The processor may refer to the processor included in the controller. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc.

[0069] Any process or method description in the flowchart or otherwise herein can be understood as representing a module, segment, or portion of code comprising one or more executable instructions for implementing a particular logical function or process, and the scope of the preferred embodiments of the invention includes additional implementations in which functions may be performed not in the order shown or discussed, including substantially simultaneously or in reverse order depending on the functions involved, as will be understood by those skilled in the art to which embodiments of the invention pertain.

[0070] Those skilled in the art will recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of each example have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementations should not be considered beyond the scope of this invention.

[0071] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing neural network samples, characterized in that, include: Replay sample selection stage: uniform sampling points are generated by sampling, and the uniform sampling points are matched with the sample points in the old samples to obtain replay sample points; Learning consolidation phase: The replayed sample points are combined with the data from the current phase to participate in the training of the neural network; Then select the learning and consolidation phase: repeat the steps of the sample selection phase and the learning and consolidation phase, and merge all sample points into the old samples.

2. The neural network sample construction method according to claim 1, characterized in that, During the replay sample selection phase: Uniform sampling points are generated using the Latin hypercube method.

3. The neural network sample construction method according to claim 1, characterized in that, The replay sample selection phase also includes: A nearest neighbor search is constructed for the sample points in the old sample to obtain the nearest neighbors of the sample points in the old sample. The uniform sampling points and the nearest neighbor points are matched to obtain the replay sample points.

4. A method for optimizing microwave absorbing materials based on the neural network sample construction method according to any one of claims 1-3, characterized in that, include: The trained neural network is used as the evaluation function; With strong wave absorption performance as the target, the structure of the wave absorbing material is continuously optimized using a differential evolution algorithm. The strong wave absorption performance includes at least the wave absorbing material having a reflectivity lower than a preset reflectivity r in a preset frequency band and having the maximum wave absorption performance.

5. The method for optimizing absorbing materials according to claim 4, characterized in that, The step of using the trained neural network as an evaluation function, with strong microwave absorption performance as the objective, and continuously optimizing the structure of the microwave absorbing material using a differential evolution algorithm includes: Set sampling points according to preset unit frequency bands within the preset frequency band range; Obtain the structural parameters a of the absorbing material 1~3 ,w,r 1~3 ,d 1~3 ; The reflectivity of the absorbing material at a single sampling point is obtained based on the evaluation function and structural parameters: f(a 1~3 ,w,r 1~3 ,d 1~3 )-+r; Configure the penalty function P; Using the minimum sum of reflectivity of the absorbing material at each sampling point in a preset frequency band as the basic objective function, an objective function with a penalty function is constructed:

6. The method for optimizing absorbing materials according to claim 5, characterized in that, It also includes dividing the preset frequency band range into low-frequency band, mid-frequency band, and high-frequency band; The method of constructing an objective function with a penalty function, using the minimum sum of reflectivity of the absorbing material at each sampling point in a preset frequency band as the basic objective function, further includes:

7. The method for optimizing absorbing materials according to claim 6, characterized in that, The low-frequency band is less than 2 GHz, the high-frequency band is greater than 18 GHz, and the intermediate frequency band is between the low-frequency band and the high-frequency band.

8. A neural network sample construction system, characterized in that, include: The replay sample selection unit is used at least to generate uniform sampling points, and to match the uniform sampling points with the sample points in the old samples to obtain replay sample points. The learning consolidation unit is used at least to jointly train the neural network with the replayed sample points and the data of the current stage; Then select a learning and consolidation unit, at least to repeat the steps of the sample selection phase and the learning and consolidation phase, and merge all sample points into the old sample.

9. A microwave absorbing material optimization system, characterized in that, include: The replay sample selection unit is used at least to sample and generate uniform sampling points, and to match the uniform sampling points with the sample points in the old samples to obtain replay sample points. The learning consolidation unit is used at least to jointly train the neural network with the replayed sample points and the data of the current stage; An evaluation function determination unit is used, at least to use the trained neural network as an evaluation function. The structural optimization unit is used to continuously optimize the structure of the absorbing material using a differential evolution algorithm with the goal of strong absorption performance. The strong absorption performance includes at least the absorbing material having a reflectivity lower than a preset reflectivity r and the maximum absorption performance in a preset frequency band.

10. A readable storage medium storing a computer program, characterized in that, The computer program is executed by a processor using the steps of the method as described in any one of claims 1-7.