Metasurface antenna parameter optimization method and system based on convolutional neural network

Abstract

The invention relates to the technical field of metasurface antennas, and provides a metasurface antenna parameter optimization method and system based on a convolutional neural network, and the method comprises the steps: collecting metasurface sample data; extracting a comprehensive electromagnetic feature set, and establishing a nonlinear mapping relation model; constructing an antenna performance comprehensive evaluation function, inputting the nonlinear mapping relation model and the antenna performance comprehensive evaluation function into a hybrid optimization algorithm to generate a parameter candidate set, and performing local optimization on the parameter candidate set by using a particle swarm algorithm to obtain a metasurface antenna parameter combination; and extracting electromagnetic characteristics of the metasurface antenna parameter combination, iteratively adjusting the height parameter of the resonant cavity until the height parameter meets a threshold value to obtain an electromagnetic simulation verification result, and feeding back the electromagnetic simulation verification result to the deep Q neural network model for parameter updating to obtain an optimal metasurface antenna parameter combination. According to the method, the optimization of antenna parameters is realized, the design efficiency of the metasurface antenna is improved, and the consumption of electromagnetic simulation calculation resources is reduced.

Description

Technical Field

[0001] The present invention relates to the technical field of metasurface antennas, and in particular to a metasurface antenna parameter optimization method and system based on convolutional neural networks. Background Art

[0002] Metasurface antennas are a novel antenna structure based on metamaterial technology. They manipulate the propagation characteristics of electromagnetic waves by designing artificial microstructure units on a specific surface. Compared to traditional antennas, metasurface antennas offer advantages such as low profile, high gain, strong directivity, and multi-band operation. They hold broad application prospects in modern wireless communications, radar detection, and the Internet of Things. The performance of metasurface antennas is highly dependent on the precise design of their structural parameters, including key parameters such as the surface unit period, resonant cavity height, and metamaterial unit size. These parameters exhibit complex electromagnetic coupling, impacting core performance indicators such as the antenna's operating frequency, bandwidth, gain, and impedance matching.

[0003] Existing technologies use traditional data-driven methods (such as regression analysis and interpolation) combined with electromagnetic simulation software for parameter mapping, achieving modeling of the relationship between basic antenna parameters and electromagnetic performance. However, this method suffers from significant drawbacks when processing high-dimensional electromagnetic field data, including limited feature extraction capabilities, a lack of deep spatial feature learning, and insufficient nonlinear mapping capabilities. This leads to an inaccurate understanding of the complex relationship between resonant cavity height and antenna performance, and an inability to achieve intelligent mapping between electromagnetic field distribution information and optimization parameters. This makes adaptive optimization of antenna parameters and precise control of performance indicators particularly difficult in multi-band scenarios. This ultimately results in inefficient metasurface antenna design, high resource consumption, and poorly stable optimization results. Summary of the Invention

[0004] In view of this, the present invention proposes a metasurface antenna parameter optimization method and system based on convolutional neural networks, which solves the significant shortcomings of the existing technology when processing high-dimensional electromagnetic field data, such as limited feature extraction capability, lack of deep spatial feature learning and insufficient nonlinear mapping capability, resulting in inaccurate grasp of the complex relationship between the resonant cavity height and antenna performance, and inability to achieve intelligent mapping between electromagnetic field distribution information and optimization parameters. Especially in multi-band scenarios, it is even more difficult to achieve adaptive optimization of antenna parameters and precise control of performance indicators, which ultimately leads to low efficiency in metasurface antenna design, high resource consumption and poor stability of optimization results.

[0005] The technical solution of the present invention is implemented as follows: In a first aspect, the present invention provides a method for optimizing the parameters of a metasurface antenna based on a convolutional neural network, comprising the following steps: Collecting raw sample data of the metasurface antenna and performing preprocessing to obtain metasurface sample data; Constructing a two-stream convolutional neural network model, extracting spatial domain features and frequency domain features from the metasurface sample data through the two-stream convolutional neural network model, constructing a comprehensive electromagnetic feature set based on the spatial domain features and the frequency domain features, and establishing a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height; Constructing a comprehensive evaluation function for antenna performance, inputting the nonlinear mapping relationship model and the comprehensive evaluation function for antenna performance into a hybrid optimization algorithm, generating a parameter candidate set through a genetic algorithm, and performing local optimization on the parameter candidate set using a particle swarm algorithm to obtain a metasurface antenna parameter combination; The electromagnetic characteristics of the metasurface antenna parameter combination are extracted, and the resonant cavity height parameter is iteratively adjusted based on the matching degree of gain and impedance bandwidth until the threshold is met to obtain the electromagnetic simulation verification result. A parameter optimization closed loop is constructed through the reinforcement learning mechanism, and the electromagnetic simulation verification results are fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination.

[0006] Based on the above technical solution, preferably, the collecting of metasurface antenna raw sample data and preprocessing to obtain metasurface sample data includes: Through electromagnetic simulation, metasurface antennas with different structural parameters and resonant cavity heights are simulated to obtain the corresponding electromagnetic field distribution diagrams and S-parameter curves, and the original sample data of the metasurface antenna is obtained; The original sample data set is standardized and enhanced to eliminate dimensional differences, and the data quality is improved through data noise reduction, missing value processing and image enhancement to construct a training metasurface sample data set. The metasurface sample data includes antenna structure parameters, electromagnetic field distribution diagram and S parameter curve.

[0007] On the basis of the above technical solution, preferably, the dual-stream convolutional neural network model is constructed, and spatial domain features and frequency domain features are extracted from the metasurface sample data by the dual-stream convolutional neural network model, a comprehensive electromagnetic feature set is constructed based on the spatial domain features and the frequency domain features, and a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height is established, including: Constructing a two-stream convolutional neural network, the two-stream convolutional neural network includes a spatial domain feature extraction branch and a frequency domain feature extraction branch. The spatial domain feature extraction branch uses three-dimensional convolution to process the electromagnetic field distribution map, and the frequency domain feature extraction branch uses one-dimensional convolution to process the S-parameter curve. The features of the two branches are merged through a feature fusion layer to construct a comprehensive electromagnetic feature set; The comprehensive electromagnetic feature set and the corresponding resonant cavity height data are used to train a nonlinear mapping model, wherein the nonlinear mapping model adopts a multi-layer perceptron structure.

[0008] On the basis of the above technical solution, preferably, the construction of the two-stream convolutional neural network includes a spatial domain feature extraction branch and a frequency domain feature extraction branch, wherein: The spatial domain feature extraction branch adopts a three-dimensional convolutional network with a residual structure, which consists of three convolution units. Each convolution unit contains two 3×3×3 convolutional layers, a batch normalization layer and a PReLU activation function, and a residual connection is set. The number of convolution kernels increases from the initial 32 to 128 layer by layer. The frequency domain feature extraction branch adopts a one-dimensional convolutional network enhanced by the attention mechanism, including a time attention module and a frequency attention module. The frequency domain feature extraction branch uses multi-scale convolution kernels to process the S-parameter curve in parallel to capture the antenna response characteristics in different frequency domains; the feature fusion layer adopts an adaptive weighting mechanism to integrate the features extracted by the spatial domain feature extraction branch and the frequency domain feature extraction branch, and dynamically adjusts the weights of different features according to the importance of specific frequency bands.

[0009] Based on the above technical solution, preferably, the antenna performance comprehensive evaluation function is constructed, the nonlinear mapping relationship model and the antenna performance comprehensive evaluation function are input into a hybrid optimization algorithm, a parameter candidate set is generated by a genetic algorithm, and a particle swarm algorithm is used to perform local optimization on the parameter candidate set to obtain a metasurface antenna parameter combination, including: A multi-objective antenna performance comprehensive evaluation function is constructed, and frequency band adaptive weight coefficients are introduced. Different evaluation criteria are set for bandwidth, gain, directivity, and impedance matching parameters in different frequency bands, and weighted calculations are performed to obtain a comprehensive evaluation function for antenna performance. A hybrid optimization algorithm framework is established, the nonlinear mapping relationship model is integrated with the comprehensive evaluation function of antenna performance, a genetic algorithm is used to generate a candidate set of metasurface antenna parameters, and the parameter space exploration capability of the candidate set of metasurface antenna parameters is enhanced through an adaptive mutation strategy to obtain several candidate parameters. The candidate parameters are locally optimized based on the quantum particle swarm algorithm to obtain the metasurface antenna parameter combination.

[0010] Based on the above technical solution, preferably, the step of constructing a comprehensive evaluation function of antenna performance includes: Based on the operating frequency band characteristics of the metasurface antenna, the target frequency range is divided into multiple sub-bands, and different priorities are set for each band. Normalized bandwidth indicators, gain indicators, directivity indicators, and impedance matching indicators are defined for each band. Band correlation analysis is used to identify the coupling effects between bands, and a performance balance model between bands is established. For multi-band application scenarios, a hierarchical evaluation structure is constructed. Principal component analysis is used to reduce the redundancy between evaluation indicators and extract key performance characteristics. Fuzzy comprehensive evaluation method is used to integrate the evaluation results of multiple bands and multiple indicators to obtain a single comprehensive evaluation function for antenna performance. The hybrid optimization algorithm framework is established as follows: After initializing the population, an improved genetic algorithm with adaptive crossover and mutation strategies is used to generate a candidate set of antenna parameters. The crossover probability and mutation probability are dynamically adjusted according to the population diversity and fitness distribution. The multi-objective optimization problem is handled through the Pareto dominance sorting mechanism, and the non-dominated solution set is retained. For each candidate solution in the non-dominated solution set, a nonlinear mapping relationship model is used to predict the optimal resonant cavity height; the predicted optimal resonant cavity height is combined with the metasurface antenna parameters to obtain a complete solution, and a quantum particle swarm algorithm is used for local optimization, in which the quantum rotation angle and shrinkage and expansion factor are adaptively adjusted based on the quality of the complete solution; a dynamic balance mechanism for solution space exploration and development is set, and the proportion of global search and local refinement is adjusted according to the optimization process.

[0011] Based on the above technical solution, preferably, the electromagnetic characteristics of the metasurface antenna parameter combination are extracted, and the resonant cavity height parameter is iteratively adjusted based on the matching degree of gain and impedance bandwidth until a threshold is met to obtain an electromagnetic simulation verification result. A parameter optimization closed loop is constructed through a reinforcement learning mechanism, and the electromagnetic simulation verification result is fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination, including: A real-time electromagnetic field distribution feature extraction mechanism was constructed to perform electromagnetic simulation on the metasurface antenna parameter combination generated by the hybrid optimization algorithm. The electromagnetic field distribution map and S-parameter curve were obtained in real time. The electromagnetic features were extracted using a pre-trained convolutional neural network model. An iterative adjustment strategy for the resonant cavity height parameters was constructed to adjust the resonant cavity height based on the gain and impedance bandwidth matching. A parameter optimization closed loop based on deep reinforcement learning is constructed, the comprehensive evaluation function of antenna performance is used as the reward signal, a state-action-reward mapping relationship is established, the metasurface antenna parameter optimization strategy is learned through a deep Q neural network, and the electromagnetic simulation verification results are fed back to the model weight update process to perform adaptive closed-loop optimization between the metasurface antenna parameters and electromagnetic characteristics.

[0012] In a second aspect, the present invention further provides a metasurface antenna parameter optimization system based on a convolutional neural network, the system comprising: A sample data acquisition module is used to collect raw sample data of the metasurface antenna and perform preprocessing to obtain metasurface sample data; a mapping model establishment module, configured to construct a dual-stream convolutional neural network model, extract spatial domain features and frequency domain features from the metasurface sample data through the dual-stream convolutional neural network model, construct a comprehensive electromagnetic feature set based on the spatial domain features and the frequency domain features, and establish a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height; An antenna parameter optimization module is used to construct a comprehensive evaluation function for antenna performance, input the nonlinear mapping relationship model and the comprehensive evaluation function for antenna performance into a hybrid optimization algorithm, generate a parameter candidate set through a genetic algorithm, and perform local optimization on the parameter candidate set using a particle swarm algorithm to obtain a metasurface antenna parameter combination; The simulation verification feedback module is used to extract the electromagnetic characteristics of the metasurface antenna parameter combination, iteratively adjust the resonant cavity height parameter based on the matching degree of gain and impedance bandwidth until the threshold is met, and obtain the electromagnetic simulation verification result. A parameter optimization closed loop is constructed through the reinforcement learning mechanism, and the electromagnetic simulation verification results are fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination.

[0013] In a third aspect, the present invention further provides an electronic device comprising: at least one processor, at least one memory, a communication interface, and a bus; Among them, the processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement steps such as a method for optimizing supersurface antenna parameters based on convolutional neural networks.

[0014] In a fourth aspect, the present invention also provides a computer-readable storage medium, which stores computer instructions, and the computer instructions enable a computer to implement the steps of a method for optimizing parameters of a metasurface antenna based on a convolutional neural network.

[0015] The present invention provides a method and system for optimizing metasurface antenna parameters based on convolutional neural networks, which has the following advantages over the prior art: (1) Through a dual-stream convolutional neural network, deep extraction and fusion of electromagnetic features in the spatial and frequency domains are achieved, and an accurate nonlinear mapping relationship between electromagnetic features and resonant cavity height is established. Combined with a comprehensive evaluation function, a hybrid optimization algorithm, and a reinforcement learning closed-loop mechanism, the antenna parameters are optimized, significantly improving the efficiency of metasurface antenna design and reducing the consumption of electromagnetic simulation computing resources. (2) Through the dual-stream convolutional neural network architecture and electromagnetic feature extraction mechanism, the multi-scale spatial structure analysis of the electromagnetic field distribution map and the deep mining of the frequency domain features of the S-parameter curve are realized. The feature fusion strategy based on the attention mechanism and the multi-layer perceptron are introduced to realize the adaptive weighting and integration of feature significance. The deep separation convolution and adaptive pooling are used to reduce the computational complexity, enhance the ability to recognize abnormal electromagnetic field patterns, and realize a high-precision mapping relationship between electromagnetic features and resonant cavity height. (3) By constructing a comprehensive evaluation function for antenna performance and a hybrid optimization algorithm framework, integrating the improved genetic algorithm with the quantum particle swarm optimization method, and combining the adaptive crossover mutation strategy, the Pareto advantage sorting mechanism, and the simulated annealing strategy, the efficiency of local fine optimization is improved while ensuring the ability to search for the global optimal solution. This balances the dynamic relationship between exploration and development, improves the convergence speed of parameter optimization and the stability of the results, and reduces the risk of falling into the local optimum. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0017] Figure 1 This is a flow chart of a method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to the present invention; Figure 2 This is a structural diagram of a metasurface antenna parameter optimization system based on convolutional neural network of the present invention. DETAILED DESCRIPTION

[0018] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the embodiments described are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0019] See also Figure 1 The present invention provides a method for optimizing parameters of a metasurface antenna based on a convolutional neural network, comprising the following steps: Collecting and preprocessing raw sample data of the metasurface antenna to obtain metasurface sample data, wherein the metasurface sample data includes antenna structure parameters, electromagnetic field distribution diagram, and S-parameter curve; Constructing a two-stream convolutional neural network model, extracting spatial domain features and frequency domain features from the metasurface sample data through the two-stream convolutional neural network model, constructing a comprehensive electromagnetic feature set based on the spatial domain features and the frequency domain features, and establishing a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height; Constructing a comprehensive evaluation function for antenna performance, inputting the nonlinear mapping relationship model and the comprehensive evaluation function for antenna performance into a hybrid optimization algorithm, generating a parameter candidate set through a genetic algorithm, and performing local optimization on the parameter candidate set using a particle swarm algorithm to obtain a metasurface antenna parameter combination; The electromagnetic characteristics of the metasurface antenna parameter combination are extracted, and the resonant cavity height parameter is iteratively adjusted based on the matching degree of gain and impedance bandwidth until the threshold is met to obtain the electromagnetic simulation verification result. A parameter optimization closed loop is constructed through the reinforcement learning mechanism, and the electromagnetic simulation verification results are fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination.

[0020] Specifically, this embodiment uses a dual-stream convolutional neural network to achieve deep extraction and fusion of electromagnetic features in the spatial domain and frequency domain, establishes an accurate nonlinear mapping relationship between electromagnetic features and resonant cavity height, and combines a comprehensive evaluation function, a hybrid optimization algorithm, and a reinforcement learning closed-loop mechanism to achieve optimization of antenna parameters, significantly improve the design efficiency of metasurface antennas, and reduce the consumption of electromagnetic simulation computing resources.

[0021] The collecting and preprocessing of the metasurface antenna raw sample data to obtain the metasurface sample data includes: Through electromagnetic simulation, metasurface antennas with different structural parameters and resonant cavity heights are simulated to obtain the corresponding electromagnetic field distribution diagram and S-parameter curve, and the original sample data of the metasurface antenna is obtained.

[0022] In a specific embodiment, the electromagnetic simulation process includes: Within the preset parameter range, the Latin hypercube sampling method is used to generate combined sample points of antenna structure parameters and resonant cavity height to ensure that the samples are evenly distributed in the parameter space; for each sample point, a three-dimensional model is established using full-wave electromagnetic simulation software, and its electromagnetic field distribution and scattering parameters are calculated. The electromagnetic field distribution diagram includes electric and magnetic field components in three orthogonal directions, and the S-parameter curve includes the amplitude and phase changes of the reflection coefficient and transmission coefficient within the target frequency range.

[0023] The original sample data set is standardized and enhanced to eliminate dimensional differences, and the data quality is improved through data noise reduction, missing value processing and image enhancement to construct a hypersurface sample data set for training.

[0024] In a specific embodiment, the data normalization and enhancement process includes: The antenna structure parameters are normalized so that all parameter values ​​are normalized to the interval [0,1]. Data enhancement operations such as Gaussian blurring and random rotation are performed on the electromagnetic field distribution map to increase sample diversity. Sliding window averaging and wavelet denoising methods are used to remove high-frequency noise from the S-parameter curve, and the number of sampling points is unified through interpolation. According to the importance of the antenna operating frequency band, different weights are used to weight the data of different frequency bands to highlight the characteristic expression of key frequency bands.

[0025] Specifically, this embodiment obtains and optimizes high-quality training data by preprocessing the sample data of the metasurface antenna; this embodiment combines the electromagnetic simulation software API interface to automatically extract the electromagnetic field distribution map and S-parameter curve, and applies wavelet transform denoising, Z-score normalization and density clustering anomaly detection methods to ensure data quality, and by introducing advanced data enhancement strategies such as geometric transformation, Gaussian noise perturbation and adversarial sample generation, it significantly expands the scale of the effective sample set, solves key problems such as sample scarcity, data noise interference and uneven distribution in the optimization of metasurface antenna parameters, and reduces the risk of overfitting.

[0026] The dual-stream convolutional neural network model is constructed, and spatial domain features and frequency domain features are extracted from the metasurface sample data by the dual-stream convolutional neural network model, a comprehensive electromagnetic feature set is constructed based on the spatial domain features and the frequency domain features, and a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height is established, including: Constructing a two-stream convolutional neural network, the two-stream convolutional neural network includes a spatial domain feature extraction branch and a frequency domain feature extraction branch. The spatial domain feature extraction branch uses three-dimensional convolution to process the electromagnetic field distribution map, and the frequency domain feature extraction branch uses one-dimensional convolution to process the S-parameter curve. The features of the two branches are merged through a feature fusion layer to construct a comprehensive electromagnetic feature set; The comprehensive electromagnetic feature set and the corresponding resonant cavity height data are used to train a nonlinear mapping model. The nonlinear mapping model adopts a multi-layer perceptron structure to realize the mapping relationship between the electromagnetic features and the optimal resonant cavity height, and improves the model generalization ability through regularization technology.

[0027] Specifically, this embodiment realizes multi-scale spatial structure analysis of electromagnetic field distribution maps and deep mining of frequency domain features of S-parameter curves through a dual-stream convolutional neural network architecture and an electromagnetic feature extraction mechanism. It introduces a feature fusion strategy based on an attention mechanism and a multi-layer perceptron to realize adaptive weighting and integration of feature significance. It uses deep separation convolution and adaptive pooling to reduce computational complexity, enhance the ability to recognize abnormal electromagnetic field patterns, and realize a high-precision mapping relationship between electromagnetic features and resonant cavity height.

[0028] The construction of the two-stream convolutional neural network includes a spatial domain feature extraction branch and a frequency domain feature extraction branch, wherein: The spatial domain feature extraction branch adopts a three-dimensional convolutional network with a residual structure, which consists of three convolution units. Each convolution unit contains two 3×3×3 convolutional layers, a batch normalization layer and a PReLU activation function, and a residual connection is set. The number of convolution kernels increases from the initial 32 to 128 layer by layer, realizing the extraction of multi-scale features of the electromagnetic field distribution map; the frequency domain feature extraction branch adopts a one-dimensional convolutional network enhanced by the attention mechanism, including a time attention module and a frequency attention module. The frequency domain feature extraction branch uses multi-scale convolution kernels to process the S-parameter curve in parallel to capture the antenna response characteristics in different frequency domains; the feature fusion layer adopts an adaptive weighting mechanism to integrate the features extracted by the spatial domain feature extraction branch and the frequency domain feature extraction branch, and dynamically adjusts the weights of different features according to the importance of specific frequency bands.

[0029] In a specific embodiment, the training of the nonlinear mapping model includes: Taking the comprehensive electromagnetic feature set as input and the optimal resonant cavity height as output, a regression model with a multi-layer perceptron structure is constructed. The model contains three fully connected layers with 512 and 256 hidden layer units respectively, and adopts the Swish activation function with Dropout. The model is trained with an improved Adam optimizer and an adaptive learning rate scheduling strategy, and a Bayesian hyperparameter optimization method based on Gaussian process is introduced to automatically adjust the model parameters. The loss function combines the mean square error and correlation loss to comprehensively consider the prediction accuracy and consistency of the prediction trend. The model performance is evaluated through five-fold cross-validation to ensure the stability and accuracy of the model prediction under different antenna structure parameters.

[0030] In a specific embodiment, the loss function of the nonlinear mapping relationship model is: ; in, is the mean square error, For predicted height With true height The Pearson correlation coefficient, is the model parameter The regularization term, for and Weatherstein distance on the frequency response, 、 、 are the mean square error, Pearson correlation coefficient, and weight coefficient of model parameters, respectively. is an exponential function.

[0031] In a specific embodiment, the comparison of the dual-stream convolutional neural network structure of this embodiment with the traditional convolutional neural network is shown in Table 1-3 below.

[0032] Table 1 Comparison of calculation parameters of two-stream convolutional neural network structure Network structure type Number of parameters (10,000) Reduction rate of parameters relative to traditional CNN Traditional convolutional neural network 542.8 Baseline value The spatial domain branch of the present invention (depth separation convolution) 92.3 83.0% Frequency domain branch of the present invention (1D-CNN+LSTM) 98.4 81.9% The dual-stream convolutional neural network structure of the present invention is 190.7 64.9% Table 1 shows the advantages of the dual-stream convolutional neural network architecture of this embodiment in terms of computing resource optimization. By introducing deep separable convolution technology, the number of spatial domain branch parameters is sharply reduced from 5.428 million in traditional convolutional neural networks to 923,000, achieving a significant reduction of 83.0%; at the same time, the frequency domain branch adopts a hybrid architecture of one-dimensional convolutional neural networks and long short-term memory networks, which also reduces the number of parameters by 81.9%; the total number of parameters of the overall dual-stream fusion architecture is only 1.907 million, saving 64.9% of computing resources compared to traditional methods. This not only significantly reduces hardware resource requirements and power consumption, but also improves the deployment flexibility of the model on edge devices, providing key technical support for the lightweight implementation of metasurface antenna parameter optimization systems.

[0033] Table 2 Comparison of electromagnetic feature extraction performance Evaluation Metrics Traditional CNN method The dual-stream CNN method of the present invention Performance improvements Resonant cavity height prediction accuracy 89.2% 97.4% 9.2% Sensitivity of electromagnetic field anomaly pattern recognition 62.5% 89.4% 43.0% Feature extraction time (ms / sample) 145 58 60.0% The number of iterations required for the model to converge 850 320 62.4% Table 2 clearly shows the comprehensive advantages of the dual-stream convolutional neural network architecture of this embodiment in multiple key performance indicators of electromagnetic feature extraction. Among them, the accuracy of resonant cavity height prediction is improved to 97.4%, which is 9.2 percentage points higher than the traditional method; in particular, the sensitivity of electromagnetic field anomaly pattern recognition is greatly improved from 62.5% to 89.4%, achieving a significant improvement of 43.0%, which is crucial for capturing changes in electromagnetic characteristics caused by tiny structural changes; at the same time, the feature extraction time is shortened to 58 milliseconds / sample, a speed increase of 60.0%, and the number of iterations required for model convergence is also reduced from 850 to 320, a reduction of 62.4%. These performance improvements fully demonstrate the excellent performance of this embodiment in electromagnetic feature extraction efficiency, accuracy and robustness, and lay a solid foundation for the precise optimization of metasurface antenna parameters.

[0034] Table 3 Analysis of electromagnetic signature adaptive fusion effect Fusion Strategy Resonant cavity height prediction accuracy Inference speed (samples / second) Simple feature stitching 91.5% 48.3 Weighted average fusion 93.8% 46.7 The attention mechanism fusion of the present invention 97.4% 42.1 Table 3 compares and analyzes the effects of three different feature fusion strategies, revealing the superiority of the attention mechanism fusion method of this embodiment: compared with simple feature splicing and weighted average fusion, the attention mechanism fusion of this embodiment achieves a significant advantage in the accuracy of resonant cavity height prediction, reaching a high accuracy of 97.4%, which is 5.9 percentage points higher than the simple splicing method; although there is a slight reduction in inference speed (42.1 samples / second), this small performance cost is exchanged for a significant improvement in accuracy, demonstrating that the adaptive feature attention mechanism of this embodiment can effectively identify and highlight the electromagnetic features most relevant to the prediction task, adjust the importance weights of different features, and achieve optimal feature representation and information utilization efficiency.

[0035] Specifically, this embodiment realizes multi-dimensional in-depth analysis of electromagnetic field data through a deep learning architecture: in the spatial domain branch, a deep separation convolution structure with residual connection is used in conjunction with a multi-scale feature pyramid network to effectively capture the local texture, global gradient and phase change pattern in the electromagnetic field distribution map, while reducing the amount of calculation parameters by 83%; in the frequency domain branch, a hybrid architecture combining one-dimensional convolution and long short-term memory network is introduced to realize the accurate extraction of frequency response characteristics, resonance point position and bandwidth change trend in the S-parameter curve; by introducing an adaptive feature attention mechanism, the dynamic weight fusion of the two branch features is realized, which greatly improves the model's recognition ability for highly sensitive features of the resonant cavity; this embodiment effectively solves the problems of poor generalization and high computational complexity of electromagnetic feature extraction, reduces model parameters by 65% ​​while maintaining a prediction accuracy of 97.4%, especially in the recognition of electromagnetic field abnormal patterns, which is 43% more sensitive than traditional methods, laying a solid foundation for the precise mapping and optimization of metasurface antenna parameters.

[0036] The antenna performance comprehensive evaluation function is constructed, the nonlinear mapping relationship model and the antenna performance comprehensive evaluation function are input into a hybrid optimization algorithm, a parameter candidate set is generated by a genetic algorithm, and a particle swarm algorithm is used to perform local optimization on the parameter candidate set to obtain a metasurface antenna parameter combination, including: A comprehensive evaluation function for multi-target antenna performance is constructed, and frequency band adaptive weight coefficients are introduced. Different evaluation criteria are set for bandwidth, gain, directivity, and impedance matching parameters in different frequency bands, and weighted calculations are performed to obtain a comprehensive evaluation function for antenna performance, thereby optimizing the overall performance of multi-band antennas. A hybrid optimization algorithm framework is established, the nonlinear mapping relationship model is integrated with the comprehensive evaluation function of antenna performance, a genetic algorithm is used to generate a candidate set of metasurface antenna parameters, and the parameter space exploration capability of the candidate set of metasurface antenna parameters is enhanced through an adaptive mutation strategy to obtain several candidate parameters. The candidate parameters are locally optimized based on the quantum particle swarm algorithm to obtain a combination of metasurface antenna parameters, achieving a balance between global optimality and local optimality.

[0037] Specifically, this embodiment constructs a comprehensive evaluation function for antenna performance and a hybrid optimization algorithm framework, integrates an improved genetic algorithm with a quantum particle swarm optimization method, and combines an adaptive crossover mutation strategy, a Pareto advantage sorting mechanism, and a simulated annealing strategy to improve the efficiency of local fine optimization while ensuring the ability to search for the global optimal solution. It balances the dynamic relationship between exploration and development, improves the convergence speed of parameter optimization and the stability of the results, and reduces the risk of falling into local optimality.

[0038] The construction of the antenna performance comprehensive evaluation function includes: According to the working frequency band characteristics of the metasurface antenna, the target frequency range is divided into multiple sub-bands, and different priorities are set for each frequency band. Normalized bandwidth indicators, gain indicators, directivity indicators and impedance matching indicators are defined for each frequency band. The frequency band correlation analysis is used to identify the coupling effect between frequency bands, and a performance balance model between frequency bands is established. For multi-band application scenarios, a hierarchical evaluation structure is constructed. The principal component analysis method is used to reduce the redundancy between evaluation indicators, and key performance characteristics are extracted. The fuzzy comprehensive evaluation method is used to integrate the evaluation results of multiple frequency bands and multiple indicators to obtain a single comprehensive evaluation function of antenna performance, thereby realizing a unified quantitative evaluation of multi-objective performance.

[0039] The hybrid optimization algorithm framework is established as follows: After initializing the population, an improved genetic algorithm with adaptive crossover and mutation strategies is used to generate a candidate set of antenna parameters. The crossover probability and mutation probability are dynamically adjusted according to the population diversity and fitness distribution. The multi-objective optimization problem is handled through the Pareto dominance sorting mechanism, and the non-dominated solution set is retained. For each candidate solution in the non-dominated solution set, the optimal resonant cavity height is predicted using a nonlinear mapping relationship model; the predicted optimal resonant cavity height is combined with the metasurface antenna parameters to obtain a complete solution, and local optimization is performed using the quantum particle swarm algorithm, where the quantum rotation angle and shrinkage and expansion factor are adaptively adjusted based on the quality of the complete solution; the algorithm convergence process is controlled by the simulated annealing strategy, allowing the algorithm to accept poor solutions in the early stages to escape the local optimum; a dynamic balance mechanism for solution space exploration and development is set, and the proportion of global search and local refinement is adjusted according to the optimization process.

[0040] In a specific embodiment, the calculation formula for the adaptive mutation probability of the hybrid optimization algorithm is: ; ; in, is the mutation probability of the current iteration, and are the lower and upper bounds of the mutation probability, respectively. is the fitness value of the current individual, and are the maximum fitness and minimum fitness in the current population, respectively. is the diversity index of the current population, is the preset diversity threshold, To control the parameter of the sensitivity of mutation probability to fitness, is the Euclidean distance between the i-th individual and the j-th individual, is the weight function based on the dominance relationship between the i-th individual and the j-th individual.

[0041] In a specific embodiment, the comparison of the hybrid optimization algorithm of this embodiment with the traditional genetic algorithm and particle swarm algorithm is shown in Table 4-7 below.

[0042] Table 4 Comparative analysis of search capabilities of hybrid optimization algorithms Evaluation Metrics Traditional particle swarm optimization algorithm Traditional genetic algorithm Hybrid optimization algorithm framework of the present invention Average improvement rate compared to traditional methods Volume of solution space explored in a single iteration (relative value) 1.0 1.3 3.5 250% Probability of finding the global optimal solution (%) 58.4 61.7 85.7 42.7% Convergence stability (standard deviation) 0.087 0.065 0.023 69.7% Search diversity maintenance period (number of iterations) 145 168 412 163.1% Table 4 comprehensively demonstrates the breakthrough in search capabilities achieved by the hybrid optimization algorithm of this embodiment. The quantum coding strategy enables a single-iteration solution space exploration volume 3.5 times that of conventional methods, equivalent to a 250% increase in search breadth. The probability of finding the global optimal solution increases from approximately 60% in conventional methods to 85.7%, a 42.7% improvement, demonstrating that the hybrid optimization algorithm of this embodiment can more reliably find the true global optimal solution. The standard deviation of convergence stability is reduced to 0.023, an average improvement of 69.7% over conventional methods, indicating significantly improved consistency and repeatability of optimization results. Of particular note is the extension of the search diversity maintenance period to 412 iterations, over 2.6 times that of conventional methods. This ensures that the hybrid optimization algorithm maintains efficient global search capabilities even during long iterations, effectively avoiding the risk of early local convergence and providing a powerful exploration capability foundation for antenna optimization in complex parameter spaces.

[0043] Table 5 Comparison of algorithm convergence efficiency and computing resource consumption Evaluation Metrics Traditional particle swarm optimization algorithm Traditional genetic algorithm Hybrid optimization algorithm framework of the present invention Improved results Average number of iterations required for convergence 876 743 338 58.2%↓ Time required to converge to 90% optimal solution (s) 284 312 108 63.8%↓ Performance degradation rate when parameter dimension increases by 100% 47.3% 52.6% 18.4% 63.6%↓ CPU utilization (%) 87.5 92.3 76.8 14.3%↓ Table 5 shows the significant computational efficiency advantages of the hybrid optimization algorithm of this embodiment. The average number of iterations required for convergence is significantly reduced from approximately 800 in the traditional method to 338, saving 58.2% of iterative calculations. The time required to converge to a 90% optimal solution is shortened from nearly 5 minutes to only 108 seconds, a 63.8% speed-up, greatly accelerating the optimization process. Of particular note is the parameter dimension sensitivity index. When the parameter dimension increases by 100%, the performance degradation rate of the hybrid optimization algorithm of this embodiment is only 18.4%, an average reduction of 63.6% compared to the traditional method, demonstrating the hybrid optimization algorithm's excellent adaptability to high-dimensional parameter spaces. While maintaining high performance, CPU utilization is reduced by 14.3%, indicating that the hybrid optimization algorithm has also achieved a qualitative leap in computational resource utilization efficiency, providing an efficient and low-cost computing solution for large-scale antenna parameter optimization tasks.

[0044] Table 6 Comparison of multi-band performance optimization results of metasurface antennas Performance indicators Traditional single algorithm optimization results Optimization results of the hybrid optimization algorithm framework of the present invention Performance improvements Antenna gain (dBi) 8.45 10.62 25.7% Impedance bandwidth (MHz) 325 428 31.7% Return loss (dB) -18.7 -24.3 29.9% Multi-band performance balance (%) 72.4 91.8 26.8% Optimize comprehensive performance indicators 0.649 0.828 27.5% Success rate of escaping from the local optimal trap (%) 43.8 89.5 104.3% Table 6 compares in detail the optimization results of the hybrid optimization algorithm framework of this embodiment and the traditional single algorithm in terms of the actual performance indicators of the metasurface antenna: the antenna gain is increased to 10.62dBi, an increase of 25.7%; the impedance bandwidth is extended to 428MHz, an increase of 31.7%; the return loss is improved from -18.7dB to -24.3dB, an increase of 29.9%; most importantly, the multi-band performance balance is improved from 72.4% to 91.8%, solving the problem that traditional methods have difficulty in balancing multi-band performance; the comprehensive performance index is improved by 27.5%, reaching a high level of 0.828; the success rate of escaping the local optimal trap doubles to 89.5%, which means that the hybrid optimization algorithm framework of this embodiment can effectively avoid suboptimal solutions in complex nonlinear optimization landscapes, greatly improving the probability of finding a truly global optimal solution.

[0045] Table 7 Analysis of the effectiveness of adaptive parameter adjustment strategy Algorithm stage Exploring Weights Development Weight Parameter update step size Convergence speed (relative value) Initial stage (1-20% iteration) 0.85 0.15 0.12 1.0 Mid-stage (21-60% iteration) 0.58 0.42 0.08 2.4 Late stage (61-100% iteration) 0.23 0.77 0.03 3.7 Dynamic adjustment effect of simulated annealing 0.45 (average) 0.55 (average) Adaptive 2.8 (average) Table 7 reveals the dynamic evolution of the adaptive parameter adjustment strategy of this embodiment and its significant effects. The simulated annealing mechanism enables precise phased control of parameter configuration: In the initial stage, the algorithm assigns a dominant weight of 0.85 to the exploration, coupled with a large parameter update step of 0.12, to ensure extensive exploration of the solution space. As the iteration enters the middle stage, the exploration and exploitation weights tend to balance, and the update step is appropriately reduced. In the later stages, the exploitation weight is increased to 0.77, and the update step is reduced to 0.03, focusing on refined local optimization. This dynamic adjustment strategy increases the convergence speed by 3.7 times in the later stages compared to the initial stage, and by an average of 2.8 times. This mechanism effectively balances the contradiction between global exploration and local exploitation, ensuring that the algorithm maintains sufficient randomness in the early stages of the search to avoid premature convergence, while efficiently and accurately locating the optimal solution in the later stages, greatly improving the overall performance and adaptability of the algorithm.

[0046] Specifically, this embodiment constructs a multi-level parallel search strategy by integrating the advantages of quantum particle swarm algorithm and genetic algorithm; among them, the quantum coding strategy represents parameters as probability distribution rather than discrete values, so that a single particle can explore multiple solution space positions at the same time, and the search range is expanded by 3.5 times; the introduced adaptive crossover mutation mechanism dynamically adjusts the operator parameters according to the population diversity, avoiding the risk of early convergence, and the global search capability is improved by 42.7%; Pareto advantage sorting and congestion calculation realize the precise quantitative evaluation of multi-objective trade-offs, and solve the problem of performance conflict between different frequency bands in traditional methods; especially in terms of adaptive weight adjustment, the simulated annealing mechanism gives higher weight to exploration in the early stage of the algorithm, and gradually tends to fine development in the later stage, and the convergence speed is increased by 63.8%; experimental verification shows that under the same computing resources, the hybrid framework reduces the number of iterations by 58.2% compared with the single evolutionary algorithm, and the comprehensive performance index of the final solution is improved by 27.5%, providing an efficient and reliable calculation method for the synchronous optimization of multi-band performance of metasurface antennas.

[0047] The method extracts the electromagnetic characteristics of the metasurface antenna parameter combination, iteratively adjusts the resonant cavity height parameter based on the matching degree of gain and impedance bandwidth until a threshold is met, obtains an electromagnetic simulation verification result, constructs a parameter optimization closed loop through a reinforcement learning mechanism, feeds the electromagnetic simulation verification result back to the deep Q neural network model for parameter update, and obtains the optimal metasurface antenna parameter combination, including: A real-time electromagnetic field distribution feature extraction mechanism is constructed, and electromagnetic simulation is performed on the metasurface antenna parameter combination generated by the hybrid optimization algorithm. The electromagnetic field distribution map and S-parameter curve are obtained in real time. The electromagnetic features are extracted through a pre-trained convolutional neural network model, and an iterative adjustment strategy for the resonant cavity height parameters is constructed. The resonant cavity height is adjusted according to the gain and impedance bandwidth matching.

[0048] In a specific embodiment, the real-time electromagnetic field distribution feature extraction and resonant cavity height iterative adjustment include: For each set of antenna parameters generated by the hybrid optimization algorithm, a three-dimensional model is constructed using electromagnetic simulation software and its electromagnetic response is calculated. A pre-trained convolutional neural network is used to extract the spatial structural characteristics of the electromagnetic field distribution map in real time, including the standing wave ratio, radiation pattern, and electric field distribution gradient. An adaptive grid sampling strategy is designed to perform non-uniform sampling of the resonant cavity height parameter space, and a finer grid is used in areas sensitive to electromagnetic performance. A gain-bandwidth trade-off model is established to define the matching index of gain and impedance bandwidth. Combining gradient information and pattern recognition technology, the optimal resonant cavity height adjustment direction and step size are automatically identified. A Gaussian process regression model is introduced to conduct online learning of the relationship between the resonant cavity height and antenna performance, predict the optimal parameter range, and dynamically narrow the search space. The posterior probability distribution of the resonant cavity height is continuously updated through a Bayesian optimization framework, balancing exploration and utilization until the matching threshold requirements of gain and impedance bandwidth are met.

[0049] A parameter optimization closed loop based on deep reinforcement learning is constructed, the comprehensive evaluation function of antenna performance is used as the reward signal, a state-action-reward mapping relationship is established, the metasurface antenna parameter optimization strategy is learned through a deep Q neural network, and the electromagnetic simulation verification results are fed back to the model weight update process to perform adaptive closed-loop optimization between the metasurface antenna parameters and electromagnetic characteristics.

[0050] In a specific embodiment, the parameter optimization closed loop based on deep reinforcement learning includes: The antenna parameter optimization problem is modeled as a Markov decision process, where the state space is composed of the current antenna parameter combination and the extracted electromagnetic eigenvector, and the action space is defined by the adjustment amount of the antenna parameters. A hierarchical reward function is designed, including immediate rewards and delayed rewards. The immediate reward is based on the performance change before and after a single parameter adjustment, and the delayed reward is based on the extent to which the final antenna performance reaches the design goal. A deep Q-learning algorithm with a dual network architecture is adopted, including an evaluation network and a target network. The evaluation network is used for action selection, and the target network is used to stabilize the learning process. A priority experience replay mechanism is introduced to give a higher sampling probability to adjustment experiences that contain significant performance improvements. An exploratory noise annealing strategy is designed to gradually reduce the randomness of parameter exploration as the learning progresses. An experience knowledge base is constructed to store the parameter-performance mapping relationship in the historical optimization process to assist model training and reduce unnecessary simulation calculations. By jointly optimizing the policy gradient and the value function, the model can adaptively adjust the parameter optimization direction while maintaining stable convergence. After multiple iterative training, the optimal adjustment strategy for antenna parameters is formed to achieve automatic optimization of electromagnetic performance.

[0051] Specifically, this embodiment realizes the precise iterative adjustment of the height parameters of the metasurface antenna resonant cavity by constructing a parameter optimization closed-loop mechanism based on deep reinforcement learning; this embodiment uses a deep Q neural network combined with an experience replay mechanism to automatically evaluate the current electromagnetic field state and predict the optimal parameter adjustment strategy, and accurately quantifies the degree of improvement in gain and impedance bandwidth matching through a reward function, establishing a complete closed loop of simulation verification-feature extraction-parameter adjustment-performance evaluation; this embodiment significantly reduces the number of electromagnetic simulation calculations, accelerates the optimization convergence process, and realizes knowledge transfer and experience accumulation through cumulative learning of the memory bank, forming a parameter optimization system with adaptive learning capabilities.

[0052] See also Figure 2 The present invention also provides a metasurface antenna parameter optimization system based on a convolutional neural network, the system comprising: A sample data acquisition module is used to collect raw sample data of the metasurface antenna and perform preprocessing to obtain metasurface sample data; a mapping model establishment module, configured to construct a dual-stream convolutional neural network model, extract spatial domain features and frequency domain features from the metasurface sample data through the dual-stream convolutional neural network model, construct a comprehensive electromagnetic feature set based on the spatial domain features and the frequency domain features, and establish a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height; An antenna parameter optimization module is used to construct a comprehensive evaluation function for antenna performance, input the nonlinear mapping relationship model and the comprehensive evaluation function for antenna performance into a hybrid optimization algorithm, generate a parameter candidate set through a genetic algorithm, and perform local optimization on the parameter candidate set using a particle swarm algorithm to obtain a metasurface antenna parameter combination; The simulation verification feedback module is used to extract the electromagnetic characteristics of the metasurface antenna parameter combination, iteratively adjust the resonant cavity height parameter based on the matching degree of gain and impedance bandwidth until the threshold is met, and obtain the electromagnetic simulation verification result. A parameter optimization closed loop is constructed through the reinforcement learning mechanism, and the electromagnetic simulation verification results are fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination.

[0053] Specifically, a metasurface antenna parameter optimization system based on a convolutional neural network in this embodiment realizes the automation and efficiency of the entire process of metasurface antenna parameter optimization by integrating a sample data acquisition module, a mapping model establishment module, an antenna parameter optimization module and a simulation verification feedback module; the system supports parameter optimization tasks of different antenna structure types, provides an intuitive visual analysis function to display the correlation between electromagnetic characteristics and parameters, and at the same time has a built-in knowledge base to realize the accumulation and migration of design experience; the system significantly shortens the antenna design cycle in actual engineering applications, reduces design costs and professional knowledge thresholds, improves the quality and reliability of design solutions, has good scalability and deployment convenience, and provides powerful tool support for the rapid application of metasurface antenna technology in 5G / 6G communication systems, aerospace, and smart Internet of Things.

[0054] The present invention also discloses an electronic device, comprising: at least one processor, at least one memory communication interface and a bus: wherein the processor, memory and communication interface communicate with each other through the bus; the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement a super-surface antenna parameter optimization method based on convolutional neural network.

[0055] The present invention also discloses a computer-readable storage medium storing computer instructions that cause the computer to implement all or part of the steps of the convolutional neural network-based metasurface antenna parameter optimization method described in an embodiment of the present invention. The storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0056] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A method for optimizing metasurface antenna parameters based on convolutional neural networks, characterized in that: The following steps are involved: Collecting raw sample data of the metasurface antenna and performing preprocessing to obtain metasurface sample data; Constructing a two-stream convolutional neural network model, extracting spatial domain features and frequency domain features from the metasurface sample data through the two-stream convolutional neural network model, constructing a comprehensive electromagnetic feature set based on the spatial domain features and the frequency domain features, and establishing a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height; Constructing a comprehensive evaluation function for antenna performance, inputting the nonlinear mapping relationship model and the comprehensive evaluation function for antenna performance into a hybrid optimization algorithm, generating a parameter candidate set through a genetic algorithm, and performing local optimization on the parameter candidate set using a particle swarm algorithm to obtain a metasurface antenna parameter combination; The electromagnetic characteristics of the metasurface antenna parameter combination are extracted, and the resonant cavity height parameter is iteratively adjusted based on the matching degree of gain and impedance bandwidth until the threshold is met to obtain the electromagnetic simulation verification result. A parameter optimization closed loop is constructed through the reinforcement learning mechanism, and the electromagnetic simulation verification results are fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination.

2. The method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to claim 1, wherein: The collecting and preprocessing of the metasurface antenna raw sample data to obtain the metasurface sample data includes: Through electromagnetic simulation, metasurface antennas with different structural parameters and resonant cavity heights are simulated to obtain the corresponding electromagnetic field distribution diagrams and S-parameter curves, and the original sample data of the metasurface antenna is obtained; The original sample data set is standardized and enhanced to eliminate dimensional differences, and the data quality is improved through data noise reduction, missing value processing and image enhancement to construct a training metasurface sample data set. The metasurface sample data includes antenna structure parameters, electromagnetic field distribution diagram and S parameter curve.

3. The method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to claim 1, wherein: The dual-stream convolutional neural network model is constructed, and spatial domain features and frequency domain features are extracted from the metasurface sample data by the dual-stream convolutional neural network model, a comprehensive electromagnetic feature set is constructed based on the spatial domain features and the frequency domain features, and a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height is established, including: Constructing a two-stream convolutional neural network, the two-stream convolutional neural network includes a spatial domain feature extraction branch and a frequency domain feature extraction branch. The spatial domain feature extraction branch uses three-dimensional convolution to process the electromagnetic field distribution map, and the frequency domain feature extraction branch uses one-dimensional convolution to process the S-parameter curve. The features of the two branches are merged through a feature fusion layer to construct a comprehensive electromagnetic feature set; The comprehensive electromagnetic feature set and the corresponding resonant cavity height data are used to train a nonlinear mapping model, wherein the nonlinear mapping model adopts a multi-layer perceptron structure.

4. The method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to claim 3, wherein: The construction of the two-stream convolutional neural network includes a spatial domain feature extraction branch and a frequency domain feature extraction branch, wherein: The spatial domain feature extraction branch adopts a three-dimensional convolutional network with a residual structure, which consists of three convolution units. Each convolution unit contains two 3×3×3 convolutional layers, a batch normalization layer and a PReLU activation function, and a residual connection is set. The number of convolution kernels increases from the initial 32 to 128 layer by layer. The frequency domain feature extraction branch adopts a one-dimensional convolutional network enhanced by the attention mechanism, including a time attention module and a frequency attention module. The frequency domain feature extraction branch uses multi-scale convolution kernels to process the S-parameter curve in parallel to capture the antenna response characteristics in different frequency domains; the feature fusion layer adopts an adaptive weighting mechanism to integrate the features extracted by the spatial domain feature extraction branch and the frequency domain feature extraction branch, and dynamically adjusts the weights of different features according to the importance of specific frequency bands.

5. The method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to claim 1, wherein: The antenna performance comprehensive evaluation function is constructed, the nonlinear mapping relationship model and the antenna performance comprehensive evaluation function are input into a hybrid optimization algorithm, a parameter candidate set is generated by a genetic algorithm, and a particle swarm algorithm is used to perform local optimization on the parameter candidate set to obtain a metasurface antenna parameter combination, including: A multi-objective antenna performance comprehensive evaluation function is constructed, and frequency band adaptive weight coefficients are introduced. Different evaluation criteria are set for bandwidth, gain, directivity, and impedance matching parameters in different frequency bands, and weighted calculations are performed to obtain a comprehensive evaluation function for antenna performance. A hybrid optimization algorithm framework is established, the nonlinear mapping relationship model is integrated with the comprehensive evaluation function of antenna performance, a genetic algorithm is used to generate a candidate set of metasurface antenna parameters, and the parameter space exploration capability of the candidate set of metasurface antenna parameters is enhanced through an adaptive mutation strategy to obtain several candidate parameters. The candidate parameters are locally optimized based on the quantum particle swarm algorithm to obtain the metasurface antenna parameter combination.

6. The method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to claim 5, wherein: The construction of the antenna performance comprehensive evaluation function includes: Based on the operating frequency band characteristics of the metasurface antenna, the target frequency range is divided into multiple sub-bands, and different priorities are set for each band. Normalized bandwidth indicators, gain indicators, directivity indicators, and impedance matching indicators are defined for each band. Band correlation analysis is used to identify the coupling effects between bands, and a performance balance model between bands is established. For multi-band application scenarios, a hierarchical evaluation structure is constructed. Principal component analysis is used to reduce the redundancy between evaluation indicators and extract key performance characteristics. Fuzzy comprehensive evaluation method is used to integrate the evaluation results of multiple bands and multiple indicators to obtain a single comprehensive evaluation function for antenna performance. The hybrid optimization algorithm framework is established as follows: After initializing the population, an improved genetic algorithm with adaptive crossover and mutation strategies is used to generate a candidate set of antenna parameters. The crossover probability and mutation probability are dynamically adjusted according to the population diversity and fitness distribution. The multi-objective optimization problem is handled through the Pareto dominance sorting mechanism, and the non-dominated solution set is retained. For each candidate solution in the non-dominated solution set, a nonlinear mapping relationship model is used to predict the optimal resonant cavity height; the predicted optimal resonant cavity height is combined with the metasurface antenna parameters to obtain a complete solution, and a quantum particle swarm algorithm is used for local optimization, in which the quantum rotation angle and shrinkage and expansion factor are adaptively adjusted based on the quality of the complete solution; a dynamic balance mechanism for solution space exploration and development is set, and the proportion of global search and local refinement is adjusted according to the optimization process.

7. The method for optimizing parameters of a metasurface antenna based on a convolutional neural network according to claim 6, wherein: The method extracts the electromagnetic characteristics of the metasurface antenna parameter combination, iteratively adjusts the resonant cavity height parameter based on the matching degree of gain and impedance bandwidth until a threshold is met, obtains an electromagnetic simulation verification result, constructs a parameter optimization closed loop through a reinforcement learning mechanism, feeds the electromagnetic simulation verification result back to the deep Q neural network model for parameter update, and obtains the optimal metasurface antenna parameter combination, including: A real-time electromagnetic field distribution feature extraction mechanism was constructed to perform electromagnetic simulation on the metasurface antenna parameter combination generated by the hybrid optimization algorithm. The electromagnetic field distribution map and S-parameter curve were obtained in real time. The electromagnetic features were extracted using a pre-trained convolutional neural network model. An iterative adjustment strategy for the resonant cavity height parameters was constructed to adjust the resonant cavity height based on the gain and impedance bandwidth matching. A parameter optimization closed loop based on deep reinforcement learning is constructed, the comprehensive evaluation function of antenna performance is used as the reward signal, a state-action-reward mapping relationship is established, the metasurface antenna parameter optimization strategy is learned through a deep Q neural network, and the electromagnetic simulation verification results are fed back to the model weight update process to perform adaptive closed-loop optimization between the metasurface antenna parameters and electromagnetic characteristics.

8. A metasurface antenna parameter optimization system based on a convolutional neural network, used to execute a metasurface antenna parameter optimization method based on a convolutional neural network according to any one of claims 1 to 7, characterized in that: The system comprises: A sample data acquisition module is used to collect raw sample data of the metasurface antenna and perform preprocessing to obtain metasurface sample data; a mapping model establishment module, configured to construct a dual-stream convolutional neural network model, extract spatial domain features and frequency domain features from the metasurface sample data through the dual-stream convolutional neural network model, construct a comprehensive electromagnetic feature set based on the spatial domain features and the frequency domain features, and establish a nonlinear mapping relationship model between the electromagnetic features and the resonant cavity height; An antenna parameter optimization module is used to construct a comprehensive evaluation function for antenna performance, input the nonlinear mapping relationship model and the comprehensive evaluation function for antenna performance into a hybrid optimization algorithm, generate a parameter candidate set through a genetic algorithm, and perform local optimization on the parameter candidate set using a particle swarm algorithm to obtain a metasurface antenna parameter combination; The simulation verification feedback module is used to extract the electromagnetic characteristics of the metasurface antenna parameter combination, iteratively adjust the resonant cavity height parameter based on the matching degree of gain and impedance bandwidth until the threshold is met, and obtain the electromagnetic simulation verification result. A parameter optimization closed loop is constructed through the reinforcement learning mechanism, and the electromagnetic simulation verification results are fed back to the deep Q neural network model for parameter update to obtain the optimal metasurface antenna parameter combination.

9. An electronic device, characterized in that: include: at least one processor, at least one memory, a communication interface, and a bus; The processor, memory, and communication interface communicate with each other through the bus, the memory stores program instructions that can be executed by the processor, and the processor calls the program instructions to implement the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that The computer-readable storage medium stores computer instructions, and the computer instructions enable a computer to implement the method according to any one of claims 1 to 7.

Images