Design method and system of a dual-state transmissive polarization conversion metasurface

Abstract

This invention discloses a design method for a dual-state transmission-type polarization conversion metasurface. The method includes: establishing a parameterized model of the metasurface unit; sampling and acquiring multiple sets of structural parameters; simulating the electromagnetic responses of each structural parameter in the first and second operating states; extracting dual-state performance indicators to construct a standard sample set; training a surrogate model and constructing a unified objective function; using a differential evolution algorithm to perform a global search on the surrogate model, selecting feedback candidate points from the search history for simulation feedback, and updating the sample set; after retraining the surrogate model, using the globally optimal solution obtained by differential evolution as the center, employing a covariance matrix adaptive evolution strategy to perform local optimization to obtain locally optimal structural parameters; and performing simulation verification on the locally optimal structural parameters to determine whether they meet the design requirements. The design method of this invention has the advantages of balancing dual-state performance indicators while improving design efficiency and result reliability.

Description

Technical Field

[0001] This invention belongs to the field of metasurface design technology, specifically relating to a design method and system for a dual-state transmission polarization conversion metasurface. Background Technology

[0002] Sea surface temperature, sea surface wind field, and polar ice and snow parameters are important physical quantities in marine environmental monitoring, global climate change research, and water cycle analysis. Microwave radiometers retrieve relevant parameters by receiving microwave signals emitted by a target's natural radiation. The C-band, especially around 6.9 GHz, is highly sensitive to changes in sea surface temperature and wind field, making it valuable for applications in spaceborne and airborne microwave remote sensing systems. In recent years, metasurfaces have become an important research direction in microwave device design due to their planarization, miniaturization, and ability to flexibly control the amplitude, phase, and polarization state of electromagnetic waves. Compared to traditional reflective structures, transmissive metasurfaces, while maintaining a thin and lightweight structure, can directly control the transmitted wavefront and polarization state, showing great application potential in microwave radiometer detection systems, polarization modulation devices, and low-profile transmissive functional devices.

[0003] Currently, the design of transmissive polarization-conversion metasurfaces mainly relies on design experience and iterative parameter scanning, resulting in long design cycles and high costs for full-wave electromagnetic simulation. For transmissive metasurfaces in reconfigurable devices, the same set of structural parameters often corresponds to different transmission responses under different operating states. If optimization is only performed on a single state, a single frequency, or a single performance index, it is difficult to simultaneously consider key performance aspects such as cross-polarization transmission amplitude, co-polarization suppression capability, phase difference between states, and polarization conversion rate under two different operating states. On the other hand, traditional methods based on manual scanning or direct full-wave optimization are inefficient in high-dimensional parameter spaces, making it difficult to meet the rapid design requirements for engineering applications.

[0004] Existing reverse engineering methods employ deep neural networks, generative adversarial networks, or simulation combined with surrogate models to achieve rapid metasurface design. However, most of these methods target dielectric metasurfaces, coded metasurfaces, or general spectral response problems, and many schemes emphasize image coding, generative design, or the mapping between single-state target responses and structures. In the field of microwave radiometer applications, where diodes are modeled separately and surrogate models are combined with optimization algorithms to optimize the design of transmission-type polarization conversion metasurfaces, existing methods typically optimize only for a single operating state, failing to address the technical challenge of joint optimization of two states in reconfigurable metasurfaces. For reconfigurable devices that require excellent performance in both operating states and satisfy specific phase difference and amplitude balance requirements between the two states, existing single-state design methods struggle to simultaneously optimize these multiple performance indicators.

[0005] Therefore, developing an inverse design method that can take into account both dual-state performance indicators and improve the design efficiency and reliability of reconfigurable transmissive polarization conversion metasurfaces is a technical problem that urgently needs to be solved in this field. Summary of the Invention

[0006] This invention provides a design method and system for a dual-state transmission polarization conversion metasurface that can take into account both dual-state performance indicators and improve design efficiency and result reliability, in order to solve the problem that existing single-state design methods are difficult to simultaneously optimize multiple performance indicators.

[0007] To achieve the above objectives, this invention provides a design method for a dual-state transmission-type polarization conversion metasurface, comprising the following steps: A parameterized model of a transmission-type polarization conversion metasurface unit is established. Multiple sets of structural parameters are sampled within the range of design variables. The electromagnetic response of the metasurface unit in the first and second operating states under each set of structural parameters is simulated. Based on the electromagnetic response, the dual-state performance index corresponding to each structural parameter is obtained. The dual-state performance index includes the target frequency phase difference error, the average phase difference error in the band, the average cross-polarization transmission amplitude in the first operating state, the average cross-polarization transmission amplitude in the second operating state, the minimum polarization conversion rate in the first operating state, the minimum polarization conversion rate in the second operating state, and the dual-state amplitude imbalance. A standard sample set is constructed based on each structural parameter and its corresponding two-state performance index. A preset surrogate model is trained based on the standard sample set to establish a nonlinear mapping between the structural parameters and the two-state performance index. After training, a unified objective function of the surrogate model is constructed to quickly evaluate the comprehensive performance of any set of structural parameters based on the difference between the two-state performance index predicted by the surrogate model and the preset target performance index. A differential evolution algorithm is used to perform a global search on the surrogate model to obtain a global optimal solution set and search history. Several feedback candidate points are selected from the search history. Each feedback candidate point is simulated and the corresponding dual-state performance index is obtained. Each feedback candidate point and the corresponding dual-state performance index are added as new feedback samples to the standard sample set to obtain a new sample set. Based on the new sample set, the surrogate model is retrained. On the retrained surrogate model, the locally optimal structural parameters are obtained by using the covariance matrix adaptive evolution strategy as the center, with the globally optimal solution obtained by the differential evolution algorithm as the center. Simulations are performed based on the locally optimal structural parameters to obtain the optimal simulated electromagnetic response of the metasurface unit in the first and second operating states. The optimal simulated electromagnetic response is then used to determine whether the polarization conversion and phase modulation requirements within the target frequency band are met.

[0008] In one embodiment, the unified objective function is: , in, A set of structural parameters, For the dimensions of two-state performance metrics, The first two-state performance index predicted by the surrogate model Dimensional performance indicators The first of the preset target performance indicators Dimensional performance indicators For the first The weighting coefficients corresponding to the performance indicators.

[0009] In one embodiment, constructing a standard sample set based on each structural parameter and its corresponding two-state performance index includes: Each set of structural parameters and its corresponding two-state performance index is used as a sample to construct an initial sample set; The structural parameters and two-state performance indices in the initial sample set are normalized according to the following formulas:

[0010] in, These are the structural parameters before normalization. and These are the mean and standard deviation of the structural parameters, respectively. These are the normalized structural parameters; The two-state performance index before normalization. and These are the mean and standard deviation of the two-state performance index, respectively. The normalized two-state performance index; Based on the normalized structural parameters and the corresponding normalized two-state performance index, a standard sample set is obtained.

[0011] In one embodiment, the step of using a differential evolution algorithm to perform a global search on the surrogate model to obtain a globally optimal solution set and search history includes the following steps: S31: Initialize the population, which contains multiple individuals, each individual corresponding to a set of structural parameters; S32: For the current population, for each individual in the population Perform the following operations: S33: Randomly select three distinct individuals from the current population, and then apply the formula... Construct a mutation vector, where, For individuals in the current population The mutation vector, , , For from the first Three distinct individuals randomly selected from the population. This is the difference scaling factor; S34: Transfer the mutation vector With the individual Perform cross operations to generate test vectors ; S35: Transfer the test vector With the individual The individuals with better fitness are compared and retained to enter the next generation of the population, where fitness is the output value of the unified objective function; S36: Iterate through steps S32 to S35 until the termination condition is met, and output the global optimal solution and search history.

[0012] In one embodiment, the step of filtering a plurality of feedback candidate points from the search history includes: Select the candidate point with the best fitness from the search history as the first feedback candidate point; The remaining candidate points are screened in descending order of fitness until the number of selected feedback candidate points reaches the preset value. The step of filtering the remaining candidate points in descending order of fitness includes: Following the order of fitness from best to worst, each time the candidate point with the best fitness is selected from the remaining candidate points as the current candidate point for that screening. For each current candidate point, calculate the minimum normalized distance between the current candidate point and each of the selected feedback candidate points. The calculation formula is as follows: , in, As the current candidate point, The selected candidate points constitute the set of all selected feedback candidate points. A feedback candidate point is one of the selected candidate point sets. and These are the upper and lower bounds for each design variable, respectively. If the minimum normalized distance is greater than a preset threshold, the current candidate point is selected as the next feedback candidate point.

[0013] In one embodiment, the step of using the globally optimal solution obtained by the differential evolution algorithm as the center and employing a covariance matrix adaptive evolution strategy to perform local optimization to obtain locally optimal structure parameters includes the following steps: S41: Obtain multiple globally optimal solutions output by the differential evolution algorithm; S42: For each of the globally optimal solutions, perform the following local optimization operations: S421: Use this globally optimal solution as the initial center point. Initialize step size Covariance Matrix ; S422: For the current generation of search, sample and generate as follows: A new individual: , in, Let be the mean vector of the current generation. The current step size, Given the current covariance matrix, With a mean of 0 and a covariance matrix of The normal distribution; S423: Calculate the fitness of each new individual based on the retrained surrogate model, and select the individual with the best fitness. Each individual is an elite individual, among whom The proportion of elites; S424: Calculate the weighted average of the elite individuals as the updated mean vector. And update the current step size based on the elite individuals. and the current covariance matrix ; S425: Iteratively execute steps S422~S424 until the termination condition is met, and output the local optimization result corresponding to the global better solution; S43: Compare the local optimization results corresponding to each of the globally optimal solutions, and select the one with the best fitness as the final local optimal structure parameter.

[0014] In one embodiment, the loss function of the proxy model is: , in, The data-driven loss term represents the two-state performance metric predicted by the surrogate model. Two-state performance metrics obtained from simulation The mean square error between them The parameters of the proxy model are... is the regularization coefficient.

[0015] Based on the same inventive concept, this invention also proposes a dual-state transmission polarization conversion metasurface obtained using the design method described in any of the preceding claims, comprising: A periodically arranged metasurface unit, wherein the metasurface unit is provided with a top metal gate, an upper dielectric layer, an intermediate functional layer, a lower dielectric layer and a bottom metal gate in sequence from top to bottom; The top metal grid and the bottom metal grid are arranged orthogonally to each other; The upper dielectric layer and the lower dielectric layer are made of dielectric materials with the same dielectric constant; The intermediate functional layer is a metal structure layer, including a functional layer outer frame with the geometric center coincident, a central rectangle, and two open-circuit branches disposed between the functional layer outer frame and the central rectangle. One end of each open-circuit branch is connected to an inner corner of the functional layer outer frame, and the other end is connected to the midpoint of one side of the central rectangle through a diode. A metal cylinder connects the bottom metal grid to the outer frame of the functional layer, and another metal cylinder connects the feed line of the bottom metal grid to the central rectangle. By switching the diode between its on and off states, the metasurface unit switches between a first operating state and a second operating state to perform transmission-type polarization conversion on the incident electromagnetic wave.

[0016] Based on the same inventive concept, this invention also proposes a design system for a dual-state transmission-type polarization conversion metasurface, comprising: The sample construction module is used to establish a parameterized model of the transmission-type polarization conversion metasurface unit. Within the range of design variables, multiple sets of structural parameters are sampled, and the electromagnetic response of the metasurface unit in the first and second operating states under each set of structural parameters is simulated. Based on the electromagnetic response, the dual-state performance index corresponding to each structural parameter is obtained. The dual-state performance index includes the target frequency phase difference error, the average phase difference error in the band, the average cross-polarization transmission amplitude in the first operating state, the average cross-polarization transmission amplitude in the second operating state, the minimum polarization conversion rate in the first operating state, the minimum polarization conversion rate in the second operating state, and the dual-state amplitude imbalance. The surrogate model training module is used to construct a standard sample set based on each structural parameter and its corresponding two-state performance index, train a preset surrogate model based on the standard sample set, establish a nonlinear mapping between the structural parameters and the two-state performance index, and after training, construct a unified objective function for the surrogate model, which is used to quickly evaluate the comprehensive performance of any set of structural parameters based on the difference between the two-state performance index predicted by the surrogate model and the preset target performance index. The global search module is used to perform a global search on the surrogate model using the differential evolution algorithm to obtain a global optimal solution set and search history. It selects several feedback candidate points from the search history, simulates each feedback candidate point and obtains the corresponding dual-state performance index, and adds each feedback candidate point and the corresponding dual-state performance index as new feedback samples to the standard sample set to obtain a new sample set. The local optimization module is used to retrain the surrogate model based on the new sample set. On the retrained surrogate model, the module uses the globally optimal solution obtained by the differential evolution algorithm as the center and adopts the covariance matrix adaptive evolution strategy to perform local optimization to obtain the local optimal structure parameters. The simulation verification module is used to perform simulation based on the local optimal structural parameters to obtain the final simulated electromagnetic response of the metasurface unit in the first and second working states, and to determine whether the polarization conversion and phase modulation requirements in the target frequency band are met based on the final simulated electromagnetic response.

[0017] Based on the same inventive concept, the present invention also proposes an electronic device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor, when executing the program, implements the design method as described in any of the preceding claims.

[0018] Based on the same inventive concept, embodiments of the present invention also propose a computer storage medium storing at least one executable instruction that causes a processor to perform the design method described in any of the preceding claims.

[0019] Compared with the prior art, the beneficial effects of the present invention are: (1) This invention uses dual-state joint evaluation as the design basis, which can simultaneously take into account key indicators such as cross-polarization transmission capability, phase difference, and polarization conversion rate under two working states. Compared with existing design methods that only optimize a single working state, this invention incorporates the performance indicators of the first and second working states into the same optimization framework, and introduces the phase difference and amplitude imbalance between states as evaluation dimensions, thus solving the technical problem that single-state optimization in reconfigurable metasurface design cannot guarantee the synergistic performance of dual states.

[0020] (2) This invention integrates differential evolution algorithm and covariance matrix adaptive evolution strategy. First, differential evolution algorithm is used for a preliminary global search, which can quickly locate potential optimal regions within a large parameter space and provide a high-quality starting point for screening real feedback candidate points. This achieves efficient exploration of a large parameter space while maintaining low algorithm complexity. Then, the covariance matrix adaptive evolution strategy is used for subsequent local optimization. This strategy can adaptively adjust the search step size and covariance structure using search history, further improving optimization accuracy within the updated optimal region. It has the advantages of robustness, speed, and minimal user settings in electromagnetic optimization problems. Compared with using differential evolution algorithm or particle swarm optimization algorithm alone, the covariance matrix adaptive evolution strategy can automatically learn the correlation between design variables, showing significant advantages in handling metasurface parameter optimization problems with strongly correlated variables. Experimental results show that the average score of the method in this invention is improved by approximately 0.05% compared to the differential evolution algorithm alone, and the standard deviation is reduced by approximately 39%. The stability and reliability of the optimization results are significantly better than existing single optimization algorithms.

[0021] (3) This invention corrects and verifies the surrogate model results through real full-wave feedback and final dual-state verification, avoiding reliance solely on the surrogate model to obtain conclusions, thus improving the reliability and engineering usability of the final design results. Compared with existing technical solutions that use the surrogate model after training, this invention actively introduces real simulation feedback and retrains the surrogate model based on it after differential evolution search and before the covariance matrix adaptive evolution strategy fine search, so that subsequent searches are based on a more accurate surrogate model, effectively overcoming the technical risk that the prediction bias of the surrogate model may lead to the failure of the optimization results. Attached Figure Description

[0022] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0023] Figure 1 This is a flowchart illustrating a design method for a dual-state transmission-type polarization conversion metasurface according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the five-layer structure of the transmission-type polarization conversion metasurface unit in an embodiment of the present invention; Figure 3 for Figure 2 A schematic diagram of the parameterized structure of the intermediate functional layer of the transmissive polarization conversion metasurface unit is shown. Figure 4 This is a schematic diagram of the dual-state full-wave electromagnetic simulation process in an embodiment of the present invention; Figure 5 This is a schematic diagram of the differential evolution global search process based on the surrogate model in an embodiment of the present invention; Figure 6 This is a schematic diagram of the proxy model retraining and local optimization process in an embodiment of the present invention; Figure 7 This is a schematic diagram of the dual-state transmission amplitude according to an embodiment of the present invention; wherein, state "0" represents the first working state and state "1" represents the second working state; Figure 8 This is a schematic diagram of the dual-state transmission phase according to an embodiment of the present invention; wherein, state "0" represents the first working state and state "1" represents the second working state; Figure 9 This is a schematic diagram of the polarization conversion rate according to an embodiment of the present invention; Figure 10 This is a schematic diagram of the structure of a design system for a dual-state transmission-type polarization conversion metasurface according to an embodiment of the present invention; Figure 11 This is a schematic diagram of an electronic device in an embodiment of the present invention. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this disclosure clearer, the following detailed description is provided in conjunction with specific embodiments and the accompanying drawings.

[0025] It should be noted that, unless otherwise defined, the technical or scientific terms used in the embodiments of this invention should have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms "first," "second," and similar terms used in the embodiments of this invention do not indicate any order, quantity, or importance, but are merely used to distinguish different components. Terms such as "comprising" or "including" mean that the element or object preceding the word encompasses the elements or objects listed following the word and their equivalents, without excluding other elements or objects. Terms such as "connected" or "linked" are not limited to physical or mechanical connections, but can include electrical connections, whether direct or indirect. Terms such as "upper," "lower," "left," and "right" are only used to indicate relative positional relationships; when the absolute position of the described object changes, the relative positional relationship may also change accordingly.

[0026] like Figure 1 As shown, this embodiment of the invention provides a design method for a dual-state transmission-type polarization conversion metasurface, comprising the following steps: S1: Establish a parameterized model of the transmission-type polarization conversion metasurface unit. Sample multiple sets of structural parameters within the range of design variables. Simulate the electromagnetic response of the metasurface unit in the first and second operating states under each set of structural parameters. Obtain the dual-state performance index corresponding to each structural parameter based on the electromagnetic response. The dual-state performance index includes the target frequency phase difference error, the average phase difference error in the band, the average cross-polarization transmission amplitude in the first operating state, the average cross-polarization transmission amplitude in the second operating state, the minimum polarization conversion rate in the first operating state, the minimum polarization conversion rate in the second operating state, and the dual-state amplitude imbalance.

[0027] S2: Construct a standard sample set based on each structural parameter and its corresponding two-state performance index. Train a pre-defined surrogate model based on the standard sample set to establish a non-linear mapping between structural parameters and two-state performance indices. After training, construct a unified objective function for the surrogate model, which is used to quickly evaluate the comprehensive performance of any set of structural parameters based on the difference between the two-state performance indices predicted by the surrogate model and the pre-defined target performance indices. The dimensions of the pre-defined target performance indices and the indices of each dimension correspond one-to-one with the two-state performance indices.

[0028] S3: A differential evolution algorithm is used to perform a global search on the surrogate model to obtain a globally optimal solution set and search history. Several feedback candidate points are selected from the search history. Each feedback candidate point is simulated and the corresponding dual-state performance index is obtained. Each feedback candidate point and its corresponding dual-state performance index are added to the standard sample set as new feedback samples to obtain a new sample set.

[0029] S4: Based on the new sample set, the agent model is retrained. On the agent model that has been retrained, the globally optimal solution obtained by the differential evolution algorithm is used as the center, and the covariance matrix adaptive evolution strategy is used to find the local optimal structure parameters.

[0030] S5: Simulation is performed based on local optimal structural parameters to obtain the optimal simulated electromagnetic response of the metasurface unit in the first and second working states. The optimal simulated electromagnetic response is used to determine whether the polarization conversion and phase modulation requirements in the target frequency band are met.

[0031] In this embodiment, a transmissive polarization conversion metasurface unit operating in the 6.2 GHz to 7 GHz frequency band is used as an example to illustrate the design method. Figure 2As shown, the metasurface unit adopts a five-layer structure, consisting of a top metal gate, an upper dielectric layer, an intermediate functional layer, a lower dielectric layer, and a bottom metal gate, from top to bottom. The top and bottom metal gates are orthogonally arranged. The upper and lower dielectric layers are made of the same material with a dielectric constant of 2.55. The intermediate functional layer integrates a diode, and the switching between the first and second operating states is achieved by controlling the diode's on / off state through external power supply. The diode's power supply method and connection structure can be conventionally designed according to actual needs, and will not be elaborated here.

[0032] The intermediate functional layer is the main optimization target in this embodiment, such as... Figure 3 As shown.

[0033] This embodiment has a metasurface unit periodicity. Taking 22.5mm as the thickness, the optimization focus is on the thickness parameters of the intermediate functional layer and interlayer layers. To facilitate subsequent optimization, the structural parameters to be optimized are defined as a design variable vector: , where T denotes the transpose of the matrix. Indicates the side length of the functional layer's outer frame. Indicates the width of the outer frame of the functional layer. Indicates the dimensions of the central rectangular structure. Indicates the width of the open branch. Indicates the length of the open branch. Indicates the diode connection gap. Indicates the thickness of the upper dielectric layer. Indicates the thickness of the underlying dielectric layer. Indicates the thickness of the intermediate air layer. This indicates the radius of the metal cylinder. Two metal cylinders are provided, both located within the lower dielectric layer. One connects the bottom metal gate to the functional layer frame; the other connects to the feed line of the bottom metal gate (the feed line is located at the center of the bottom metal gate, see details). Figure 2 The design variables are the central rectangle of the intermediate functional layer and the metasurface unit. These 10 design variables together constitute the optimization space. A set of design variables is a set of structural parameters of the metasurface unit, which is a 10-dimensional vector.

[0034] In this embodiment, the diode is configured with two operating states for dual-state simulation: the first operating state (state "0") is that the upper diode is on and the lower diode is off; the second operating state (state "1") is that the upper diode is off and the lower diode is on. By calculating the transmitted electromagnetic response separately in both operating states, joint optimization of the dual-state transmission polarization conversion performance and phase difference performance is achieved. Figure 4 As shown, for any set of design variables Full-wave electromagnetic simulations were performed using CST (CST Studio Suite, a full-wave electromagnetic simulation software) under both of the aforementioned working states. The cross-polarization transmission coefficient, common-polarization transmission coefficient, polarization conversion rate, and phase response, among other indicators, were extracted from the simulation results under both states.

[0035] In this embodiment, the optimization objective requires not only maintaining high cross-polarization transmission capability within the 6.2 GHz to 7 GHz frequency band, but also a predetermined phase difference between the two states near the target frequency. 6.9 GHz is selected as the key optimization frequency, and the dual-state phase difference error, in-band average phase difference error, dual-state average cross-polarization transmission amplitude, dual-state minimum polarization conversion rate, and dual-state amplitude balance are used as the main evaluation metrics. To this end, a 7-dimensional target vector is defined. y The dual-state performance indicators described in this invention specifically include: target frequency phase difference error, in-band average phase difference error, average cross-polarization transmission amplitude in the first operating state, average cross-polarization transmission amplitude in the second operating state, minimum polarization conversion rate in the first operating state, minimum polarization conversion rate in the second operating state, and dual-state amplitude imbalance. By incorporating a total of seven performance indicators from the first and second operating states into a unified optimization framework and performing dual-state joint optimization, the problem that existing single-state design methods cannot simultaneously consider the dual-state synergistic performance of reconfigurable metasurfaces can be solved.

[0036] In this embodiment, parameter sampling is performed within the upper and lower bounds of the design variables. 800 sets of structural parameters are taken as initial valid samples. For each set of structural parameters, CST is called to perform full-wave electromagnetic simulation under both diode operating states, i.e., 1600 simulations. Then, the dual-state performance indicators corresponding to each set of structural parameters are obtained. Specifically, the cross-polarization transmission coefficient, common-polarization transmission coefficient, polarization conversion rate, and phase response under the dual states are extracted from the simulation results. Based on these indicators, the target frequency phase difference error, in-band average phase difference error, average cross-polarization transmission amplitude in the first operating state, average cross-polarization transmission amplitude in the second operating state, minimum polarization conversion rate in the first operating state, minimum polarization conversion rate in the second operating state, and dual-state amplitude imbalance (i.e., dual-state performance indicators) corresponding to each set of design variables (i.e., each set of structural parameters) are calculated. The polarization conversion rate can be expressed as: , This represents the cross-polarization transmission coefficient. This represents the same polarization transmission coefficient. The closer the PCR value is to 1, the more complete the polarization conversion.

[0037] Then, proceed to step S2 above, where a standard sample set is constructed based on each structural parameter and its corresponding dual-state performance index, including: S21: Construct an initial sample set by treating each set of structural parameters and its corresponding two-state performance index as a sample. After obtaining the initial sample set, standardize the structural parameters as input and the two-state performance index as the output target index.

[0038] S22: Normalize the structural parameters and two-state performance indices in the initial sample set according to the following formulas:

[0039] in, These are the structural parameters before normalization. and These are the mean and standard deviation of the structural parameters, respectively. These are the normalized structural parameters; The two-state performance index before normalization. and These are the mean and standard deviation of the two-state performance index, respectively. This refers to the normalized two-state performance metrics. Normalization unifies structural parameters (different dimensions and orders of magnitude for each design variable) and two-state performance metrics (different units and ranges for phase difference error, amplitude, conversion rate, etc.) to the same scale, preventing variables with large values ​​from dominating the optimization process. Normalized data has a more balanced distribution, which is beneficial for gradient descent optimization of neural networks, accelerating convergence and improving prediction accuracy. A unified format specification is defined for the standard sample set, facilitating data management for subsequent training, validation, and testing of surrogate models.

[0040] S23: Based on the normalized structural parameters and the corresponding normalized dual-state performance index, a standard sample set is obtained for subsequent surrogate model training and search optimization.

[0041] In this embodiment, the surrogate model is implemented using a Multilayer Perceptron (MLP) to establish a nonlinear mapping from 10-dimensional structural parameters to 7-dimensional two-state performance metrics. An MLP is a feedforward neural network whose basic structure includes an input layer, multiple hidden layers, and an output layer. Each neuron in each layer receives the output of the previous layer, performs a weighted summation, and then undergoes a nonlinear transformation using an activation function (such as ReLU, Sigmoid, or Tanh) before passing the output layer layer by layer. In this embodiment, the MLP has 10 nodes in the input layer (corresponding to 10 design variables), 7 nodes in the output layer (corresponding to 7 two-state performance metrics), and several fully connected hidden layers. The number of layers and neurons can be adjusted based on training performance. Compared to deep residual networks (such as ResNet-18), MLP offers more lightweight training and faster prediction speeds.

[0042] After the standard sample set is constructed, it is divided into training, validation, and test sets, and then used to train the surrogate model, establishing a nonlinear mapping between 10-dimensional structural parameters and 7-dimensional two-state performance metrics. The loss function of the surrogate model is: , in, The data-driven loss term represents the two-state performance metric predicted by the surrogate model. Two-state performance metrics obtained from simulation The mean square error between them For proxy model parameters, is the regularization coefficient. The prediction accuracy of the surrogate model directly affects the optimization results of the difference evolution and covariance matrix adaptive evolution strategies. This loss function combines data-driven approaches with regularization to ensure that the surrogate model has both accuracy and generalization ability.

[0043] Training ends when the surrogate model reaches the preset maximum number of training rounds, or when the validation set error no longer decreases for several consecutive rounds, and the surrogate model with the smallest validation set error is saved.

[0044] After training, a unified objective function is constructed for the surrogate model, used for rapid evaluation in the subsequent optimization phase. In this embodiment, the unified objective function is: ,in, To represent a 10-dimensional vector of structural parameters, For the dimensions of two-state performance metrics, , The first two-state performance index predicted by the surrogate model Dimensional performance indicators The first of the preset target performance indicators Dimensional performance indicators For the first The weighting coefficients corresponding to the performance indicators. The output value is the fitness value of that set of structural parameters. The smaller the fitness value, the better the overall performance of that set of structural parameters. Weighting coefficients It can be flexibly adjusted according to engineering needs, so that the design method can adapt to the different emphases of various performance indicators in different application scenarios.

[0045] In step S3, such as Figure 5 As shown, this embodiment first employs the Differential Evolution Algorithm (DE) to perform a global search on the surrogate model, obtaining the globally optimal solution and search history in the sense of the surrogate model, including the following steps: First, initialize the population, which contains multiple individuals, each corresponding to a set of structural parameters. .

[0046] Then, for the current population, for each individual in the population... Perform mutation and crossover operations to obtain the trial vector. Specifically, for the... The middle generation individual From the current number Three distinct individuals are randomly selected from the population. , , According to the formula Construct a mutation vector, where, For individuals in the current population The mutation vector, This is the differential scaling factor.

[0047] Subsequently, the mutation vector With the individual Perform cross operations to generate test vectors : ; in, Represents an individual test vector The The values ​​of each dimension (parameter) are candidate solutions generated by the cross operation. Represents an individual Mutation vector The Values ​​of each dimension Indicates the first The middle generation individual The The values ​​of each dimension, that is, the original genes of the current individual. Represents the crossover probability, controlling the probability of taking a value from the mutation vector; its value range is typically [range missing]. , It is a random number. To force the variation dimension index, from A randomly selected integer ( (For parameter dimensions), ensuring that at least one dimension of the experimental vector comes from the mutation vector. For the current individual Generate a random number for each dimension. When random number Less than or equal to crossover probability Or, when the dimension is a randomly selected forced mutation dimension, the test vector The value in this dimension is taken from the mutation vector. The corresponding dimension, otherwise taken from the current individual. The corresponding dimension.

[0048] Then, the population is updated using a greedy criterion: , experimental vectors With individuals The comparison is made, and individuals with better fitness are retained for the next generation of the population. Fitness is the unified objective function. The output value, fitness value, reflects the comprehensive performance of structural parameters. The smaller the fitness value, the closer the set of structural parameters are to the preset target performance index, and the better the comprehensive performance.

[0049] Then, using the updated next-generation population as the new current population, the above operations are iteratively executed until the termination condition is met, and the global best solution and search history are output. There are generally multiple global best solutions.

[0050] In this embodiment, the differential evolution algorithm parameters are set as follows: population size NP = 30, maximum iteration max Gen = 40, differential scaling factor F = 0.6, and crossover probability CR = 0.9. The total evaluation budget for the differential evolution algorithm is approximately 1200 iterations, providing a basis for subsequent candidate point selection.

[0051] In step S3, after the differential evolution algorithm completes the global search, several feedback candidate points are selected from the search history to return to the real CST for dual-state simulation to obtain more reliable real samples, which are then used to correct the surrogate model.

[0052] Specifically, several candidate feedback points are selected from the search history, including: Select the candidate point with the best fitness from the search history as the first feedback candidate point; then filter the remaining candidate points in order of fitness from best to worst until the number of selected feedback candidate points reaches the preset value.

[0053] The remaining candidate points are then screened in descending order of fitness, including: S301: In order of fitness from best to worst, each time select the candidate point with the best fitness from the remaining candidate points as the current candidate point for the current screening. S302: For each current candidate point, calculate the minimum normalized distance between the current candidate point and each of the selected feedback candidate points. The calculation formula is as follows: , in, As the current candidate point, The selected candidate points constitute the set of all selected feedback candidate points. A feedback candidate point is one of the selected candidate point sets. and These are the upper and lower bounds for each design variable, respectively. , , and All are 10-dimensional vectors corresponding to the design variables. Representing vectors sum vector Subtraction of each dimension This represents the division of vectors dimension by dimension.

[0054] S303: If the minimum normalized distance is greater than the preset threshold, the current candidate point will be selected as the next feedback candidate point.

[0055] In this embodiment, the number of feedback candidate points is set to 5, and the minimum relative distance threshold between candidate points is set to 0.05. Five candidate points with relatively good surrogate model evaluation and non-overlapping distribution are selected, and each of them is subjected to real dual-state full-wave simulation to form a real feedback closed loop. By selecting candidate points from the search history for real electromagnetic simulation and adding the simulation results back to the sample set as new feedback samples, the technical risk that the prediction bias of the pure surrogate model may cause the optimization results to fail can be overcome.

[0056] In step S4, such as Figure 6 As shown, after completing the real feedback and updating the standard dataset, the surrogate model is retrained. The termination condition for the second training of the surrogate model is: when the retrained surrogate model reaches the preset maximum number of training rounds, or when the error on the new validation set no longer decreases for several consecutive rounds, the retraining ends, and the surrogate model with the smallest error on the new validation set is saved.

[0057] Then, based on the updated surrogate model, and centering on the globally optimal solution obtained by the differential evolution algorithm, a covariance matrix adaptive evolution strategy (CMA-ES) is used for local optimization to obtain the locally optimal structure parameters, including: S41: Obtain multiple globally better solutions output by the differential evolution algorithm.

[0058] S42: For each globally optimal solution, perform the following local optimization operations: S421: Use this globally optimal solution as the initial center point. Initialize step size Covariance Matrix ; S422: For the current generation of search, sample and generate as follows: A new individual: , in, Let be the mean vector of the current generation. The current step size, Given the current covariance matrix, With a mean of 0 and a covariance matrix of The normal distribution; S423: Calculate the fitness of each new individual based on the retrained surrogate model, and select the individual with the best fitness. Each individual is an elite individual, among whom For the proportion of elites, 0 < <1; S424: Calculate the weighted average of the elite individuals as the updated mean vector. And update the current step size based on elite individuals. and the current covariance matrix ; S425: Iterate through steps S422~S424 until the termination condition is met, and output the local optimization result corresponding to the global better solution; S43: Compare the local optimization results corresponding to each globally optimal solution, and select the one with the best fitness as the final local optimal structure parameter.

[0059] In this embodiment, the parameters of the covariance matrix adaptive evolution strategy are set as follows: number of samples per round. Maximum number of iterations: 30; initial step size: Elite ratio =0.5, step size decay coefficient 0.9, covariance update rate 0.3. The local optimization budget is approximately 600 surrogate model evaluations. After local optimization, the locally optimal structural parameters in the sense of retraining and updating the surrogate model can be obtained.

[0060] In this embodiment, after obtaining the locally optimal structural parameters through local optimization, CST is called again to perform a two-state full-wave simulation on these locally optimal structural parameters to obtain the optimal simulated electromagnetic response. The optimal simulated electromagnetic response is then compared with the preset target performance indicators to determine whether it meets the transmission polarization conversion and phase modulation requirements within the target frequency band. If it does, the locally optimal structural parameters are output as the final design result.

[0061] In some embodiments, if the conditions are not met, the process may return to the step of constructing a standard sample set based on each structural parameter and its corresponding two-state performance index, or return to the step of performing a global search on the surrogate model using a differential evolution algorithm.

[0062] In this embodiment, the final result Lf =21.5651, Wf =1.8498, Ld =10.8753, Wb =1.2984, Lb =7.4202, Gd =0.1500, hsub1 =4.0041, hsub2 =1.8549, hair =3.9955, r =0.4872 (unit: mm). Figure 7 , Figure 8 , Figure 9 The transmission amplitude, phase difference, and polarization conversion efficiency of the transmission unit structure under two different conditions were demonstrated, all of which met the design standards.

[0063] Furthermore, to verify the effectiveness of the proposed optimization design method compared to a single optimization algorithm, under the same dual-state objective evaluation method, design variable range, surrogate model, and total evaluation budget as the above embodiments, particle swarm optimization algorithm, individual differential evolution algorithm, and genetic algorithm were used as comparative examples for statistical comparison. To facilitate unified statistical analysis and intuitive comparison of the surrogate model evaluation results of different algorithms, in this embodiment and each comparative example, the fitness value output by the surrogate model is further converted into a positive scoring index, defined as: Score = exp(-fitness), where fitness is the output value of the unified objective function of the surrogate model, i.e., the fitness value. Since the exponential function is a monotonic function, the above transformation does not change the ranking relationship between different candidate solutions; the smaller the fitness, the larger the score, indicating better performance. Therefore, the score can be used as a positive evaluation index for statistical comparison between the embodiments and comparative examples.

[0064] In this embodiment, the design method proposed in this invention is run 20 times independently to obtain the score statistics. In Comparative Example 1, the particle swarm optimization algorithm is used and run 20 times independently under the same conditions to obtain the score statistics. In Comparative Example 2, a separate differential evolution algorithm is used and run 20 times independently under the same conditions to obtain the score statistics. In Comparative Example 3, a genetic algorithm is used and run 20 times independently under the same conditions to obtain the score statistics. The specific score statistics are shown in the table below.

[0065]

[0066] As can be seen from the comparison results of the above embodiments and comparative examples, under the same two-state evaluation conditions, the same surrogate model, and the same total evaluation budget, the design method adopted in this invention achieved the highest average score among all methods, and the standard deviation was smaller. This indicates that the method of this invention is superior to comparative examples one, two, and three in terms of overall optimization effect and repeated running stability. In particular, compared with comparative example two, which uses the differential evolution algorithm alone, the method of this invention performs better in terms of average score, optimal score, and worst score, indicating that this method can further improve the overall optimization effect while maintaining good stability. Compared with comparative examples one and three, the advantages of the method of this invention are more obvious, showing that when using particle swarm optimization algorithm or genetic algorithm alone, the results fluctuate relatively greatly and the consistency of repeated running is insufficient, while the fusion optimization and real feedback retraining route proposed in this invention can obtain more stable and better results.

[0067] This invention does not simply use two algorithms side-by-side, but rather employs the differential evolution algorithm for the initial global search, fully leveraging its ability to rapidly locate potential optimal regions within a large, continuous parameter space. Building upon this, a covariance matrix adaptive evolution strategy is used for the subsequent local optimization, allowing it to adaptively adjust the search step size and covariance structure using search history, further refining the search direction and scale within the optimal region. Simultaneously, by introducing real feedback samples and a surrogate model retraining process between the two stages, this invention forms a dedicated optimization process suitable for the inverse design of two-state transmissive polarization conversion metasurfaces. The covariance matrix adaptive evolution strategy is robust, fast, requires fewer user parameters, and is suitable for handling continuous optimization problems with strong variable correlations. Its complementary relationship with the differential evolution algorithm makes it even more suitable for the two-state transmissive polarization conversion metasurface design problem involved in this invention.

[0068] The foregoing has described specific embodiments of the present invention. In some cases, the actions or steps described in the embodiments of the present invention may be performed in a different order than that shown in the embodiments and the desired results may still be achieved. Furthermore, the processes depicted in the drawings do not necessarily require a specific or sequential order to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0069] Based on the same concept, embodiments of the present invention also provide a design system for a dual-state transmission-type polarization conversion metasurface. This is applied to servers. Figure 10 As shown, the design system for a dual-state transmission polarization conversion metasurface includes: a sample construction module 100, a surrogate model training module 200, a global search module 300, a local optimization module 400, and a simulation verification module 500.

[0070] The sample construction module 100 is used to establish a parameterized model of the transmission-type polarization conversion metasurface unit. Within the range of design variables, multiple sets of structural parameters are sampled, and the electromagnetic response of the metasurface unit in the first and second working states under each set of structural parameters is simulated. Based on the electromagnetic response, the dual-state performance index corresponding to each structural parameter is obtained. The dual-state performance index includes the target frequency phase difference error, the average phase difference error in the band, the average cross-polarization transmission amplitude in the first working state, the average cross-polarization transmission amplitude in the second working state, the minimum polarization conversion rate in the first working state, the minimum polarization conversion rate in the second working state, and the dual-state amplitude imbalance.

[0071] The surrogate model training module 200 is used to construct a standard sample set based on each structural parameter and its corresponding two-state performance index, train a preset surrogate model based on the standard sample set, establish a nonlinear mapping between structural parameters and two-state performance index, and after training, construct a unified objective function for the surrogate model, which is used to quickly evaluate the comprehensive performance of any set of structural parameters based on the difference between the two-state performance index predicted by the surrogate model and the preset target performance index.

[0072] The global search module 300 is used to perform a global search on the surrogate model using the differential evolution algorithm to obtain a global optimal solution set and search history. Several feedback candidate points are selected from the search history, each feedback candidate point is simulated and the corresponding dual-state performance index is obtained. Each feedback candidate point and its corresponding dual-state performance index are added as new feedback samples to the standard sample set to obtain a new sample set.

[0073] The local optimization module 400 is used to retrain the surrogate model based on a new sample set. On the retrained surrogate model, the module uses the globally optimal solution obtained by the differential evolution algorithm as the center and adopts the covariance matrix adaptive evolution strategy to perform local optimization and obtain the locally optimal structural parameters.

[0074] The simulation verification module 500 is used to perform simulations based on locally optimal structural parameters to obtain the final simulated electromagnetic response of the metasurface unit in the first and second working states. Based on the final simulated electromagnetic response, it is determined whether the polarization conversion and phase modulation requirements in the target frequency band are met.

[0075] For ease of description, the above system is described by dividing it into various modules based on their functions. Of course, in implementing the embodiments of the present invention, the functions of each module can be implemented in one or more software and / or hardware.

[0076] The system described in the above embodiments is applied to the corresponding method in the foregoing embodiments and has the beneficial effects of the corresponding method embodiments, which will not be repeated here.

[0077] Based on the same inventive concept, embodiments of the present invention also provide an electronic device, which includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the program, it implements the method described in any of the above embodiments.

[0078] This invention provides a non-volatile computer storage medium storing at least one executable instruction that can execute the method described in any of the above embodiments.

[0079] Figure 11 This embodiment illustrates a more specific hardware structure of an electronic device, which may include a processor 501, a memory 502, an input / output interface 503, a communication interface 504, and a bus 505. The processor 501, memory 502, input / output interface 503, and communication interface 504 are interconnected internally via the bus 505.

[0080] The processor 501 can be implemented using a general-purpose CPU (Central Processing Unit), microprocessor, application-specific integrated circuit (ASIC), or one or more integrated circuits, and is used to execute relevant programs to implement the technical solutions provided in the embodiments of the present invention.

[0081] The memory 502 can be implemented in the form of ROM (Read Only Memory), RAM (Random Access Memory), static storage device, dynamic storage device, etc. The memory 502 can store the operating system and other application programs. When the technical solution provided by the method embodiment of the present invention is implemented by software or firmware, the relevant program code is stored in the memory 502 and is called and executed by the processor 501.

[0082] Input / output interface 503 is used to connect input / output modules to realize information input and output. Input / output modules can be configured as components in the device (not shown in the figure) or externally connected to the device to provide corresponding functions. Input devices may include keyboards, mice, touch screens, microphones, various sensors, etc., and output devices may include displays, speakers, vibrators, indicator lights, etc.

[0083] Communication interface 504 is used to connect a communication module (not shown in the figure) to enable communication between this device and other devices. The communication module can communicate via wired means (such as USB, Ethernet cable, etc.) or wireless means (such as mobile network, WIFI, Bluetooth, etc.).

[0084] Bus 505 includes a pathway for transmitting information between various components of the device (e.g., processor 501, memory 502, input / output interface 503, and communication interface 504).

[0085] It should be noted that although the above-described device only shows the processor 501, memory 502, input / output interface 503, communication interface 504, and bus 505, in specific implementations, the device may also include other components necessary for normal operation. Furthermore, those skilled in the art will understand that the above-described device may only include the components necessary for implementing the embodiments of the present invention, and does not necessarily include all the components shown in the figures.

[0086] Those skilled in the art should understand that the discussion of any of the above embodiments is merely exemplary and is not intended to imply that the scope of this application is limited to these examples; within the framework of this application, the technical features of the above embodiments or different embodiments can also be combined, the steps can be implemented in any order, and there are many other variations of the different aspects of this application as described above, which are not provided in detail for the sake of brevity.

[0087] This application is intended to cover all such substitutions, modifications, and variations that fall within the broad scope of the embodiments of this invention. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the embodiments of this invention should be included within the protection scope of this application.

Claims

1. A design method of a dual-state transmissive polarization conversion metasurface, characterized in that, Includes the following steps: A parameterized model of a transmission-type polarization conversion metasurface unit is established. Multiple sets of structural parameters are sampled within the range of design variables. The electromagnetic response of the metasurface unit in the first and second operating states under each set of structural parameters is simulated. Based on the electromagnetic response, the dual-state performance index corresponding to each structural parameter is obtained. The dual-state performance index includes the target frequency phase difference error, the average phase difference error in the band, the average cross-polarization transmission amplitude in the first operating state, the average cross-polarization transmission amplitude in the second operating state, the minimum polarization conversion rate in the first operating state, the minimum polarization conversion rate in the second operating state, and the dual-state amplitude imbalance. A standard sample set is constructed based on each structural parameter and its corresponding two-state performance index. A preset surrogate model is trained based on the standard sample set to establish a nonlinear mapping between the structural parameters and the two-state performance index. After training, a unified objective function of the surrogate model is constructed to quickly evaluate the comprehensive performance of any set of structural parameters based on the difference between the two-state performance index predicted by the surrogate model and the preset target performance index. A differential evolution algorithm is used to perform a global search on the surrogate model to obtain a global optimal solution set and search history. Several feedback candidate points are selected from the search history. Each feedback candidate point is simulated and the corresponding dual-state performance index is obtained. Each feedback candidate point and the corresponding dual-state performance index are added as new feedback samples to the standard sample set to obtain a new sample set. Based on the new sample set, the surrogate model is retrained. On the retrained surrogate model, the locally optimal structural parameters are obtained by using the covariance matrix adaptive evolution strategy as the center, with the globally optimal solution obtained by the differential evolution algorithm as the center. Simulations are performed based on the locally optimal structural parameters to obtain the optimal simulated electromagnetic response of the metasurface unit in the first and second operating states. The optimal simulated electromagnetic response is then used to determine whether the polarization conversion and phase modulation requirements within the target frequency band are met.

2. The design method for a dual-state transmission-type polarization conversion metasurface as described in claim 1, characterized in that, The unified objective function is: , wherein, is a set of structural parameters, is a dimension of the dual-state performance index, is a first dimensional performance index predicted by the proxy model, is a first dimensional performance index of a preset target performance index, is a weight coefficient corresponding to the first dimensional performance index.

3. The design method for a dual-state transmission-type polarization conversion metasurface as described in claim 1, characterized in that, The construction of a standard sample set based on each structural parameter and its corresponding dual-state performance index includes: Each set of structural parameters and its corresponding two-state performance index is used as a sample to construct an initial sample set; The structural parameters and two-state performance indices in the initial sample set are normalized according to the following formulas: in, These are the structural parameters before normalization. and These are the mean and standard deviation of the structural parameters, respectively. These are the normalized structural parameters; The two-state performance index before normalization. and These are the mean and standard deviation of the two-state performance index, respectively. The normalized two-state performance index; Based on the normalized structural parameters and the corresponding normalized two-state performance index, a standard sample set is obtained.

4. The design method for a dual-state transmission-type polarization conversion metasurface as described in claim 1, characterized in that, The step of performing a global search using the differential evolution algorithm on the surrogate model to obtain a globally optimal solution set and search history includes the following steps: S31: Initialize the population, which contains multiple individuals, each individual corresponding to a set of structural parameters; S32: For the current population, for each individual in the population Perform the following operations: S33: Randomly select three distinct individuals from the current population, and then apply the formula... Construct a mutation vector, where, For individuals in the current population The mutation vector, , , For from the first Three distinct individuals randomly selected from the population. This is the difference scaling factor; S34: Transfer the mutation vector With the individual Perform cross operations to generate test vectors ; S35: Transfer the test vector With the individual The individuals with better fitness are compared and retained to enter the next generation of the population. The fitness is the output value of the unified objective function. S36: Iterate through steps S32 to S35 until the termination condition is met, and output the global optimal solution and search history.

5. The design method for a dual-state transmission-type polarization conversion metasurface as described in claim 1, characterized in that, The step of filtering several feedback candidate points from the search history includes: Select the candidate point with the best fitness from the search history as the first feedback candidate point; The remaining candidate points are screened in descending order of fitness until the number of selected feedback candidate points reaches the preset value. The step of filtering the remaining candidate points in descending order of fitness includes: Following the order of fitness from best to worst, each time the candidate point with the best fitness is selected from the remaining candidate points as the current candidate point for that screening. For each current candidate point, calculate the minimum normalized distance between the current candidate point and each of the selected feedback candidate points. The calculation formula is as follows: ， in, As the current candidate point, The selected candidate points constitute the set of all selected feedback candidate points. A feedback candidate point is one of the selected candidate point sets. and These are the upper and lower bounds for each design variable, respectively. If the minimum normalized distance is greater than a preset threshold, the current candidate point is selected as the next feedback candidate point.

6. The design method for a dual-state transmission-type polarization conversion metasurface as described in claim 1, characterized in that, The process of using the globally optimal solution obtained by the differential evolution algorithm as the center, and employing a covariance matrix adaptive evolution strategy to perform local optimization to obtain locally optimal structure parameters includes the following steps: S41: Obtain multiple globally optimal solutions output by the differential evolution algorithm; S42: For each of the globally optimal solutions, perform the following local optimization operations: S421: Use this globally optimal solution as the initial center point. Initialize step size Covariance Matrix ; S422: For the current generation of search, sample and generate as follows: A new individual: , in, Let be the mean vector of the current generation. The current step size, Given the current covariance matrix, The mean is 0 and the covariance matrix is The normal distribution; S423: Calculate the fitness of each new individual based on the retrained surrogate model, and select the individual with the best fitness. Each individual is an elite individual, among whom The proportion of elites; S424: Calculate the weighted average of the elite individuals as the updated mean vector. And update the current step size based on the elite individuals. and the current covariance matrix ; S425: Iteratively execute steps S422~S424 until the termination condition is met, and output the local optimization result corresponding to the global better solution; S43: Compare the local optimization results corresponding to each of the globally optimal solutions, and select the one with the best fitness as the final local optimal structure parameter.

7. A dual-state transmission-type polarization conversion metasurface obtained using the design method described in any one of claims 1-6, characterized in that, include: A periodically arranged metasurface unit, wherein the metasurface unit is provided with a top metal gate, an upper dielectric layer, an intermediate functional layer, a lower dielectric layer and a bottom metal gate in sequence from top to bottom; The top metal grid and the bottom metal grid are arranged orthogonally to each other; The upper dielectric layer and the lower dielectric layer are made of dielectric materials with the same dielectric constant; The intermediate functional layer is a metal structure layer, including a functional layer outer frame with the geometric center coincident, a central rectangle, and two open-circuit branches disposed between the functional layer outer frame and the central rectangle. One end of each open-circuit branch is connected to an inner corner of the functional layer outer frame, and the other end is connected to the midpoint of one side of the central rectangle through a diode. A metal cylinder connects the bottom metal grid to the outer frame of the functional layer, and another metal cylinder connects the feed line of the bottom metal grid to the central rectangle. By switching the diode between its on and off states, the metasurface unit switches between a first operating state and a second operating state to perform transmission-type polarization conversion on the incident electromagnetic wave.

8. A design system for a dual-state transmission-type polarization conversion metasurface, characterized in that, include: The sample construction module is used to establish a parameterized model of the transmission-type polarization conversion metasurface unit. Within the range of design variables, multiple sets of structural parameters are sampled, and the electromagnetic response of the metasurface unit in the first and second operating states under each set of structural parameters is simulated. Based on the electromagnetic response, the dual-state performance index corresponding to each structural parameter is obtained. The dual-state performance index includes the target frequency phase difference error, the average phase difference error in the band, the average cross-polarization transmission amplitude in the first operating state, the average cross-polarization transmission amplitude in the second operating state, the minimum polarization conversion rate in the first operating state, the minimum polarization conversion rate in the second operating state, and the dual-state amplitude imbalance. The surrogate model training module is used to construct a standard sample set based on each structural parameter and its corresponding two-state performance index, train a preset surrogate model based on the standard sample set, establish a nonlinear mapping between the structural parameters and the two-state performance index, and after training, construct a unified objective function for the surrogate model, which is used to quickly evaluate the comprehensive performance of any set of structural parameters based on the difference between the two-state performance index predicted by the surrogate model and the preset target performance index. The global search module is used to perform a global search on the surrogate model using the differential evolution algorithm to obtain a global optimal solution set and search history. It selects several feedback candidate points from the search history, simulates each feedback candidate point and obtains the corresponding dual-state performance index, and adds each feedback candidate point and the corresponding dual-state performance index as new feedback samples to the standard sample set to obtain a new sample set. The local optimization module is used to retrain the surrogate model based on the new sample set. On the retrained surrogate model, the module uses the globally optimal solution obtained by the differential evolution algorithm as the center and adopts the covariance matrix adaptive evolution strategy to perform local optimization to obtain the local optimal structure parameters. The simulation verification module is used to perform simulation based on the local optimal structural parameters to obtain the final simulated electromagnetic response of the metasurface unit in the first and second working states, and to determine whether the polarization conversion and phase modulation requirements in the target frequency band are met based on the final simulated electromagnetic response.

9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor executes the program, it implements the design method as described in any one of claims 1-6.

10. A computer storage medium, characterized in that, The storage medium stores at least one executable instruction that causes the processor to perform the design method as described in any one of claims 1-6.

Images