A wearable antenna intelligent design method based on multi-working condition electromagnetic performance stability

Abstract

The application discloses a kind of based on multi-working condition electromagnetic performance stable wearable antenna intelligent design method, it is related to wireless communication and electromagnetic design technical field.A kind of based on multi-working condition electromagnetic performance stable wearable antenna intelligent design method, comprising the following steps: constructing parameterized antenna model, multi-working condition electromagnetic performance screening, acquisition multi-working condition electromagnetic simulation data, forward proxy model training and reverse optimization design.The application is by constructing parameterized antenna model, with medium parameter and antenna geometric dimension as design variable, form the candidate antenna structure in discrete parameter space, and introduce multi-working condition electromagnetic performance screening step, the candidate structure is discriminated in plane working condition with reflection coefficient, and the invalid sample that does not satisfy preset threshold is screened out, solves the technical problem of large amount of invalid simulation calculation in early stage of multi-working condition antenna design, and greatly improves design efficiency.

Description

Technical Field

[0001] This invention relates to the fields of wireless communication and electromagnetic design technology, and in particular to a smart design method for wearable antennas based on stable electromagnetic performance under multiple operating conditions. Background Technology

[0002] With the rapid development of wearable electronic devices, flexible communication systems, and human body sensing networks, wearable antennas are increasingly widely used in smart terminals, health monitoring, and human body communication. Compared to traditional rigid antennas, wearable antennas are inevitably affected by various working conditions such as bending, attachment, and deformation during actual use, which can cause significant changes in their electromagnetic performance, manifesting as resonant frequency shift, impedance mismatch, and decreased radiation performance, seriously affecting the stability and reliability of communication systems.

[0003] Currently, the design of wearable antennas mainly relies on electromagnetic simulation or empirical parameter adjustments under a single operating condition, making it difficult to simultaneously meet the comprehensive performance requirements under various deformation conditions. While some studies have attempted to introduce multi-condition simulation analysis, the sheer volume of simulation calculations, long design cycles, and lack of systematic data modeling and reverse design methods make it difficult to achieve efficient design while ensuring performance stability under multiple operating conditions. Furthermore, traditional trial-and-error or rule-based optimization methods struggle to fully exploit potential optimal solutions in the design parameter space, resulting in design results that are sensitive to changes in operating conditions and limited engineering applicability.

[0004] Therefore, there is an urgent need for a smart design method for wearable antennas that can achieve stable electromagnetic performance, high design efficiency, and engineering feasibility under multiple operating conditions. Summary of the Invention

[0005] This invention proposes a smart design method for wearable antennas that can achieve stable electromagnetic performance, high design efficiency, and engineering feasibility under multiple operating conditions.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows: A smart design method for wearable antennas based on stable electromagnetic performance under multiple operating conditions includes the following steps: A parameterized antenna model is constructed, taking the periodic electromagnetic bandgap antenna structure as the object, and using the dielectric material parameters and antenna geometry as design variables to construct candidate antenna structures in the discrete parameter space; Multi-condition electromagnetic performance screening: Under the plane condition with zero bending radius, the reflection coefficient performance of candidate antenna structures is evaluated. Only when the minimum reflection coefficient of the candidate structure in the target operating frequency band is less than or equal to the preset threshold is it retained and proceeded to the next step. Multi-condition electromagnetic simulation data is collected, and electromagnetic simulation calculations are performed on the selected candidate antenna structures under various bending radius conditions to obtain the reflection coefficient curve, resonant frequency, and frequency offset information under each condition. Based on the active learning mechanism, a multi-condition positive surrogate model is constructed using a Gaussian process regression model on the current simulation data. The information gain of unsimulated samples is evaluated according to the acquisition function, and high-value samples are iteratively selected to supplement the training set until the model converges. Forward surrogate model training: Using the constructed multi-condition electromagnetic performance dataset, a forward surrogate model that can characterize the mapping relationship between antenna structure parameters and electromagnetic response under various conditions is trained. The reverse optimization design utilizes a trained forward surrogate model to solve the antenna structure parameters in reverse through an optimization algorithm, ensuring that the reflection coefficients meet the preset constraints under all operating conditions. This results in a periodic electromagnetic bandgap antenna structure with stable electromagnetic performance under multiple operating conditions.

[0007] As a preferred embodiment of the present invention, in the multi-condition electromagnetic performance screening step, the reflection coefficient evaluation is performed only within the target operating frequency band; if the minimum reflection coefficient of the candidate structure in this frequency band is higher than a preset threshold, the candidate structure is directly determined to be an invalid structure, and no further multi-condition simulation calculations are performed on it.

[0008] As a preferred technical solution of the present invention, in the multi-condition electromagnetic simulation data acquisition step, the bending radius conditions used include at least two different bending radius conditions; and an active learning strategy based on Gaussian process regression is adopted, the initial sample set is obtained by Latin hypercube sampling, and in each subsequent iteration, the unsimulated sample with the largest amount of information is selected for simulation according to the acquisition function, wherein the acquisition function simultaneously considers the coverage of the sample in the parameter space, the uncertainty of the model prediction, and the degree to which the sample is close to the design boundary.

[0009] As a preferred technical solution of the present invention, in the forward proxy model training step, the reflection coefficient curve is first subjected to principal component analysis for dimensionality reduction, and then transformed into a low-dimensional feature vector; at the same time, the bending radius operating condition parameter is encoded; then the antenna structure parameter and the operating condition encoding are jointly input into the neural network model, so that the model can simultaneously predict the electromagnetic response under different operating conditions.

[0010] As a preferred technical solution of the present invention, the neural network model is a multilayer perceptron structure, in which a random deactivation layer is introduced in the hidden layer. During the training phase, some neurons are randomly discarded with a preset probability. During the prediction phase, all neurons are retained and the mean and variance are output, thereby quantifying the uncertainty of the model prediction and improving the robustness of multi-condition electromagnetic performance prediction.

[0011] As a preferred technical solution of the present invention, during the multi-condition electromagnetic simulation data acquisition process, a unique sample identification number is assigned to each candidate antenna structure. This identification number remains unchanged throughout the entire process of data acquisition, model training, and reverse design, and is used to realize sample tracking and management.

[0012] As a preferred technical solution of the present invention, in the reverse optimization design step, an optimization objective function is constructed with the goal of minimizing the maximum deviation of each resonant frequency relative to the target frequency under multiple operating conditions, and the reflection coefficient in the target frequency band under all operating conditions is less than or equal to a preset threshold is used as a hard constraint condition.

[0013] As a preferred technical solution of the present invention, the reverse optimization design step adopts the differential evolution algorithm for global optimization, including population initialization, mutation operation, crossover operation and selection operation; in the mutation operation, the mutation vector of each target individual is obtained by adding the weighted difference of three randomly selected different individuals to the basis vector; in the selection operation, if the offspring individual satisfies all hard constraints of reflection coefficient and its objective function value is better than that of the parent individual, then the offspring replaces the parent and enters the next generation of population.

[0014] As a preferred technical solution of the present invention, the comprehensive analysis includes: the antenna structure parameters include the relative permittivity of the dielectric, the dielectric thickness, the length of the radiating patch and the width of the periodic unit; each design variable is searched within a preset value range, and the geometric parameter space is generated by an equidistant grid sampling method to generate candidate structures, and the sampling interval is determined according to the value range of each design variable and the preset number of sampling points.

[0015] A computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements a smart design method for wearable antennas based on stable electromagnetic performance under multiple operating conditions.

[0016] The present invention has the following advantages: This invention constructs a parameterized antenna model, using dielectric parameters and antenna geometry as design variables to form candidate antenna structures in a discrete parameter space. It also introduces a multi-condition electromagnetic performance screening step to quickly determine the reflection coefficient of candidate structures under planar conditions, filtering out invalid samples that do not meet the preset threshold. This solves the technical problem of large amount of invalid simulation calculations in the early stage of multi-condition antenna design and greatly improves design efficiency.

[0017] This invention constructs a multi-condition positive proxy model by using a Gaussian process regression model based on active learning. It dynamically selects the sample with the most information through the acquisition function for iterative supplementation and performs electromagnetic simulation evaluation under various bending radius conditions to obtain reflection coefficient curves, resonant frequencies and frequency offset information. It then constructs a multi-condition electromagnetic performance dataset, solving the technical problems of blind sample selection and low training set construction efficiency in the multi-condition electromagnetic simulation data acquisition process, and realizing adaptive optimization of data acquisition.

[0018] This invention achieves rapid prediction of the antenna's electromagnetic response by performing principal component analysis to reduce the dimensionality of the reflection coefficient curve and constructing a neural network forward surrogate model based on the encoding of operating conditions. On this basis, with the optimization objective of minimizing the maximum deviation of the resonant frequency under multiple operating conditions and the reflection coefficient constraint as a hard constraint, a differential evolution algorithm is used for inverse parameter optimization design. This solves the technical problem that traditional design methods cannot balance the stability of performance under multiple operating conditions and design efficiency. It obtains a periodic electromagnetic bandgap antenna structure with stable electromagnetic performance under multiple operating conditions, which significantly improves the engineering applicability of wearable antennas under various operating conditions such as bending, attachment, and deformation. Attached Figure Description

[0019] 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 in the following description are only schematic diagrams of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort. Figure 1 This is a flowchart illustrating a smart design method for wearable antennas based on stable electromagnetic performance under multiple operating conditions, as used in an embodiment of the present invention.

[0020] Figure 2 The optimal antenna structure obtained by reverse design in the embodiments of the present invention is S under the Flat condition. 11 Parameter curve graph.

[0021] Figure 3 This is a specific absorption rate (SAR) distribution diagram of the optimal antenna structure in the Flat operating condition in the embodiment of the present invention.

[0022] Figure 4 S is the optimal antenna structure in this embodiment of the invention after bending along the x-axis under multiple operating conditions. 11 Parameter curve graph.

[0023] Figure 5 The S-axis of the optimal antenna structure in this embodiment of the invention after bending along the y-axis under multiple operating conditions. 11 Parameter curve graph. Detailed Implementation

[0024] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this invention, and not all embodiments. Based on the embodiments of this invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this invention.

[0025] Example 1: A smart design method for wearable antennas based on stable electromagnetic performance under multiple operating conditions, such as... Figures 1-5 As shown, it includes the following steps: Step S1: Construct a parameterized antenna model. Taking the periodic electromagnetic bandgap antenna structure as the object, the dielectric material parameters and the antenna geometry are used as design variables to construct candidate antenna structures in the discrete parameter space. In this step, the periodic electromagnetic bandgap antenna structure includes at least a dielectric substrate, a metallic transmission line, periodic electromagnetic bandgap elements, an impedance matching component, and a feeding structure. The dielectric material parameters include at least the relative permittivity εᵣ of the dielectric substrate, the dielectric loss tangent tanδ, and the dielectric thickness H. t The antenna's geometric dimensions include at least the length L1 of the radiating patch, the width W2 of the electromagnetic bandgap element, the element spacing, and the metal linewidth. The range of design variable values ​​is determined based on the antenna's operating frequency band and the physical size constraints of the wearable scenario.

[0026] When generating the candidate antenna structure parameter space, each design variable is discretely sampled. The sampling method uses equally spaced grid sampling, with a sampling interval of... Determined by the following formula: in The preset number of sampling points for the i-th design variable.

[0027] For example, setting the number of sampling points for the relative permittivity εᵣ of the dielectric substrate. =5, Medium thickness H t Number of sampling points =6, the number of sampling points of the length L1 of the radiating patch unit. =5, the number of sampling points for width W2 If the value is 5, then a total of 5×6×5×5=750 candidate antenna structures are generated.

[0028] Each candidate structure is assigned a unique sample identifier (ID) for sample tracking and management throughout the subsequent data acquisition, model training, and reverse engineering processes. The parameter space of the candidate antenna structure is represented as follows: in This is a vector of antenna structure parameters. , where m is the number of design variables.

[0029] Step S2, Multi-condition electromagnetic performance screening: Under the plane condition with zero bending radius, the reflection coefficient performance of the candidate antenna structure is evaluated. Only when the minimum reflection coefficient of the candidate structure in the target operating frequency band is less than or equal to the preset threshold is it retained and proceeded to the next step. In this step, the reflection coefficient performance of all candidate antenna structures is first rapidly evaluated under planar conditions (bending radius R = ∞, i.e., zero bending deformation). Using full-wave electromagnetic simulation software (such as HFSS or CST), the reflection coefficient performance of each candidate structure in the target operating frequency band is calculated. Internal reflection coefficient .

[0030] Set the target operating frequency band as The preset threshold for the reflection coefficient is γ=-10dB.

[0031] For each candidate structure, extract the minimum reflection coefficient within that frequency band. ; The selection criteria are: if and only if When the value is less than or equal to γ, the candidate structure is deemed to have passed the screening, is retained, and enters the subsequent multi-condition simulation steps; otherwise, it is directly deemed an invalid structure and is eliminated.

[0032] The screening mechanism can be expressed as a formula: like If so, then the candidate structure is eliminated.

[0033] For example, among 750 initial candidate structures, after screening under planar conditions, valid samples that meet the reflection coefficient constraints are retained, thereby significantly reducing the amount of computation required for subsequent multi-condition simulations.

[0034] Step S3: Multi-condition electromagnetic simulation data acquisition. For the selected candidate antenna structures, electromagnetic simulation calculations are performed under various bending radius conditions to obtain the reflection coefficient curve, resonant frequency, and frequency offset information under each condition. Based on the active learning mechanism, a Gaussian process regression model is used to construct a multi-condition positive surrogate model for the current simulation data. The information gain of unsimulated samples is evaluated according to the acquisition function, and high-value samples are iteratively selected to supplement the training set until the model converges. In this step, the candidate structures selected in step S2 are subjected to full-wave electromagnetic simulation under various bending radius conditions. These bending radius conditions include at least two different bending radius conditions to characterize the changes in the antenna's electromagnetic performance under different structural deformation states.

[0035] This embodiment selects four typical bending radius conditions: R1=30mm, R2=40mm, R3=50mm, and R4=60mm, to simulate the deformation state of the wearable antenna under different wearing parts or different degrees of bending, such as the arm, wrist, and torso.

[0036] For each candidate structure x that passes the screening, electromagnetic simulation calculations are performed under the four bending conditions mentioned above. Obtain the reflection coefficient curves under various operating conditions. Resonant frequency and frequency offset information ,in =2.45GHz is the target resonant frequency.

[0037] After the simulation is completed, the obtained data is associated with and stored with the corresponding structural parameter x, operating condition parameter R and sample identifier ID to form an initial multi-operating condition electromagnetic performance dataset.

[0038] To improve data acquisition efficiency and reduce redundant simulations, this step introduces an active learning mechanism based on Gaussian process regression. The specific process is as follows: First, from the candidate structures that have passed the screening, the Latin hypercube sampling method is used to select... One initial sample (in this embodiment) =32), complete the full-wave electromagnetic simulation of it under all four bending conditions, and construct the initial training set. ; Then, based on the current training set, a temporary positive surrogate model is constructed using Gaussian process regression. This model takes the antenna structure parameter x as input and outputs the electromagnetic performance under multiple operating conditions (such as the resonant frequency under each operating condition). ).

[0039] Gaussian process regression models can output predicted mean. and prediction variance The latter is used to quantify the uncertainty of the model at the current sample point.

[0040] Next, for each unsimulated candidate structure Calculate its acquisition function value The acquisition function used in this embodiment comprehensively considers the coverage of samples in the parameter space, the uncertainty of model prediction, and the degree to which samples are close to the design boundary. Its expression is: in, The coverage term is calculated based on the minimum Euclidean distance between the candidate sample and the set of simulated samples, and is used to improve the coverage uniformity of the parameter space. The uncertainty term is given by the variance predicted by the Gaussian process regression model and is used to explore areas where the model's understanding is insufficient. The boundary term is calculated based on the distance from the candidate sample to the parameter space boundary, and is used to encourage exploration of the parameter space boundary region.

[0041] The weighting coefficients are set as follows: =1.0, =1.0, =0.6.

[0042] In each iteration, the one with the largest acquisition function value is selected. One unsimulated sample (in this embodiment) =4) Perform the next batch of simulations and add the results to the training set to update the Gaussian process regression model.

[0043] Repeat the above iterative process until the convergence condition is met: the change in the root mean square error of the Gaussian process regression model on the validation set is less than a preset threshold (e.g., the change in the root mean square error of the resonant frequency prediction is less than 1MHz), or the maximum number of evaluation samples is reached (in this embodiment, the maximum number of evaluation samples is set to 400).

[0044] Ultimately, this results in an electromagnetic performance dataset covering multiple operating conditions. This is used for subsequent training of the positive proxy model.

[0045] Step S4: Forward surrogate model training. Using the constructed multi-condition electromagnetic performance dataset, a forward surrogate model is trained that can characterize the mapping relationship between antenna structure parameters and electromagnetic response under various conditions. In this step, the multi-condition electromagnetic performance dataset constructed in step S3 is used. A positive surrogate model M is trained to quickly predict the electromagnetic response under different structural parameters x and operating conditions R.

[0046] Due to the reflection coefficient curve High dimensionality (e.g., 1002 frequency points) directly used as model output leads to high computational complexity and training difficulties.

[0047] Therefore, this step first reduces the dimensionality of the reflection coefficient curve, and then trains the neural network model.

[0048] Principal component analysis (PCA) was used for dimensionality reduction.

[0049] Specifically, the dataset Return loss curves of all samples (Unit: dB) is used as the processing object.

[0050] Principal component analysis is performed on the set of RL curves to calculate the eigenvector matrix U and the mean vector μ. The top k principal components whose sum of contribution rates exceeds 99% are retained, and each original curve is compressed from Nf dimensions (Nf=1002) to a k-dimensional eigenvector. In this embodiment, k is the optimal value automatically selected by Optuna from {30,40,50,60,80}.

[0051] PCA transformation satisfies: In subsequent model training, feature vectors As a dimension-reduced representation of the reflection coefficient curve, the reconstruction formula is: .

[0052] The operating condition parameters are encoded using one-hot encoding. For the four bending operating conditions R∈{30,40,50,60}mm, they are encoded as e(R=30)=[1,0,0,0], e(R=40)=[0,1,0,0], e(R=50)=[0,0,1,0], and e(R=60)=[0,0,0,1]. The antenna structure parameter x is concatenated with the operating condition encoding vector e(R) as the model input.

[0053] The forward surrogate model employs a multilayer perceptron (MLP) neural network structure, and its expression is: The network structure contains three hidden layers, with 256, 256, and 128 neurons in each layer, respectively, and ReLU is used as the activation function.

[0054] To prevent overfitting, a Monte Carlo Dropout (MC Dropout) layer is introduced into the hidden layers, randomly dropping some neurons with a preset probability pdrop during the training phase. The model output layer contains k+1 neurons, each corresponding to a k-dimensional feature vector. and predicting the resonant frequency .

[0055] The model training uses the Adam optimizer, and the loss function is based on the reconstructed return loss curve. While ensuring the overall fitting accuracy, the deep region is weighted, and a smoothing regularization term is introduced to suppress non-physical oscillations.

[0056] Training round settings N The epochs are 2000, the learning rate and batch size are automatically determined through Bayesian optimization, and an adaptive learning rate decay strategy based on validation loss is adopted. During training, the model performance is monitored using a validation set to prevent overfitting.

[0057] After training, the forward surrogate model can achieve rapid prediction. Since MC Dropout remains enabled during the prediction phase, a set of predictions can be obtained through multiple forward propagations (e.g., 10 times), allowing the calculation of the mean μ and variance σ², quantifying the uncertainty of the model's predictions. , .

[0058] The model takes less than 0.1 seconds to predict a single sample and has a root mean square error of less than 5 MHz for predicting the resonant frequency, which fully meets the real-time requirements of reverse optimization design.

[0059] Step S5: Reverse optimization design. Using the trained forward surrogate model, under the premise of ensuring that the reflection coefficients meet the preset constraints under all operating conditions, the antenna structure parameters are solved in reverse by the optimization algorithm, thereby obtaining a periodic electromagnetic bandgap antenna structure with stable electromagnetic performance under multiple operating conditions.

[0060] In this step, based on the forward surrogate model M trained in step S4, an inverse optimization problem under multiple operating conditions is constructed to solve for the optimal antenna structure parameters. The optimization objective is to minimize the maximum deviation of the resonant frequency from the target frequency under all bending conditions, thus ensuring the frequency stability of the antenna under multiple operating conditions. The objective function expression is: in, =2.45GHz is the target resonant frequency. The objective function aims to make the resonant frequency of the antenna as close as possible to the target frequency under four bending conditions, thereby ensuring frequency stability under multiple conditions.

[0061] The hard constraint is: for all operating conditions R∈{30,40,50,60}mm and all frequency points f∈[2.2,2.8]GHz, the reflection coefficient is... ≤-10dB (i.e., return loss) ≥10dB).

[0062] The constraint is predicted and determined by the forward surrogate model M. When the predicted reflection coefficient curve under any operating condition has a point higher than -10dB in the target frequency band, the candidate solution is considered to not satisfy the constraint.

[0063] This embodiment uses the differential evolution algorithm to solve the above optimization problem. The algorithm parameters are set as follows: population size NP=20, maximum number of iterations Gmax=120, and mutation factor F is... Random values ​​are selected within the range, with a crossover probability CR = 0.7.

[0064] The algorithm process includes: Population initialization: Randomly generate NP individuals in the parameter space X as the initial population. Each individual is an antenna structure parameter vector. .

[0065] Mutation operation: For each target individual Three distinct individuals are randomly selected from the current population. , , (and with) (Different), generate mutation vectors: like If a certain dimension exceeds the range of the design variable, it will be truncated to the boundary value.

[0066] Cross operation: target individual With the mutation vector Perform binomial crossover to generate experimental individuals. .

[0067] For each dimension j=1,2,3,4, if or ,but ,otherwise ,in, The dimension index is randomly selected, ensuring that at least one dimension comes from the mutated vector.

[0068] Selection Operation: First, use the positive surrogate model M to predict the experimental individuals. Resonant frequency under all four operating conditions and reflection coefficient curve .

[0069] like If the hard constraint on the reflection coefficient is satisfied under all operating conditions, then calculate its objective function value. Otherwise, a large penalty value will be imposed on the objective function.

[0070] Then, With parental individuals If a comparison is made, ,but Alternative Enter the next generation of the population.

[0071] Repeat the above iterations until the maximum number of iterations is reached or the population converges; finally, output the optimal individual. These are the structural parameters of a periodic electromagnetic bandgap antenna that meet the requirements for stable performance under multiple operating conditions.

[0072] The optimal antenna structure obtained by reverse optimization is verified by full-wave electromagnetic simulation. Its reflection coefficient curve and resonant frequency under various operating conditions are obtained to verify the stability of multi-condition performance. In this step, the optimized antenna structure obtained in step S5 is... Verification simulations were performed using full-wave electromagnetic simulation software (such as HFSS or CST).

[0073] The simulation settings are consistent with the multi-condition simulation in step S3, including planar conditions and four bending conditions (R=30mm, 40mm, 50mm, 60mm). The reflection coefficient curves for each condition are obtained. and resonant frequency And calculate frequency offset .

[0074] The verification results are used to test the effectiveness of the reverse optimization design. The verification results of this embodiment show that: under planar conditions, the reflection coefficient within the target frequency band is below -10dB, the resonant frequency is 2.455GHz, and the deviation from the 2.45GHz target frequency is only 5MHz; under the four conditions of bending along the x-axis, the resonant frequencies are 2.4502GHz, 2.4490GHz, 2.4520GHz, and 2.4532GHz, respectively, with a maximum frequency deviation of 3.2MHz; under the four conditions of bending along the y-axis, the resonant frequencies are 2.4490GHz, 2.4520GHz, 2.4520GHz, and 2.4550GHz, respectively, with a maximum frequency deviation of 5MHz. Under all conditions, the reflection coefficient within the target frequency band remains below -10dB, meeting the design requirements.

[0075] A wearable antenna structure design scheme with stable electromagnetic performance under multiple operating conditions is provided, including optimal structural parameters, electromagnetic performance indicators under various operating conditions, and design verification results.

[0076] In this step, the optimal antenna structure parameters obtained through reverse optimization design will be... The electromagnetic performance predicted by the forward surrogate model is correlated and summarized with the full-wave simulation verification results to form a complete design scheme output. The output should include at least: Optimal antenna structure parameters: relative permittivity of the dielectric εᵣ, dielectric thickness H t The specific values ​​for the length L1 of the radiating patch and the width W2 of the periodic unit; Electromagnetic performance indicators under various operating conditions: resonant frequency, -10dB impedance bandwidth, and reflection coefficient curve data under planar operating conditions and four bending operating conditions; Biocompatibility assessment: The specific absorption rate distribution and maximum value under planar operating conditions are used to verify that it complies with FCC and ICNIRP safety standards; Design verification results: The error comparison between the predicted values ​​of the forward surrogate model and the full-wave simulation results proves that the model prediction accuracy meets the engineering requirements.

[0077] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart design method for wearable antennas based on stable electromagnetic performance under multiple operating conditions, characterized in that, Includes the following steps: A parameterized antenna model is constructed, taking the periodic electromagnetic bandgap antenna structure as the object, and using the dielectric material parameters and antenna geometry as design variables to construct candidate antenna structures in the discrete parameter space; Multi-condition electromagnetic performance screening: Under the plane condition with zero bending radius, the reflection coefficient performance of candidate antenna structures is evaluated. Only when the minimum reflection coefficient of the candidate structure in the target operating frequency band is less than or equal to the preset threshold is it retained and proceeded to the next step. Collect electromagnetic simulation data under multiple operating conditions, and perform electromagnetic simulation calculations on the selected candidate antenna structures under various bending radius conditions to obtain the reflection coefficient curve, resonant frequency and frequency offset information under each operating condition. Based on the active learning mechanism, a multi-condition positive surrogate model is constructed using a Gaussian process regression model on the current simulation data. The information gain of unsimulated samples is evaluated according to the acquisition function, and high-value samples are iteratively selected to supplement the training set until the model converges. Forward surrogate model training: Using the constructed multi-condition electromagnetic performance dataset, a forward surrogate model that can characterize the mapping relationship between antenna structure parameters and electromagnetic response under various conditions is trained. The reverse optimization design utilizes a trained forward surrogate model to solve the antenna structure parameters in reverse through an optimization algorithm, ensuring that the reflection coefficients meet the preset constraints under all operating conditions. This results in a periodic electromagnetic bandgap antenna structure with stable electromagnetic performance under multiple operating conditions.

2. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 1, characterized in that, In the multi-condition electromagnetic performance screening step, the reflection coefficient evaluation is only performed within the target operating frequency band. If the minimum reflection coefficient of the candidate structure in this frequency band is higher than the preset threshold, the candidate structure is directly determined to be an invalid structure and will not be subjected to subsequent multi-condition simulation calculations.

3. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 1, characterized in that, In the multi-condition electromagnetic simulation data acquisition step, the bending radius conditions used include at least two different bending radius conditions; and an active learning strategy based on Gaussian process regression is adopted. The initial sample set is obtained through Latin hypercube sampling. In each subsequent iteration, the unsimulated sample with the most information is selected for simulation according to the acquisition function. The acquisition function simultaneously considers the coverage of the sample in the parameter space, the uncertainty of the model prediction, and the degree to which the sample is close to the design boundary.

4. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 1, characterized in that, In the forward proxy model training step, the reflection coefficient curve is first subjected to principal component analysis for dimensionality reduction, which transforms it into a low-dimensional feature vector; at the same time, the bending radius operating condition parameter is encoded; then the antenna structure parameter and the operating condition encoding are input into the neural network model, so that the model can simultaneously predict the electromagnetic response under different operating conditions.

5. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 4, characterized in that, The neural network model is a multilayer perceptron structure, in which a random deactivation layer is introduced in the hidden layer. During the training phase, some neurons are randomly discarded with a preset probability. During the prediction phase, all neurons are retained and the mean and variance are output, thereby quantifying the uncertainty of the model prediction and improving the robustness of electromagnetic performance prediction under multiple working conditions.

6. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 1, characterized in that, During the multi-condition electromagnetic simulation data acquisition process, a unique sample identifier number is assigned to each candidate antenna structure. This identifier number remains unchanged throughout the entire process of data acquisition, model training, and reverse design, and is used to track and manage the samples.

7. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 1, characterized in that, In the reverse optimization design step, an optimization objective function is constructed with the goal of minimizing the maximum deviation of each resonant frequency relative to the target frequency under multiple operating conditions, and the reflection coefficient in the target frequency band under all operating conditions is less than or equal to a preset threshold as a hard constraint condition.

8. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 7, characterized in that, The reverse optimization design steps employ differential evolution algorithm for global optimization, including population initialization, mutation operation, crossover operation, and selection operation. In the mutation operation, the mutation vector of each target individual is obtained by adding the weighted difference of three randomly selected different individuals to the basis vector. In the selection operation, if the offspring individual satisfies all hard constraints of reflection coefficients and its objective function value is better than that of the parent individual, then the offspring replaces the parent and enters the next generation of the population.

9. The intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability according to claim 1, characterized in that, The antenna structure parameters include the relative permittivity of the dielectric, the dielectric thickness, the length of the radiating patch, and the width of the periodic element; Each design variable is searched within a preset value range. The geometric parameter space uses an equidistant grid sampling method to generate candidate structures. The sampling interval is determined based on the value range of each design variable and the preset number of sampling points.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the intelligent design method for wearable antennas based on multi-condition electromagnetic performance stability as described in any one of claims 1 to 9.

Images