Antenna parameter optimization method and apparatus, and related device

Abstract

The invention provides an antenna parameter optimization method and device and related equipment, and relates to the technical field of antennae. According to the technical scheme of the invention, an initial function model corresponding to an antenna is established according to antenna parameters and user demands, feature extraction is carried out on the initial function model, and target features having the greatest influence on antenna performance are determined; according to the method, the initial function model is subjected to dimensionality reduction according to the target characteristics, and the antenna parameters are optimized, so that the calculation cost and the simulation time can be greatly reduced while high precision is kept, and the accuracy and the efficiency of antenna parameter optimization are improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of antennas, in particular to an antenna parameter optimization method and device and related equipment. BACKGROUND

[0002] Antenna parameter optimization is a complex process involving multiple parameters and multiple objectives, and there is a complex interaction between these parameters and objectives. In related technical solutions, when optimizing antenna parameters, only global optimization is performed, and local sensitivity fine-tuning is ignored. This leads to difficulty in correcting the precision loss caused by dimensionality reduction or model approximation after global search, and the final accurate optimal solution cannot be locked. Moreover, there is a lack of effective dimensionality reduction. When the number of antenna design parameters increases, the number of required electromagnetic simulations increases exponentially, and the computational cost is extremely high. Thus, related technologies have the problem of low accuracy and efficiency for antenna parameter optimization. SUMMARY

[0003] The present application provides an antenna parameter optimization method, device and related equipment, which solves the problem of low accuracy and efficiency for antenna parameter optimization in related technologies.

[0004] To solve the above problems, the present application is implemented as follows:

[0005] In a first aspect, an antenna parameter optimization method is provided, which includes:

[0006] establishing an initial function model corresponding to an antenna according to antenna parameters and requirement information, the antenna parameters including geometric parameters corresponding to the antenna, the requirement information including an optimization target for optimizing the antenna parameters, and the initial function model using parameters from a multi-dimensional parameter space corresponding to the antenna;

[0007] performing feature extraction on the initial function model to obtain target features, the target features being features having the greatest impact on the performance of the antenna;

[0008] performing dimensionality reduction processing on the initial function model based on the target features to obtain a target function model;

[0009] optimizing the antenna parameters according to the target function model to obtain optimized antenna parameters.

[0010] Optionally, the step of establishing an initial function model corresponding to an antenna according to antenna parameters and requirement information includes:

[0011] generating a target optimization function according to input requirement information, and analyzing the antenna to obtain antenna parameters;

[0012] generate a plurality of constraint conditions corresponding to the antenna parameters based on the requirement information, the plurality of constraint conditions being used to constrain different antenna attributes corresponding to the antenna, the antenna attributes including at least one of a frequency range of the antenna, a gain range of the antenna, an in-band axial ratio of the antenna, and a size of the antenna;

[0013] generate an initial function model corresponding to the antenna based on the antenna parameters, the plurality of constraint conditions, and the target optimization function.

[0014] Optionally, the feature extraction on the initial function model to obtain a target feature includes:

[0015] perform feature extraction and simulation on the initial function model based on a fast global sensitivity analysis algorithm (FGSA) to obtain a response set, the response set including a plurality of response samples, each of the response samples corresponding to a feature point in the multi-dimensional parameter space;

[0016] determine a target feature vector and a target feature value according to the plurality of response samples, the target feature vector being a feature vector corresponding to a response sample that has the greatest impact on the direction of the antenna among the plurality of response samples, and the target feature value being an impact value corresponding to the target feature vector;

[0017] wherein the target feature includes the target feature vector and the target feature value.

[0018] Optionally, the determination of the target feature vector and the target feature value according to the plurality of response samples includes:

[0019] determine a Euclidean distance between each first response sample and a corresponding adjacent sample among the plurality of response samples to obtain a plurality of distance values, the plurality of distance values corresponding one-to-one to the plurality of response samples, wherein the adjacent sample corresponding to the first response sample is a response sample closest to the first response sample among the plurality of response samples, and the first response sample is an arbitrary response sample;

[0020] calculate a plurality of displacement vectors corresponding to the plurality of distance values one-to-one to obtain the plurality of displacement vectors corresponding to the plurality of distance values one-to-one, and calculate a plurality of response changes corresponding to the plurality of distance values respectively to obtain a plurality of response change values corresponding to the plurality of distance values one-to-one;

[0021] generate a positioning matrix according to the plurality of displacement vectors and the plurality of response change values, an i-th row of the positioning matrix being a product of an i-th displacement vector and an i-th response change value;

[0022] determine the target feature vector and the target feature value according to the positioning matrix.

[0023] Optionally, the determining the target feature vector and the target feature value according to the positioning matrix comprises:

[0024] performing eigenvalue decomposition on the positioning matrix to obtain a plurality of feature vectors and a plurality of feature values, the plurality of feature vectors corresponding to the plurality of feature values one by one;

[0025] determining a target feature value in the plurality of feature values and a target feature vector in the plurality of feature vectors, the target feature value being a maximum feature value in the plurality of feature values, and the target feature vector being a feature vector corresponding to the target feature value;

[0026] the dimension reduction processing of the initial function model based on the target feature to obtain a target function model comprises:

[0027] the dimension reduction processing of the initial function model according to the target feature vector and the target feature value to obtain a target function model.

[0028] Optionally, the optimizing the antenna parameters according to the target function model to obtain optimized antenna parameters comprises:

[0029] generating an initial Kriging surrogate model according to the target function model;

[0030] performing global search and simulation on the initial Kriging surrogate model based on a particle swarm optimization algorithm (PSO) to obtain an updated Kriging surrogate model;

[0031] performing local fine-tuning on the updated Kriging surrogate model based on a trust region algorithm (TR) to obtain a target optimization strategy;

[0032] optimizing the antenna parameters according to the target optimization strategy to obtain optimized antenna parameters.

[0033] In a second aspect, an embodiment of the present application provides an antenna parameter optimization device, and the device comprises:

[0034] a establishing module configured to establish an initial function model corresponding to an antenna according to antenna parameters and requirement information, wherein the antenna parameters comprise geometric parameters corresponding to the antenna, the requirement information comprises an optimization target for optimizing the antenna parameters, and parameters used by the initial function model are derived from a multi-dimensional parameter space corresponding to the antenna;

[0035] an extracting module configured to perform feature extraction on the initial function model to obtain a target feature, wherein the target feature is a feature that has the greatest impact on the performance of the antenna;

[0036] a processing module, configured to perform dimension reduction processing on the initial function model based on the target feature, to obtain a target function model;

[0037] an optimization module, configured to optimize the antenna parameter according to the target function model, to obtain an optimized antenna parameter.

[0038] In a third aspect, the present application provides an electronic device, including a processor, a memory, and a computer program stored in the memory and executable in the processor, and when the computer program is executed by the processor, the steps in the method according to the first aspect are implemented.

[0039] In a fourth aspect, the present application provides a computer readable storage medium, and the computer readable storage medium stores a computer program, and when the computer program is executed by a processor, the steps in the method according to the first aspect are implemented.

[0040] In a fifth aspect, the present application provides a computer program product, including computer instructions, and when the computer instructions are executed by a processor, the steps in the method according to the first aspect are implemented.

[0041] The present application provides an antenna parameter optimization method and device and related equipment, the method including: establishing an initial function model corresponding to an antenna according to antenna parameters and requirement information, the antenna parameters including geometric parameters corresponding to the antenna, the requirement information including an optimization target for optimizing the antenna parameters, and the initial function model using parameters derived from a multi-dimensional parameter space corresponding to the antenna; performing feature extraction on the initial function model to obtain a target feature, the target feature being a feature having the greatest impact on the performance of the antenna; performing dimension reduction processing on the initial function model based on the target feature to obtain a target function model; and optimizing the antenna parameter according to the target function model to obtain an optimized antenna parameter. The technical solution of the present application establishes an initial function model corresponding to an antenna according to antenna parameters and user requirements, determines a target feature having the greatest impact on the performance of the antenna by performing feature extraction on the initial function model, performs dimension reduction on the initial function model according to the target feature, and optimizes the antenna parameter, thereby being able to greatly reduce the calculation cost and simulation time while maintaining high precision, and improving the accuracy and efficiency of antenna parameter optimization. BRIEF DESCRIPTION OF DRAWINGS

[0042] To make the technical solution of the present application clearer, the drawings needed in the description of the present application will be briefly introduced as follows. Obviously, the drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative labor.

[0043] Figure 1 A flowchart illustrating an antenna parameter optimization method provided in this application embodiment;

[0044] Figure 2 This is a schematic diagram of the structure of an antenna parameter optimization device provided in an embodiment of this application;

[0045] Figure 3 This is a schematic diagram of the structure of an electronic device provided in an embodiment of this application. Detailed Implementation

[0046] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0047] The terms "first," "second," etc., used in the embodiments of this application are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, system, product, or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products, or devices. Additionally, the use of "and / or" in this application indicates at least one of the connected objects, such as A and / or B and / or C, representing seven possibilities: including A alone, B alone, C alone, and the presence of both A and B, both B and C, both A and C, and the presence of A, B, and C.

[0048] See Figure 1 , Figure 1 This is a flowchart illustrating the antenna parameter optimization method provided in an embodiment of this application. Figure 1 As shown, the antenna parameter optimization method may include the following steps:

[0049] Step 101: Establish an initial function model for the antenna based on the antenna parameters and requirements information. The antenna parameters include the geometric parameters of the antenna, and the requirements information includes the optimization objective for optimizing the antenna parameters. The parameters used in the initial function model are derived from the multidimensional parameter space of the antenna.

[0050] In this embodiment, antenna parameters are those that need to be optimized. For example, antenna parameters may include at least one of the following: Gain: Represents the enhancement of the antenna's radiation capability in a specific direction relative to an ideal antenna (such as an isotropic antenna). Directivity: The ratio of the antenna's radiation intensity in a specific direction to its average radiation intensity in all directions, reflecting the antenna's radiation concentration capability. Beamwidth: The width of the antenna's main lobe, usually expressed in degrees, and divided into horizontal and vertical beamwidths. Standing Wave Ratio (VSWR): A parameter reflecting the degree of matching between the antenna and the feed line. To ensure optimal antenna performance, multiple parameters must be finely tuned, thus generating requirement information. One of the keys to this process is accurately quantifying the quality of the design. A scalar cost function is typically defined; the smaller the value of this function, the better the design quality. If multiple optimization objectives exist, they are converted into constraints. The requirement information can be the optimization objectives that the user needs to optimize for the antenna, such as improving or modifying data for certain antenna parameters.

[0051] An initial function model for the antenna is generated based on the antenna parameters and requirements. The parameters used in the initial function model are derived from the multidimensional parameter space corresponding to the antenna. Specifically, the multidimensional parameter space corresponding to the antenna is a high-dimensional parameter space.

[0052] Step 102: Extract features from the initial function model to obtain target features, which are the features that have the greatest impact on the performance of the antenna.

[0053] In this embodiment, feature extraction is performed on the multidimensional parameter space corresponding to the initial function model to obtain target features. It should be noted that the target features represent directions in the parameter space, where the antenna response changes decreasing. Therefore, by identifying the features that have the greatest impact on antenna performance, dimensionality reduction of the antenna's high-dimensional parameter space can be achieved. Specifically, in this embodiment, Fast Global Sensitivity Analysis (FGSA) is run in the high-dimensional parameter space, with a small amount of EM simulation sampling, to determine the target features. FGSA is a method for evaluating the sensitivity of a model's output to uncertainties in input parameters. It is particularly suitable for complex models and can quickly estimate the importance of each input parameter to the output result.

[0054] Step 103: Based on the target features, perform dimensionality reduction on the initial function model to obtain the target function model.

[0055] In this embodiment, dimensionality reduction is a data processing technique designed to reduce data complexity while retaining as much useful information as possible. When dealing with high-dimensional data, dimensionality reduction can help improve model efficiency, reduce computational costs, and potentially enhance the model's generalization ability. After dimensionality reduction, the complexity of the initial function model is reduced, and the objective function model becomes a simpler, more clearly structured model that more intuitively reveals the relationship between input and output. For example, dimensionality reduction can reduce the 10-dimensional data space corresponding to the initial function model to a 9-dimensional data space.

[0056] Step 104: Optimize the antenna parameters according to the objective function model to obtain the optimized antenna parameters.

[0057] In this embodiment, the antenna parameters initially obtained are optimized based on the generated objective function model to obtain optimized antenna parameters.

[0058] The technical solution of this application establishes an initial function model corresponding to the antenna based on antenna parameters and user requirements. By extracting features from the initial function model, the target features that have the greatest impact on antenna performance are determined. Based on these target features, the initial function model is dimensionality reduced, thereby optimizing the antenna parameters. This approach can significantly reduce computational costs and simulation time while maintaining high accuracy, thus improving the accuracy and efficiency of antenna parameter optimization.

[0059] In some feasible implementations, optionally, establishing the initial function model corresponding to the antenna based on antenna parameters and requirement information includes:

[0060] Based on the input requirements, a target optimization function is generated, and the antenna is analyzed to obtain antenna parameters;

[0061] Based on the aforementioned requirement information, multiple constraints are generated corresponding to the antenna parameters. These multiple constraints are used to constrain different antenna attributes corresponding to the antenna. The antenna attributes include at least one of the following: the antenna's frequency range, the antenna's gain range, the antenna's in-band axial ratio, and the antenna's size.

[0062] Based on the antenna parameters, the multiple constraints, and the objective optimization function, an initial function model corresponding to the antenna is generated.

[0063] In this embodiment, by analyzing the input requirements, a target optimization function is constructed. This function measures antenna performance metrics, such as minimizing loss, maximizing gain, and optimizing bandwidth. The antenna analysis process includes analyzing the antenna structure and operating principle to obtain key antenna parameters from the design. Antenna attributes include at least one of the following: frequency range, gain range, in-band axial ratio, and dimensions. Based on the requirement information, multiple constraints on the antenna parameters are generated. Then, using the antenna parameters, multiple constraints, and the target optimization function, an initial function model corresponding to the antenna is generated.

[0064] Specifically, the requirement information is the antenna optimization task, which is expressed by the following formula.

[0065]

[0066] Where, x * This represents the optimal parameter vector, where x is the geometric structure parameter determined based on the actual antenna conditions, and U(x) is the cost function defined according to user requirements. The antenna response is evaluated through EM simulation. Typically, this task is accompanied by a series of constraints:

[0067]

[0068] To simplify the above problem, constraints can be handled implicitly, that is, by introducing a penalty function to transform it into an unconstrained optimization problem:

[0069]

[0070] Among them, U P(x) Let the new objective function be represented by the original cost function U(x) and a series of penalty functions:

[0071]

[0072] Among them, c k (x) represents the degree of violation of the constraint, β k It is a coefficient that controls the weight of the penalty function.

[0073] Table 1 lists the symbols and terminology used in antenna design tasks. Table 2 illustrates various scenarios for antenna design optimization, where the described penalty functions explain relative constraint violations related to the assumed acceptable level. The use of quadratic powers guarantees U... P It becomes a differentiable function that violates constraints at the boundary of the feasible region, thus facilitating the exploration of the feasible region, since one or more constraints are variable in the optimal design.

[0074] Table 1. Symbols and Terminology in Antenna Optimization

[0076] Table 2 Examples of Antenna Structure Parameter Tuning Scenarios

[0077] The terms are explained below:

[0078] f - frequency;

[0079] In design and frequency The reflection coefficient modulus below;

[0080] G( - Achieved gain;

[0081] AR - Shaft ratio;

[0082] A( - Antenna size (e.g., area occupied).

[0083] The operating frequency range F of an antenna can be a single frequency band (for a single-frequency antenna) or a combination of multiple frequency bands. For a multi-frequency antenna, the target frequency range can be expressed as:

[0084]

[0085] Where N represents the number of frequency bands the antenna operates on.

[0086] Optionally, the step of extracting features from the initial function model to obtain target features includes:

[0087] Based on the Fast Global Sensitivity Analysis (FGSA) algorithm, feature extraction and simulation are performed on the initial function model to obtain a response set. The response set includes multiple response samples, and each response sample corresponds to a feature point in the multidimensional parameter space.

[0088] Based on the multiple response samples, a target feature vector and a target feature value are determined. The target feature vector is the feature vector corresponding to the response sample that has the greatest impact on the direction of the antenna among the multiple response samples, and the target feature value is the impact value corresponding to the target feature vector.

[0089] The target features include the target feature vector and the target feature value.

[0090] In this embodiment, based on the Fast Global Sensitivity Analysis (FGSA) algorithm, feature extraction and simulation are performed on the initial function model to obtain a response set. Each response sample represents the performance of the sample feature point in the multidimensional parameter space, and multiple response samples provide the output information of the model under different input conditions. The FGSA provided in this embodiment needs to meet the following conditions: low computational cost (e.g., no more than 100 EM simulations) and the ability to determine important directions in the design variable space that have a significant impact on antenna characteristics, rather than identifying individual parameters.

[0091] The target eigenvector is the eigenvector corresponding to the response sample that has the greatest impact on the antenna orientation among all response samples. The system analyzes numerous response samples to determine which feature (parameter combination) has the most significant impact on the output (antenna orientation, etc.). The target eigenvalue is the influence value associated with the target eigenvector, representing the strength or importance of that feature in influencing the response variable (such as antenna orientation).

[0092] Optionally, determining the target feature vector and target feature value based on the plurality of response samples includes:

[0093] In the plurality of response samples, the Euclidean distance between each first response sample and its corresponding neighbor sample is determined to obtain a plurality of distance values, wherein the plurality of distance values ​​correspond one-to-one with the plurality of response samples, wherein the neighbor sample corresponding to the first response sample is: the response sample that is closest to the first response sample among the plurality of response samples, and the first response sample is any one of the response samples;

[0094] Calculate the normalized displacement vectors corresponding to the plurality of distance values ​​to obtain a plurality of displacement vectors that correspond one-to-one with the plurality of distance values; and calculate the response changes corresponding to the plurality of distance values ​​to obtain a plurality of response change values ​​that correspond one-to-one with the plurality of distance values.

[0095] A positioning matrix is ​​generated based on the plurality of displacement vectors and the plurality of response change values, wherein the i-th row of the positioning matrix is ​​the product of the i-th displacement vector and the i-th response change value;

[0096] Based on the positioning matrix, the target feature vector and the target feature value are determined.

[0097] In this embodiment, the sample closest to the first response sample is found in the response sample set. This is achieved by calculating the Euclidean distance, i.e., measuring the straight-line distance between two samples in multidimensional space. For each distance value, a normalized displacement vector is generated based on its magnitude. The output change of the response sample corresponding to each distance value under a specific input change is evaluated to obtain the response change value. Multiple displacement vectors and their corresponding response change values ​​are combined. The i-th row of the localization matrix is ​​the product of the i-th displacement vector and the i-th response change value. By analyzing the localization matrix, important feature vectors and corresponding feature values ​​can be extracted, further revealing the importance of features in influencing the response output. Target feature vectors typically represent the dominant input features in the response change, while target feature values ​​reflect the influence of these features. This helps decision-makers quickly identify key parameters and supports effective design and optimization processes.

[0098] The FGSA (Fast Global Sensitivity Analysis) algorithm is summarized below:

[0099] enter:

[0100] Parameter space : Represents the design parameter set, with dimensions of .

[0101] Simulation model Input parameters The corresponding system response.

[0102] Sample size : Number of samples, usually generated using the Latin hypercube sampling method.

[0103] Output:

[0104] Feature vector : Indicates the most important direction in the system response.

[0105] Eigenvalues : Indicates the degree of influence of this direction on the system response.

[0106] The algorithm process includes the following steps:

[0107] 1. Sampling and Simulation: From the parameter space Generated in random sample points Electromagnetic simulation was performed to obtain the response. .

[0108] 2. Calculate the minimum distance sample: For each sample Find other recent samples And calculate the Euclidean distance between them:

[0109]

[0110] 3. Calculate the normalized displacement vector and response change:

[0111]

[0112]

[0113] 4. Calculate and relocate the matrix. :

[0114]

[0115] 5. Spectral analysis: For the relocation matrix... Perform eigenvalue decomposition to obtain the eigenvector e. j and the corresponding eigenvalue λ j .

[0116] 6. Dimensionality reduction: Sort the features by size and select the dominant direction with larger features for dimensionality reduction.

[0117] Specifically, through the migration matrix Perform spectral analysis to obtain the eigenvector e j These vectors represent the directions in the parameter space, where the antenna response changes decreasingly. The importance of each direction is determined by its corresponding eigenvalue λ. j Quantization is performed. Vector e j (j=1,…,n) form an orthogonal basis for the design variable space X. The results of FGSA can define the dimensionality reduction domain of the surrogate model, which will be used in the global optimization phase of the algorithm. This domain is defined by a few important eigenvectors. Their number is determined by the smallest integer N. d To determine whether it satisfies the following conditions:

[0118]

[0119] In the formula, C min It is a user-defined threshold, usually set to 90%.

[0120] Optionally, determining the target feature vector and the target feature value based on the positioning matrix includes:

[0121] The positioning matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvectors and multiple eigenvalues, and the multiple eigenvectors and multiple eigenvalues ​​correspond one-to-one.

[0122] Determine a target eigenvalue among the multiple eigenvalues, and determine a target eigenvector among the multiple eigenvectors, where the target eigenvalue is the largest eigenvalue among the multiple eigenvalues, and the target eigenvector is the eigenvector corresponding to the target eigenvalue;

[0123] The dimension reduction processing of the initial function model based on the target feature to obtain a target function model includes:

[0124] Perform dimension reduction processing on the initial function model according to the target eigenvector and the target eigenvalue to obtain a target function model.

[0125] In this embodiment, the eigenvalue decomposition of the positioning matrix is performed to determine multiple eigenvectors and multiple eigenvalues, and the eigenvalue and eigenvector with the greatest influence are determined. Then, the initial function model is dimensionally reduced according to the eigenvalue and eigenvector with the greatest influence to obtain a target function model.

[0126] Specifically, the dimension reduction model domain of FGSA aims to identify N d directions in space X that are crucial for the change of antenna response. These directions (eigenvectors e j (j = 1,..., n)) are used to define the dimension reduction region X d . The set X d serves as the effective region of the fast metamodel, that is, the dimensionally reduced region, for predicting the antenna response therein. The same region will also be used as the search domain in the global optimization stage. The set X d is defined as:

[0127]

[0128] where X d is the intersection of the original domain X and the set of vectors x c + a1e1 +... + a Nd e Nd , x c = [l + u] / 2 is the center of X, and a j , (j = 1,..., N d ) are real numbers.

[0129] Dimension reduction is crucial for the accuracy of the surrogate model. In particular, since dim(X d ) = N d < n, a usable metamodel can be constructed with less training data. Reducing the training set can improve the efficiency of the search process. At the same time, the effective region of the surrogate model contains the directions crucial for the change of antenna response, thus ensuring its design utility.

[0130] Optionally, optimizing the antenna parameters according to the objective function model to obtain optimized antenna parameters includes:

[0131] Generate an initial Kriging proxy model based on the objective function model;

[0132] Based on the particle swarm optimization algorithm (PSO), a global search and simulation are performed on the initial Kriging proxy model to obtain an updated Kriging proxy model.

[0133] The updated Kriging agent model is locally fine-tuned based on the Trust Domain Algorithm (TR) to obtain the target optimization strategy.

[0134] The antenna parameters are optimized according to the target optimization strategy to obtain the optimized antenna parameters.

[0135] In this embodiment, firstly, a Kriging model is used in the dimension-reduced domain X. d An internal proxy model is established, in which the trend function is set to a second-order polynomial and the correlation function is a Gaussian correlation function.

[0136]

[0137] Where h=[h1,h2,...,h n ] T θ j These are hyperparameters. The training dataset size is N. i N d , where N i These are user-defined coefficients. (Sample) (k=1,...,N i N d ) evenly distributed in X d Above. Proxy model s tmp (x) is based on the dataset Constructed, in which The antenna response was obtained through EM simulation. Subsequently, fill points were generated by adding the mean squared error (MSE) of the current surrogate model's predictions.

[0138]

[0139] For j=1,2,..., based on the extended training set The process is refined until the relative root mean square error of cross-validation is below the user-defined threshold Emax or the total number of samples exceeds 2NiNd. This design aims to place new training samples at positions corresponding to the maximum prediction model error, improving the global accuracy of the surrogate model across the entire domain Xd. At the end of the phase, the current model s tmp (x) becomes the initial agent s (0)(x). After the Particle Swarm Optimization (PSO) algorithm is completed, local parameter optimization needs to be performed on the original design variable space X. This is to ensure a truly optimal design. Global algorithm optimization is performed in the reduced-dimensional domain X. d The algorithm, which performs local parameter optimization, covers the most important directions in X, but not the entire space. The specific method used is the trust-region (TR) algorithm. TR solves the problem in the original space X. It generates continuous points x in X through iteration. (i) (i=0,1,...) come to work. The xth (i+1) The calculation for each point is as follows:

[0140]

[0141] in Let R be the value at the current iteration point x. (i) linear approximation model U L Consistent with U, but based on L (i) (x) is calculated instead of directly based on the simulated antenna response R(x). This is achieved by explicitly specifying U L For L (i) The dependence of (x) is emphasized. The size parameter d (i) It was modified according to traditional rules. J R (x) (i) ) is the Jacobian matrix of the antenna response. If or The algorithm will stop, prioritizing the condition that occurred first. User-defined parameters. It is the control variable of the algorithm.

[0142] As shown in Table 3, Table 3 is the default parameter table for the algorithm in this embodiment:

[0143] Table 3 Algorithm Default Parameters

[0144] In addition to the parameters related to the termination condition ( , , These parameters, besides allowing adjustment of the search process resolution, have only three control variables: N r N i and E max However, none of these parameters play a decisive role. On the one hand, changing the number of random observations in FGSA has no significant impact on the sensitivity analysis results because the effect of a specific parameter space orientation is averaged over the parameter space. On the other hand, N i and E maxIt is used only for rendering the initial alternative model, which is subsequently refined in the machine learning optimization loop. This means that the algorithm does not require additional parameter tuning for any specific problem.

[0145] In this embodiment, the entire algorithm flow pseudocode can be represented as global optimization algorithm pseudocode, including global sensitivity analysis, determining the model domain and initial model rendering, the global search phase of machine learning, and local parameter tuning. Global optimization involves iteratively generating candidate designs and refining alternative models using EM data collected during the search process.

[0146] The pseudocode for the global optimization algorithm includes the algorithm input, algorithm output, and algorithm flow, specifically:

[0147] Algorithm input:

[0148] objective function Design Space (interval) ,in and These are the lower and upper limits of the designable parameters and the control parameters. Termination threshold and .

[0149] Algorithm output:

[0150] Optimal solution .

[0151] Algorithm flow:

[0152] 1. Input parameters: objective function , , , and .

[0153] 2. Generate A randomized observable is measured and the FGSA procedure is performed to determine the domain dimension of the surrogate model. and span vector .

[0154] 3. Define the domain of the dimensionality reduction proxy model. .

[0155] 4. Construct the initial proxy model .

[0156] 5. Set the iteration counter .

[0157] 6. Use the PSO algorithm to solve the following optimization problem and obtain the fill points. :

[0158]

[0159] 7. Update the dataset:

[0160] .

[0161] 8. Update the dataset: Add an iteration counter .

[0162] 9. Build the proxy model using the updated dataset. .

[0163] 10. If any of the following conditions are met, proceed to step 11; otherwise, proceed to step 6:

[0164] (Parameter convergence)

[0165] exist The objective function value did not improve in the next iteration.

[0166] 11. In the original parameter space Perform local adjustments and solve the following optimization problem. The starting point is the vector found during the global search phase. 。:

[0167]

[0168] 12. Returning the optimal solution .

[0169] The technical solution of this application establishes an initial function model corresponding to the antenna based on antenna parameters and user requirements. By extracting features from the initial function model, the target features that have the greatest impact on antenna performance are determined. Based on these target features, the initial function model is dimensionality reduced, thereby optimizing the antenna parameters. This approach can significantly reduce computational costs and simulation time while maintaining high accuracy, thus improving the accuracy and efficiency of antenna parameter optimization.

[0170] See Figure 2 , Figure 2 This is a structural diagram of the antenna parameter optimization device provided in an embodiment of this application. Figure 2 As shown, the antenna parameter optimization device 200 includes:

[0171] Module 210 is used to establish an initial function model corresponding to the antenna based on antenna parameters and requirement information. The antenna parameters include the geometric parameters corresponding to the antenna, and the requirement information includes the optimization objective for optimizing the antenna parameters. The parameters used in the initial function model are derived from the multidimensional parameter space corresponding to the antenna.

[0172] Extraction module 220 is used to extract features from the initial function model to obtain target features, wherein the target features are the features that have the greatest impact on the performance of the antenna;

[0173] Processing module 230 is used to perform dimensionality reduction processing on the initial function model based on the target features to obtain the target function model;

[0174] The optimization module 240 is used to optimize the antenna parameters according to the objective function model to obtain the optimized antenna parameters.

[0175] Optionally, module 210 includes:

[0176] The parsing submodule is used to generate the target optimization function based on the input requirement information, and to parse the antenna to obtain the antenna parameters;

[0177] The first generation submodule is used to generate multiple constraints corresponding to the antenna parameters based on the requirement information. The multiple constraints are used to constrain different antenna attributes corresponding to the antenna. The antenna attributes include at least one of the following: the frequency range of the antenna, the gain range of the antenna, the in-band axis ratio of the antenna, and the size of the antenna.

[0178] The second generation submodule is used to generate an initial function model corresponding to the antenna based on the antenna parameters, the multiple constraints, and the target optimization function.

[0179] Optionally, the extraction module 220 includes:

[0180] The extraction submodule is used to perform feature extraction and simulation on the initial function model based on the Fast Global Sensitivity Analysis (FGSA) algorithm to obtain a response set. The response set includes multiple response samples, and each response sample corresponds to a feature point in the multidimensional parameter space.

[0181] The determination submodule is used to determine the target feature vector and the target feature value based on the plurality of response samples. The target feature vector is the feature vector corresponding to the response sample that has the greatest impact on the direction of the antenna among the plurality of response samples, and the target feature value is the impact value corresponding to the target feature vector.

[0182] The target features include the target feature vector and the target feature value.

[0183] Optionally, the submodules to be determined include:

[0184] The first determining unit is configured to determine the Euclidean distance between each first response sample and its corresponding neighbor sample among the plurality of response samples, thereby obtaining a plurality of distance values, wherein the plurality of distance values ​​correspond one-to-one with the plurality of response samples, wherein the neighbor sample corresponding to the first response sample is the response sample closest to the first response sample among the plurality of response samples, and the first response sample is any one of the response samples;

[0185] The calculation unit is used to calculate the normalized displacement vectors corresponding to the plurality of distance values ​​respectively, to obtain a plurality of displacement vectors that correspond one-to-one with the plurality of distance values, and to calculate the response changes corresponding to the plurality of distance values ​​respectively, to obtain a plurality of response change values ​​that correspond one-to-one with the plurality of distance values.

[0186] A generation unit is configured to generate a positioning matrix based on the plurality of displacement vectors and the plurality of response change values, wherein the i-th row of the positioning matrix is ​​the product of the i-th displacement vector and the i-th response change value;

[0187] The second determining unit is used to determine the target feature vector and the target feature value based on the positioning matrix.

[0188] Optionally, the second determining unit includes:

[0189] The decomposition subunit is used to perform eigenvalue decomposition on the positioning matrix to obtain multiple eigenvectors and multiple eigenvalues, wherein the multiple eigenvectors and the multiple eigenvalues ​​correspond one-to-one.

[0190] A determining subunit is configured to determine a target feature value among the plurality of feature values ​​and a target feature vector among the plurality of feature vectors, wherein the target feature value is the largest feature value among the plurality of feature values ​​and the target feature vector is a feature vector corresponding to the target feature value;

[0191] Processing module 230 includes:

[0192] The processing submodule is used to perform dimensionality reduction processing on the initial function model based on the target feature vector and the target feature value to obtain the target function model.

[0193] Optionally, optimization module 240 includes:

[0194] The third generation submodule is used to generate an initial Kriging proxy model based on the objective function model.

[0195] The simulation submodule is used to perform a global search and simulation on the initial Kriging proxy model based on the particle swarm optimization algorithm (PSO) to obtain the updated Kriging proxy model.

[0196] The fine-tuning submodule is used to locally fine-tune the updated Kriging agent model based on the Trust Domain Algorithm (TR) to obtain the target optimization strategy.

[0197] The optimization submodule is used to optimize the antenna parameters according to the target optimization strategy to obtain the optimized antenna parameters.

[0198] The technical solution of this application establishes an initial function model corresponding to the antenna based on antenna parameters and user requirements. By extracting features from the initial function model, the target features that have the greatest impact on antenna performance are determined. Based on these target features, the initial function model is dimensionality reduced, thereby optimizing the antenna parameters. This approach can significantly reduce computational costs and simulation time while maintaining high accuracy, thus improving the accuracy and efficiency of antenna parameter optimization.

[0199] This application also provides an electronic device. Please refer to [link to relevant documentation]. Figure 3 The electronic device may include a processor 301, a memory 302, and a program 3021 stored in the memory 302 and capable of running on the processor 301.

[0200] When program 3021 is executed by processor 301, it can achieve the following: Figure 1 Any step in the corresponding method embodiment:

[0201] An initial function model corresponding to the antenna is established based on the antenna parameters and the requirement information. The antenna parameters include the geometric parameters corresponding to the antenna, and the requirement information includes the optimization objective for optimizing the antenna parameters. The parameters used in the initial function model are derived from the multidimensional parameter space corresponding to the antenna.

[0202] Feature extraction is performed on the initial function model to obtain target features, which are the features that have the greatest impact on the performance of the antenna;

[0203] Based on the target features, the initial function model is reduced in dimensionality to obtain the target function model;

[0204] The antenna parameters are optimized based on the objective function model to obtain the optimized antenna parameters.

[0205] Optionally, establishing the initial function model corresponding to the antenna based on antenna parameters and requirement information includes:

[0206] Based on the input requirements, a target optimization function is generated, and the antenna is analyzed to obtain antenna parameters;

[0207] Based on the required information, multiple constraints corresponding to the antenna parameters are generated. These multiple constraints are used to constrain different antenna attributes corresponding to the antenna. The antenna attributes include at least one of the following: the frequency range of the antenna, the gain range of the antenna, the in-band axial ratio of the antenna, and the size of the antenna.

[0208] Based on the antenna parameters, the multiple constraints, and the objective optimization function, an initial function model corresponding to the antenna is generated.

[0209] Optionally, the step of extracting features from the initial function model to obtain target features includes:

[0210] Based on the Fast Global Sensitivity Analysis (FGSA) algorithm, feature extraction and simulation are performed on the initial function model to obtain a response set. The response set includes multiple response samples, and each response sample corresponds to a feature point in the multidimensional parameter space.

[0211] Based on the multiple response samples, a target feature vector and a target feature value are determined. The target feature vector is the feature vector corresponding to the response sample that has the greatest impact on the direction of the antenna among the multiple response samples, and the target feature value is the impact value corresponding to the target feature vector.

[0212] The target features include the target feature vector and the target feature value.

[0213] Optionally, determining the target feature vector and target feature value based on the plurality of response samples includes:

[0214] In the plurality of response samples, the Euclidean distance between each first response sample and its corresponding neighbor sample is determined to obtain a plurality of distance values, wherein the plurality of distance values ​​correspond one-to-one with the plurality of response samples, wherein the neighbor sample corresponding to the first response sample is: the response sample that is closest to the first response sample among the plurality of response samples, and the first response sample is any one of the response samples;

[0215] Calculate the normalized displacement vectors corresponding to the plurality of distance values ​​to obtain a plurality of displacement vectors that correspond one-to-one with the plurality of distance values; and calculate the response changes corresponding to the plurality of distance values ​​to obtain a plurality of response change values ​​that correspond one-to-one with the plurality of distance values.

[0216] A positioning matrix is ​​generated based on the plurality of displacement vectors and the plurality of response change values, wherein the i-th row of the positioning matrix is ​​the product of the i-th displacement vector and the i-th response change value;

[0217] Based on the positioning matrix, the target feature vector and the target feature value are determined.

[0218] Optionally, determining the target feature vector and the target feature value based on the positioning matrix includes:

[0219] The positioning matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvectors and multiple eigenvalues, and the multiple eigenvectors and multiple eigenvalues ​​correspond one-to-one.

[0220] A target feature value is determined among the plurality of feature values, and a target feature vector is determined among the plurality of feature vectors, wherein the target feature value is the largest feature value among the plurality of feature values, and the target feature vector is the feature vector corresponding to the target feature value;

[0221] The step of reducing the dimensionality of the initial function model based on the target features to obtain the target function model includes:

[0222] The initial function model is reduced in dimensionality based on the target feature vector and the target feature value to obtain the target function model.

[0223] Optionally, optimizing the antenna parameters according to the objective function model to obtain optimized antenna parameters includes:

[0224] Generate an initial Kriging proxy model based on the objective function model;

[0225] Based on the particle swarm optimization algorithm (PSO), a global search and simulation are performed on the initial Kriging proxy model to obtain an updated Kriging proxy model.

[0226] The updated Kriging agent model is locally fine-tuned based on the Trust Domain Algorithm (TR) to obtain the target optimization strategy.

[0227] The antenna parameters are optimized according to the target optimization strategy to obtain the optimized antenna parameters.

[0228] The technical solution of this application establishes an initial function model corresponding to the antenna based on antenna parameters and user requirements. By extracting features from the initial function model, the target features that have the greatest impact on antenna performance are determined. Based on these target features, the initial function model is dimensionality reduced, thereby optimizing the antenna parameters. This approach can significantly reduce computational costs and simulation time while maintaining high accuracy, thus improving the accuracy and efficiency of antenna parameter optimization.

[0229] This application also provides a computer-readable storage medium storing a computer program. When executed by a processor, this computer program implements the various processes of the antenna parameter optimization embodiments described above and achieves the same technical effects. To avoid repetition, it will not be described again here. The computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk, etc.

[0230] This application also provides a computer program product, which is stored in a storage medium and executed by at least one processor to implement the various processes of the antenna parameter optimization method embodiments described above, and can achieve the same technical effect. To avoid repetition, it will not be described again here.

[0231] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes that element.

[0232] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a communication device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in the various embodiments of this application.

[0233] The embodiments of this application have been described above with reference to the accompanying drawings. However, this application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for optimizing antenna parameters, characterized in that, The method includes: An initial function model corresponding to the antenna is established based on the antenna parameters and the requirement information. The antenna parameters include the geometric parameters corresponding to the antenna, and the requirement information includes the optimization objective for optimizing the antenna parameters. The parameters used in the initial function model are derived from the multidimensional parameter space corresponding to the antenna. Feature extraction is performed on the initial function model to obtain target features, which are the features that have the greatest impact on the performance of the antenna; Based on the target features, the initial function model is reduced in dimensionality to obtain the target function model; The antenna parameters are optimized based on the objective function model to obtain the optimized antenna parameters.

2. The method according to claim 1, characterized in that, The step of establishing the initial function model corresponding to the antenna based on antenna parameters and requirement information includes: Based on the input requirements, a target optimization function is generated, and the antenna is analyzed to obtain antenna parameters; Based on the required information, multiple constraints corresponding to the antenna parameters are generated. These multiple constraints are used to constrain different antenna attributes corresponding to the antenna. The antenna attributes include at least one of the following: the frequency range of the antenna, the gain range of the antenna, the in-band axial ratio of the antenna, and the size of the antenna. Based on the antenna parameters, the multiple constraints, and the objective optimization function, an initial function model corresponding to the antenna is generated.

3. The method according to claim 1, characterized in that, The step of extracting features from the initial function model to obtain target features includes: Based on the Fast Global Sensitivity Analysis (FGSA) algorithm, feature extraction and simulation are performed on the initial function model to obtain a response set. The response set includes multiple response samples, and each response sample corresponds to a feature point in the multidimensional parameter space. Based on the multiple response samples, a target feature vector and a target feature value are determined. The target feature vector is the feature vector corresponding to the response sample that has the greatest impact on the direction of the antenna among the multiple response samples, and the target feature value is the impact value corresponding to the target feature vector. The target features include the target feature vector and the target feature value.

4. The method according to claim 3, characterized in that, The step of determining the target feature vector and target feature value based on the plurality of response samples includes: In the plurality of response samples, the Euclidean distance between each first response sample and its corresponding neighbor sample is determined to obtain a plurality of distance values, wherein the plurality of distance values ​​correspond one-to-one with the plurality of response samples, wherein the neighbor sample corresponding to the first response sample is: the response sample that is closest to the first response sample among the plurality of response samples, and the first response sample is any one of the response samples; Calculate the normalized displacement vectors corresponding to the plurality of distance values ​​to obtain a plurality of displacement vectors that correspond one-to-one with the plurality of distance values; and calculate the response changes corresponding to the plurality of distance values ​​to obtain a plurality of response change values ​​that correspond one-to-one with the plurality of distance values. A positioning matrix is ​​generated based on the plurality of displacement vectors and the plurality of response change values, wherein the i-th row of the positioning matrix is ​​the product of the i-th displacement vector and the i-th response change value; Based on the positioning matrix, the target feature vector and the target feature value are determined.

5. The method according to claim 4, characterized in that... The step of determining the target feature vector and the target feature value based on the positioning matrix includes: The positioning matrix is ​​decomposed into eigenvalues ​​to obtain multiple eigenvectors and multiple eigenvalues, and the multiple eigenvectors and multiple eigenvalues ​​correspond one-to-one. A target feature value is determined among the plurality of feature values, and a target feature vector is determined among the plurality of feature vectors, wherein the target feature value is the largest feature value among the plurality of feature values, and the target feature vector is the feature vector corresponding to the target feature value; The step of reducing the dimensionality of the initial function model based on the target features to obtain the target function model includes: The initial function model is reduced in dimensionality based on the target feature vector and the target feature value to obtain the target function model.

6. The method according to claim 1, characterized in that, The step of optimizing the antenna parameters according to the objective function model to obtain the optimized antenna parameters includes: Generate an initial Kriging proxy model based on the objective function model; Based on the particle swarm optimization algorithm (PSO), a global search and simulation are performed on the initial Kriging proxy model to obtain an updated Kriging proxy model. The updated Kriging agent model is locally fine-tuned based on the Trust Domain Algorithm (TR) to obtain the target optimization strategy. The antenna parameters are optimized according to the target optimization strategy to obtain the optimized antenna parameters.

7. An antenna parameter optimization device, characterized in that, The device includes: A module is established to build an initial function model corresponding to the antenna based on the antenna parameters and requirement information. The antenna parameters include the geometric parameters corresponding to the antenna, and the requirement information includes the optimization objective for optimizing the antenna parameters. The parameters used in the initial function model are derived from the multidimensional parameter space corresponding to the antenna. The extraction module is used to extract features from the initial function model to obtain target features, wherein the target features are the features that have the greatest impact on the performance of the antenna; The processing module is used to perform dimensionality reduction processing on the initial function model based on the target features to obtain the target function model; The optimization module is used to optimize the antenna parameters according to the objective function model to obtain the optimized antenna parameters.

8. An electronic device, characterized in that, include: A processor, a memory, and a program stored in the memory and executable on the processor, wherein the program, when executed by the processor, implements the steps of the method as described in any one of claims 1 to 6.

9. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the steps of the method as described in any one of claims 1 to 6.

10. A computer program product, characterized in that, Includes computer instructions that, when executed by a processor, implement the steps of the method as described in any one of claims 1 to 6.

Images