Agent model sample screening method and system and microwave device

Abstract

The invention discloses a sample screening method and system for an agent model and a microwave device, and the method comprises the steps: sampling and obtaining an initial design point, screening the design point through a fuzzy clustering algorithm, and storing the design point in a database; judging whether the database meets a preset stop condition or not, if yes, outputting the optimal design, and if not, turning to the next step; using the fuzzy clustering algorithm again to simplify the database; thirdly, designing a plurality of design targets, respectively constructing a proxy model for each design target, and optimizing prediction functions obtained by the plurality of proxy models constructed in the third step by using a multi-target optimization algorithm to obtain a target solution set; performing full-wave simulation on the target solution set, judging whether corresponding design points in the target solution set meet design requirements or not, if so, outputting the target solution, and otherwise, turning to the next step; performing candidate design point selection on the target solution in the target solution set by adopting a parallel point adding criterion, performing full-wave simulation, adding a real response value into the database, and returning to the step 2, thereby reducing the design cost of the microwave device.

Description

Technical Field

[0001] The present invention relates to the technical field of microwave device design, and in particular to a sample screening method and system of an agent model and a microwave device. Background Art

[0002] With the development of wireless communication technology and the increasingly stringent performance standards of microwave devices, high-dimensional design space, multi-objective optimization and complex and difficult-to-solve objective functions are involved. They are difficult to optimize properly, which poses new challenges to the design of high-performance microwave devices with compact size, multi-band and broadband characteristics.

[0003] At present, the engineering community mainly relies on powerful computer simulation software such as CST and HFSS for optimization design. Such microwave device design cycle is long, and such conventional design methods are powerless for antenna design with irregular structures, new structures and high performance requirements. Moreover, when optimizing microwave device structures involving multiple parameters and multiple objectives, the various parameters affect each other, the design process is lengthy, the optimization capability and efficiency become very poor, and it is difficult to obtain a better solution by relying solely on full-wave simulation.

[0004] The proxy model technology can effectively solve the above problems and has become a very popular and effective way to replace expensive numerical simulations. As a key step in building a proxy model, sample selection is crucial. The selected sample points largely determine the accuracy level of the proxy model.

[0005] In the prior art, the invention patent with the patent publication number CN115081107A discloses a method and device for constructing a Kriging approximate model of an automobile body based on the parallel point addition criterion, wherein the method includes generating an initial sample set of design variables according to the modeling target; calculating the function response value of each sample point; constructing an initial Kriging approximate model of the automobile body using the initial sample set; judging whether the model meets the optimization requirements according to the accuracy evaluation criterion; if not, adding new sample points according to the EI and improved EI point addition criteria at the same time, and adding the newly added sample points to the initial sample set, deleting redundant sample points, and re-establishing the Kriging approximate model; then judging the accuracy of the new approximate model, if it does not meet the requirements, continue to add sample points according to the parallel point addition criterion until the final optimized automobile body Kriging approximate model is obtained. The patent adds sample points according to the EI point addition criterion and the improved EI point addition criterion, and then deletes redundant sample points, including duplicate points between the newly added sample points and duplicate points between the newly added points and the original sample points, and only deletes the duplication in the data. However, some samples have high validity and some have low validity, and the patent cannot distinguish them. Summary of the invention

[0006] The technical problem to be solved by the present invention is to provide a multi-objective microwave device design method which can effectively screen sample sets, greatly reduce the design cost of microwave devices, and has a short design cycle and high efficiency.

[0007] In order to solve the above technical problems, the present invention provides the following technical solutions:

[0008] A sample screening method for a proxy model, comprising:

[0009] S10, sampling to obtain initial design points, and using fuzzy clustering algorithm to screen the design points and save them in the database;

[0010] S20, determining whether the database meets the preset stop condition, if so, outputting the optimal design, otherwise turning to step S30;

[0011] S30, again using the fuzzy clustering algorithm to simplify the database; based on the simplified database, constructing a proxy model for each design target;

[0012] S40, optimizing the prediction functions obtained by the multiple proxy models constructed in step S30 using a multi-objective optimization algorithm to obtain a target solution set;

[0013] S50, performing full-wave simulation on the target solution set, and determining whether the corresponding design point in the target solution set meets the design requirements, if so, outputting the target solution, otherwise turning to step S60;

[0014] S60, selecting candidate design points for the target solutions in the target solution set by using the parallel point addition criterion, performing full-wave simulation on the candidate design points, adding the true response values ​​to the database, and returning to step S20.

[0015] Advantages: The present invention selects validity data through fuzzy clustering and mines the deep-level properties of samples. In order to improve the prediction accuracy of the proxy model, a parallel adding strategy is used to add points.

[0016] In one embodiment of the present invention, in step S10, the fuzzy clustering algorithm selects the design points and saves them in the database, including:

[0017] S11, determine the sample design point;

[0018] S12, randomly select n c Cluster centers and set fuzzy parameters;

[0019] S13, for each sample design point, the membership degree to each cluster center is calculated according to its distance from the cluster center, and the value range is between 0 and 1;

[0020] S14, recalculating each cluster center according to the size of the membership degree to update the cluster center;

[0021] S15, repeating steps S13 and S14 until a stop condition is met;

[0022] S16, dividing the design points into clusters according to the final membership matrix and cluster centers obtained after the stop condition is met;

[0023] S17, selecting the center point of each cluster center obtained in step S16, and the q sample design points farthest from the center point in each cluster as the final initial design points;

[0024] S18, performing full-wave simulation on the initial design point, and saving the initial design point and the corresponding full-wave simulation into a database.

[0025] In one embodiment of the present invention, the fuzzy clustering algorithm is the following formula:

[0026]

[0027] In the formula, J represents the fuzzy clustering algorithm, n represents the number of sample design points, and u ij Expressed as membership degree, x i Represents the i-th sample design point, c j Represented as the jth cluster center, n c It is represented as the number of cluster centers and m is represented as the fuzzy parameter.

[0028] In one embodiment of the present invention, the initial design point X is obtained by the following formula:

[0029] X=Top q (d(x i ,c j ));

[0030] In the formula, d(x i ,c j ) is represented as the design point x i and the center point c of the cluster center j The distance between.

[0031] In one embodiment of the present invention, in step S30, the fuzzy clustering algorithm is used to simplify the database, including:

[0032] The initial design points in the database are adaptively divided into different regions through fuzzy clustering algorithm;

[0033] In each region, the average Euclidean distance between each initial design point and all other initial design points is calculated, and the initial design point with the shortest average distance to all sample points in each region is screened out and recorded as the pseudo-central design point. In each corresponding region, the pseudo-central design point is removed.

[0034] In one embodiment of the present invention, the pseudo center design point xc is obtained by the following formula:

[0035]

[0036] In the formula, d(x i ,x j ) represents the Euclidean distance between the initial design point and all other initial design points, n represents the number of design points, x i 、x j Represented as different design points.

[0037] In one embodiment of the present invention, in step S60, the target solution in the target solution set is selected using a parallel point addition criterion for candidate design points, including:

[0038] S61, optimizing the prediction functions obtained by the multiple proxy models using a multi-objective optimization algorithm, and selecting the design point with the minimum prediction value for each design objective;

[0039] S62, optimizing the prediction variance functions obtained by the multiple proxy models using a multi-objective optimization algorithm, and selecting the design point with the largest prediction variance for each design objective;

[0040] S63, using the K-fold cross-validation algorithm to calculate the cross-validation error of each design point in the streamlined database; then using the Monte Carlo sampling method to generate random sample points, and using the fuzzy clustering algorithm in step S30 to adaptively divide the random sample points into different areas to assign them to different areas; respectively, in the area where the design point with the largest cross-validation error corresponding to each target is located, select the design point farthest from the center point of the area.

[0041] The present invention also provides a system for a sample screening method of a proxy model, comprising:

[0042] The initial database module is used to sample and obtain the initial design points, and use the fuzzy clustering algorithm to screen the design points and save them in the database;

[0043] The first judgment module is used to judge whether the database meets the preset stop condition. If so, the optimal design is output; otherwise, the optimal design is turned to the database simplification and proxy model module;

[0044] The database reduction and proxy model module is used to reduce the database again using the fuzzy clustering algorithm; based on the reduced database, a proxy model is constructed for each design target;

[0045] A multi-objective optimization module is used to optimize the prediction functions obtained by constructing multiple proxy models using a multi-objective optimization algorithm to obtain a target solution set;

[0046] The second judgment module is used to perform full-wave simulation on the target solution set and judge whether the corresponding design point in the target solution set meets the design requirements. If so, the target solution is output, otherwise it turns to the parallel point adding module;

[0047] The parallel adding point module is used to select candidate design points for the target solution in the target solution set using the parallel adding point criterion, then perform full-wave simulation on the candidate design points, add the true response values ​​to the database, and return to the first judgment module.

[0048] The present invention also provides a microwave device, and the sample screening method of the above-mentioned proxy model is used to screen the design points of the microwave device.

[0049] Compared with the prior art, the beneficial effects of the present invention are as follows: the present invention screens the initial samples through fuzzy classification, fully reflects the regional characteristics of the sample set and reduces the risk of overfitting. In addition, by deleting redundant sample points during the proxy model update process, unnecessary calculations can be reduced, thereby accelerating the optimization process. The adaptability of fuzzy clustering enables it to dynamically adjust the cluster center according to the distance relationship between samples, thereby better adapting to the complexity and diversity in microwave device design. This adaptability helps to improve the robustness and reliability of the proxy model. Moreover, the method itself is completely independent of the proxy model technology, so it can be combined with various types of proxy models. At the same time, the method adopts parallel point addition, which can flexibly adjust the optimization strategy and dynamically update the proxy model according to different design requirements, thereby replacing the electromagnetic simulation process to estimate the desired function value, avoiding the direct electromagnetic simulation and precise calculation of all search samples in the global optimization design method, so as to achieve better optimization effect. The method has good versatility and practicality, and can effectively improve the accuracy and efficiency of microwave device design. This method not only reduces the computational cost, but also improves the reliability and stability of the design process. BRIEF DESCRIPTION OF THE DRAWINGS

[0050] Figure 1 The present invention is a flowchart of a sample screening method of a proxy model according to an embodiment of the present invention.

[0051] Figure 2 The present invention is a block diagram of a sample screening system of an agent model according to an embodiment of the present invention.

[0052] Figure 3Schematic diagram of a microwave device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0053] In order to facilitate those skilled in the art to understand the technical solution of the present invention, the technical solution of the present invention is further described in conjunction with the accompanying drawings of the specification.

[0054] The terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of the features. In the description of this application, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0055] See also Figure 1 As shown, the present invention provides a sample screening method of a proxy model, comprising:

[0056] S10, sampling to obtain initial design points, and using fuzzy clustering algorithm to screen the design points and save them in the database.

[0057] In one embodiment of the present invention, a large number of random points are generated, and design point sampling is performed based on Latin hypercube sampling to generate 10 x k sample points, where k is the dimension of the design variable.

[0058] In this embodiment, the samples are adaptively divided into different areas based on the initial design points obtained by sampling as sample information through a fuzzy clustering algorithm. The membership of each sample point to all cluster centers is calculated, and then the cluster center to which the sample point belongs is determined according to the size of the membership. By introducing fuzzy parameters, the data point can belong to multiple cluster centers, thereby better handling the uncertainty of the data.

[0059] In this embodiment, the goal of the fuzzy clustering algorithm is to minimize all sample points x i To its cluster center c j The sum of the weighted distances of each sample point and the membership u of all cluster centers ij The sum is 1, see formula (1).

[0060]

[0061] Where n is the number of sample design points, u ij Expressed as membership degree, x i Represents the i-th sample design point, c j It is represented as the jth cluster center, and J represents the fuzzy clustering algorithm.

[0062] The specific implementation steps are as follows:

[0063] S11, determine the sample in advance: determine the sample design point.

[0064] S12, initialization: randomly select n c cluster centers and set the fuzzy parameter m.

[0065] S13, calculating the membership degree: for each sample design point, the membership degree to each cluster center is calculated according to its distance from the cluster center, and the value range is between 0 and 1. Specifically, the Euclidean distance from a sample point to each cluster center is calculated, and the sum of these distances and the weighted result is the membership degree.

[0066] S14, updating cluster centers: recalculating each cluster center according to the size of the membership degree to update the cluster center;

[0067] S15, iterative optimization: repeat steps S13 and S14 until a stop condition is met. In this embodiment, the stop condition is met such as reaching a maximum number of iterations or the objective function converges or the change in the cluster center is less than a preset threshold.

[0068] S16, result output: divide the design points into clusters according to the final membership matrix and cluster centers obtained after the stop condition is met.

[0069] S17, selecting initial design points: selecting the center point of each cluster center obtained in step S16, and q sample design points in each cluster that are farthest from the center point as the final initial design points.

[0070] In this embodiment, the initial design point X is obtained by formula (2):

[0071] X=Top q (d(x i ,c j )), (2);

[0072] In the formula, d(x i ,c j ) is represented as the design point x i and the center point c of the cluster center j The distance between the design points x i and the center point c of the cluster center j The distance between them can be obtained by formula (3):

[0073]

[0074] S18, performing full-wave simulation on the initial design point, and saving the initial design point and the corresponding full-wave simulation into a database.

[0075] S20, determining whether the database satisfies a preset stop condition, if so, outputting the optimal design, otherwise turning to step S30.

[0076] In this embodiment, the preset stop condition is, for example, that the design goal meets the requirements or the maximum number of simulation times is reached.

[0077] S30, again using the fuzzy clustering algorithm to streamline the database; and constructing a proxy model for each design target based on the streamlined database.

[0078] In one embodiment of the present invention, the steps of simplifying the database include:

[0079] S31, adaptively dividing the initial design points in the database into different regions through a fuzzy clustering algorithm.

[0080] In this embodiment, the principle of clustering sample design points is the same, and steps S11 to S16 can be performed. The difference is that the samples prepared in advance in step S11 are the initial design points in the database. After dividing different regions, redundant samples are identified and removed in each region, and the sample set after screening can fully reflect the regional characteristics and reduce the risk of overfitting.

[0081] S32, in each region, calculate the average Euclidean distance between each initial design point and all other initial design points, screen out the initial design point in each region with the shortest average distance to all sample points, record it as the pseudo central design point, and remove the pseudo central design point in each corresponding region.

[0082] In this embodiment, the pseudo-central design point can be understood as the regional position represented by the design point, which can be represented by other design points and is removed to reduce redundancy. The pseudo-central design point xc is obtained by the following formula (4):

[0083]

[0084] In the formula, d(x i ,x j ) represents the Euclidean distance between the initial design point and all other initial design points, n represents the number of design points, x i 、x j Represented as different design points.

[0085] In one embodiment of the present invention, a proxy model is constructed for each design target based on the streamlined database, where the proxy model may be a Gaussian process regression model, a neural network proxy model, or the like.

[0086] S40, optimizing the prediction functions obtained by the multiple proxy models constructed in step S30 using a multi-objective optimization algorithm to obtain a target solution set.

[0087] In this embodiment, a multi-objective optimization algorithm is used to perform an optimization operation on the prediction function obtained by the multiple proxy models constructed in step S30, obtain the Pareto frontier, and screen the solution set. The multi-objective optimization algorithm here can be a multi-objective genetic NSGA-II algorithm, a particle swarm algorithm, etc. Among them, for multi-objective optimization problems, a single solution that makes the global optimal cannot be obtained, but a solution set, and the Pareto frontier and the Pareto optimal solution set correspond to the set of target values.

[0088] S50, perform full-wave simulation on the target solution set, and determine whether the corresponding design point in the target solution set meets the design requirements. If yes, output the target solution, otherwise go to step S60.

[0089] In this embodiment, the processing method for each target is: by taking the maximum value within the frequency band, adding or subtracting the values ​​corresponding to different angles at different frequencies, etc., the final ideal target is returned to 0 or below 0, that is, the ideal target vector for stopping the optimization is [≤0,≤0,...,≤0] 1 x p , where p is the number of design goals. So we choose a distance 1 xp dimension vector [≤0,≤0,...,≤0] 1x p If the solution corresponding to the point on the nearest Pareto front meets the design requirements, output the solution; otherwise, go to step S60.

[0090] S60, selecting candidate design points for the target solutions in the target solution set by using the parallel point addition criterion, performing full-wave simulation on the candidate design points, adding the true response values ​​to the database, and returning to step S20.

[0091] In one embodiment of the present invention, the parallel point addition criterion is used to select candidate points, and (3x p+1) sample points are generated in one optimization. The point addition process will be described in detail below, where the "1" in 3x p+1 is the distance 1 xp dimension vector [0,0,...,0] selected in step S40. 1 x p The sample points on the nearest Pareto frontier are as follows:

[0092] S61, using a multi-objective optimization algorithm to optimize the prediction functions obtained by multiple proxy models, and select the design point with the smallest prediction value for each design objective.

[0093] x new1 =argmin(F(x));

[0094] In the formula, x new1It is represented as the design point with the smallest predicted value of the design target, and F(x) is represented as the prediction function obtained by the surrogate model. In this step, there are p design points with the smallest predicted value of the design target.

[0095] S62, using a multi-objective optimization algorithm to optimize the prediction variance functions obtained by multiple proxy models, and select the design point with the largest prediction variance for each design objective.

[0096] x new2 = argmax(s 2 (x));

[0097] In the formula, x new2 It represents the design point with the largest prediction variance of the design target, s 2 (x) represents the prediction variance function obtained by the surrogate model. In this step, there are p design points with the largest prediction variance of the design target. In addition, the surrogate model needs to be able to provide the prediction variance.

[0098] S63, using the K-fold cross-validation algorithm to calculate the cross-validation error of each design point in the streamlined database; then using the Monte Carlo sampling method to generate random sample points, and using the fuzzy clustering algorithm in step S30 to adaptively divide the random sample points into different areas to assign them to different areas; respectively, in the area where the design point with the largest cross-validation error corresponding to each target is located, select the design point farthest from the center point of the area.

[0099] x new3 = argmax(d(x,xc)),x∈Rc;

[0100] In the formula, x new3 It represents the design point farthest from the center of the region, there are p of them in total, d(x,xc) represents the distance from the random sample point to the center of the region, and Rc represents the region where the random sample point is located.

[0101] In one embodiment of the present invention, the Monte Carlo sampling method is used to generate random sample points, and the target response calculation is not performed on these random sample points. The random sample points are assigned to different regions by adaptively dividing them into different regions using the fuzzy clustering algorithm in step S30, and the division principle is the same as that of the cluster sample design point.

[0102] In this embodiment, new sample points are obtained and true response values ​​are calculated according to the three point adding criteria, added to the database, and the process returns to step S20.

[0103] See also Figure 1 and Figure 2 As shown, the present invention provides a system for a sample screening method of a proxy model, comprising:

[0104] The initial database module 10 is used to sample and obtain initial design points, and to select the design points using a fuzzy clustering algorithm and save them in a database.

[0105] The first judgment module 20 is used to judge whether the database meets the preset stop condition. If so, the optimal design is output; otherwise, the optimal design is turned to the database reduction and proxy model module.

[0106] The database simplification and proxy model module 30 is used to simplify the database again using the fuzzy clustering algorithm; and to construct a proxy model for each design target based on the simplified database.

[0107] The multi-objective optimization module 40 is used to optimize the prediction functions obtained by constructing multiple proxy models using a multi-objective optimization algorithm to obtain a target solution set.

[0108] The second judgment module 50 is used to perform full-wave simulation on the target solution set and judge whether the corresponding design points in the target solution set meet the design requirements. If yes, the target solution is output, otherwise it turns to the parallel point adding module.

[0109] The parallel adding module 60 is used to select candidate design points for the target solution in the target solution set using the parallel adding criterion, perform full-wave simulation on the candidate design points, add the true response values ​​to the database, and return to the first judgment module.

[0110] See also Figures 1 to 3 As shown, the present invention also provides a microwave device, which uses the sample screening method of the proxy model described above to screen the design points of the microwave device to assist in the design of the microwave device. In this embodiment, the microwave device is a corrugated horn.

[0111] In this embodiment, the dimension of the corrugated horn design variables is 23, and the optimized design goals are standing waves less than 1.2 in the range of 6 to 8 GHz, illumination levels of plus or minus 36 degrees, cross polarization less than -30 dB, and beam equalization less than 2 dB within the illumination angle. Here, each target is processed in the following way: the final ideal target is returned to 0 or below 0 by taking the maximum value within the frequency band and adding or subtracting the values ​​corresponding to different angles at different frequencies. For the corrugated horn, 5 design goals are set, namely p = 5, corresponding to standing waves in the range of 6 to 8 GHz, the level value in the horizontal direction (Phi = 0°), the level value in the vertical direction (Phi = 90°), cross polarization and beam equalization, and the ideal target vector for optimization stop is [≤0, ≤0, ≤0, ≤0, ≤0].

[0112] The specific implementation steps are as follows:

[0113] Step 1: Perform initial design point sampling based on Latin hypercube to generate 10 x 23 sample points; use the fuzzy clustering algorithm to select the center point of each cluster and the 8 samples farthest from the center point as the final initial design points, a total of 54 design points, calculate the true response to obtain the initial database.

[0114] Step 2: Determine whether the database meets the preset stop conditions, specifically whether there is a design goal that meets the requirements or the maximum number of simulations is reached. If so, output the optimal solution in the database, otherwise go to step 3.

[0115] Step 3: Use the fuzzy clustering algorithm to filter out the pseudo-center samples in each area with the closest average distance to all samples and remove them. Based on the 48 sample points after screening, Gaussian process regression proxy models are constructed for the above-mentioned five design objectives such as standing waves.

[0116] Step 4: Use the multi-objective genetic algorithm NSGA-II to optimize the prediction function obtained from the five Gaussian process regression models constructed in step 3.

[0117] Step 5: After obtaining the Pareto frontier, screen the solution set and select the solution corresponding to the point on the Pareto frontier closest to the sample point [0,0,0,0,0] for full-wave simulation. If the preset stopping condition is met (there is a design goal that meets the requirements), output the solution; otherwise, go to step 6.

[0118] Step 6: Use the parallel addition criterion to add 15+1 sample points, where "1" is the sample point on the Pareto front closest to [0,0,0,0,0] selected in step 5.

[0119] Ⅰ. Use the NSGA-II algorithm to optimize the prediction function obtained by the five Gaussian process regression models constructed in step 3, and select five sample points with the smallest prediction value for each design target.

[0120] Ⅱ. Use the NSGA-II algorithm to optimize the prediction variance function obtained by the five Gaussian process regression models constructed in step 3, and select five sample points with the largest prediction variance for each design target.

[0121] III. Using the K-CV algorithm and Monte Carlo sampling method, select 5 sample points farthest from the center of each target cross-validation error area. Here, K is set to 5.

[0122] After denormalizing the selected new sample points, perform full-wave simulation, add them to the database, and return to step 2.

[0123] Finally, the optimization process was repeated 52 times. After adding points 480 times on the basis of the initial design point, the design target that met the requirements was found: [-0.05, -0.297, -0.452, 0, -0.596]. At this time, the standing wave, illumination level, cross polarization and beam equalization in the range of 6 to 8 GHz all met the expected goals, as shown in Table 1.

[0124] Table 1 Comparison between the actual optimization results of the corrugated speaker and the optimization target

[0125]

[0126]

[0127] See also Figures 1 to 3 As shown in the figure, when the surrogate model sample screening method based on fuzzy clustering assists the corrugated horn design, the initial design point and the optimized added point are subjected to a total of 54+480 CST full-wave simulations; without direct surrogate model assisted corrugated horn design without screening, the initial design point and the optimized added point are subjected to a total of 230+656 CST full-wave simulations. The surrogate model sample screening method based on fuzzy clustering selects the initial samples through fuzzy classification, fully reflects the regional characteristics of the sample set, and reduces the risk of overfitting. In addition, redundant sample points are deleted during the surrogate model update process, which reduces unnecessary calculations and greatly accelerates the optimization process.

[0128] It is obvious to those skilled in the art that the present invention is not limited to the details of the exemplary embodiments described above, and that the present invention can be implemented in other specific forms without departing from the spirit or essential features of the present invention. Therefore, the embodiments should be regarded as exemplary and non-limiting from any point of view, and the scope of the present invention is defined by the appended claims rather than the above description, and it is intended that all changes falling within the meaning and scope of the equivalent elements of the claims are included in the present invention, and any reference numerals in the claims should not be regarded as limiting the claims involved.

[0129] The above-described embodiments merely represent implementation methods of the invention. The protection scope of the present invention is not limited to the above-described embodiments. For those skilled in the art, several modifications and improvements may be made without departing from the concept of the present invention, which all belong to the protection scope of the present invention.

Claims

1. A sample screening method for a proxy model, characterized in that: include: S10, sampling to obtain initial design points, and using fuzzy clustering algorithm to screen the design points and save them in the database; S20, determining whether the database meets the preset stop condition, if so, outputting the optimal design, otherwise turning to step S30; S30, again using the fuzzy clustering algorithm to simplify the database; based on the simplified database, constructing a proxy model for each design target; S40, optimizing the prediction functions obtained by the multiple proxy models constructed in step S30 using a multi-objective optimization algorithm to obtain a target solution set; S50, performing full-wave simulation on the target solution set, and determining whether the corresponding design point in the target solution set meets the design requirements, if so, outputting the target solution, otherwise turning to step S60; S60, selecting candidate design points for the target solutions in the target solution set by using the parallel point addition criterion, performing full-wave simulation on the candidate design points, adding the true response values ​​to the database, and returning to step S20.

2. The sample screening method of the proxy model according to claim 1, characterized in that: In step S10, the fuzzy clustering algorithm selects the design points and saves them in the database, including: S11, determine the sample design point; S12, randomly select n c Cluster centers and set fuzzy parameters; S13, for each sample design point, the membership degree to each cluster center is calculated according to its distance from the cluster center, and the value range is between 0 and 1; S14, recalculating each cluster center according to the size of the membership degree to update the cluster center; S15, repeating steps S13 and S14 until a stop condition is met; S16, dividing the design points into clusters according to the final membership matrix and cluster centers obtained after the stop condition is met; S17, selecting the center point of each cluster center obtained in step S16, and the q sample design points farthest from the center point in each cluster as the final initial design points; S18, performing full-wave simulation on the initial design point, and saving the initial design point and the corresponding full-wave simulation into a database.

3. The sample screening method of the proxy model according to claim 1, characterized in that: The fuzzy clustering algorithm is as follows: In the formula, J represents the fuzzy clustering algorithm, n represents the number of sample design points, and u ij is represented by the degree of membership, xi represents the i-th sample design point, c j Represented as the jth cluster center, n c It is represented as the number of cluster centers and m is represented as the fuzzy parameter.

4. The sample screening method of the proxy model according to claim 2, characterized in that: The initial design point X is obtained by the following formula: X=Top q (d(x i ,c j )); In the formula, d(x i ,c j ) is represented as the design point x i and the center point c of the cluster center j The distance between.

5. The sample screening method of the proxy model according to claim 1, characterized in that: In step S30, the database is simplified using a fuzzy clustering algorithm, including: The initial design points in the database are adaptively divided into different regions through fuzzy clustering algorithm; In each region, the average Euclidean distance between each initial design point and all other initial design points is calculated, and the initial design point with the shortest average distance to all sample points in each region is screened out and recorded as the pseudo-central design point. In each corresponding region, the pseudo-central design point is removed.

6. The sample screening method of the proxy model according to claim 5, characterized in that: The pseudo central design point xc is obtained by the following formula: In the formula, d(x i ,x j ) represents the Euclidean distance between the initial design point and all other initial design points, n represents the number of design points, x i 、x j Represented as different design points.

7. The sample screening method of the proxy model according to claim 1, characterized in that: In step S60, the target solution in the target solution set is selected using a parallel point addition criterion for candidate design points, including: S61, optimizing the prediction functions obtained by the multiple proxy models using a multi-objective optimization algorithm, and selecting the design point with the minimum prediction value for each design objective; S62, optimizing the prediction variance functions obtained by the multiple proxy models using a multi-objective optimization algorithm, and selecting the design point with the largest prediction variance for each design objective; S63, using the K-fold cross-validation algorithm to calculate the cross-validation error of each design point in the streamlined database; then using the Monte Carlo sampling method to generate random sample points, and using the fuzzy clustering algorithm in step S30 to adaptively divide the random sample points into different areas to assign them to different areas; respectively, in the area where the design point with the largest cross-validation error corresponding to each target is located, select the design point farthest from the center point of the area.

8. A system for the sample screening method of the proxy model according to any one of claims 1 to 7, characterized in that: include: The initial database module is used to sample and obtain the initial design points, and use the fuzzy clustering algorithm to screen the design points and save them in the database; The first judgment module is used to judge whether the database meets the preset stop condition. If so, the optimal design is output; otherwise, the optimal design is turned to the database simplification and proxy model module; The database reduction and proxy model module is used to reduce the database again using the fuzzy clustering algorithm; based on the reduced database, a proxy model is constructed for each design target; A multi-objective optimization module is used to optimize the prediction functions obtained by constructing multiple proxy models using a multi-objective optimization algorithm to obtain a target solution set; The second judgment module is used to perform full-wave simulation on the target solution set and judge whether the corresponding design point in the target solution set meets the design requirements. If so, the target solution is output, otherwise it turns to the parallel point adding module; The parallel adding point module is used to select candidate design points for the target solution in the target solution set using the parallel adding point criterion, then perform full-wave simulation on the candidate design points, add the true response values ​​to the database, and return to the first judgment module.

9. A microwave device, characterized in that: The sample screening method of the proxy model described in any one of claims 1 to 7 is used to screen the design points of microwave devices.

Images