An online automatic design method, device and medium for complex high-dimensional electromagnetic structures

Abstract

The application discloses an online automatic design method and device for complex high-dimensional electromagnetic structure and a medium, wherein the method comprises the following steps: modeling the electromagnetic structure as a binary variable matrix, setting an optimization objective function, and constructing an initial data set; training an agent model by using the initial data set to obtain an initial agent model; performing sensitivity analysis on variables by using Sobol indexes; extracting key points according to contribution proportions and performing optimization training on the agent model; determining a starting solution and parameters of a simulated annealing algorithm, and iteratively optimizing by generating a neighborhood solution, calculating a target value and adopting a probability acceptance strategy; performing simulation verification and updating the online agent model; re-finding the key points and performing optimization, and finally screening a sample with the largest target value as an optimization result. The application can effectively improve the optimization efficiency of the electromagnetic structure, reduce the calculation complexity and resource consumption, improve the optimization accuracy and reliability, and has a wide application prospect in the field of electromagnetic structure design.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of electromagnetic structure design, and particularly relates to an online complex high-dimensional electromagnetic structure automatic design method, device and medium. BACKGROUND

[0002] With the wide application of machine learning technology in the automatic design of electromagnetic structures, how to improve the optimization efficiency in high-dimensional space has become an important research direction. Most of the existing technical solutions replace the high-cost commercial simulation software by introducing a surrogate model (such as Gaussian process, support vector machine or deep neural network) to reduce the computational complexity in the optimization process, and directly combine heuristic algorithms to search the global parameter space to find the optimal solution that meets the multi-objective design requirements.

[0003] Although the above method improves the optimization efficiency to a certain extent, the existing technical solutions still have significant deficiencies in practical application. The complexity of high-dimensional design space makes it difficult to guarantee the quality and representativeness of samples, which makes it difficult for the surrogate model to accurately fit the global characteristics of the objective function. In addition, the existing technical solutions usually optimize all input variables directly, which has high computational complexity of the system, and the optimization process wastes a lot of computing resources in redundant dimensions, which is low in efficiency and significantly limits its application effect in electromagnetic structure design.

[0004] Although the design of electromagnetic structures usually involves high-dimensional design space, a large number of variables have a weak influence on the target performance, while a few key variables play a decisive role in the optimization result. Therefore, there is an urgent need for a new technical solution that can effectively reduce the dimension of high-dimensional design space using a surrogate model without increasing the simulation cost, so as to improve the efficiency and overall performance of high-dimensional optimization. SUMMARY

[0005] To at least partially solve one of the technical problems in the prior art, the purpose of the present application is to provide an online complex high-dimensional electromagnetic structure automatic design method, device and medium based on sensitivity analysis.

[0006] The first technical solution adopted by the present application is:

[0007] An online complex high-dimensional electromagnetic structure automatic design method, comprising the following steps:

[0008] Modeling the electromagnetic structure as a binary variable matrix, setting an optimization objective function, and constructing an initial data set;

[0009] Training the surrogate model using the initial data set to obtain an initial surrogate model;

[0010] Sensitivity analysis of matrix variables was performed using the Sobol index. A sample set was generated by Latin hypercube sampling, and the relevant variance and index were calculated based on the surrogate model to quantify the contribution of variables.

[0011] Key points are extracted based on contribution percentage, and the proxy model is then optimized and trained.

[0012] Simulated annealing optimization steps: Construct a simulated annealing algorithm, and based on the optimized surrogate model, perform iterative optimization by generating neighborhood solutions, calculating target values, and adopting a probability-based acceptance strategy;

[0013] After multiple iterations, simulations were performed on a subset of samples to verify the data, and the surrogate model was then merged and updated.

[0014] The key points are recalculated and optimization is performed until the preset maximum number of optimization rounds is reached. Finally, the sample with the largest target value is selected as the optimization result.

[0015] Furthermore, the process of modeling the electromagnetic structure as a binary variable matrix, setting an optimization objective function, and constructing an initial dataset includes:

[0016] The electromagnetic structure is modeled as an n*m dimensional binary variable matrix, where each element in the matrix represents a material unit;

[0017] Define the objective function for quantifying the performance of the electromagnetic structure;

[0018] N is generated through uniform random sampling. sim The initial dataset D0 is formed by calculating the performance curves and target values ​​of each matrix using simulation software.

[0019] Furthermore, the ResNet50 model is used as the surrogate model;

[0020] The step of training the proxy model using the initial dataset to obtain the initial proxy model includes:

[0021] The initial dataset is preprocessed, and the agent model is trained using the preprocessed dataset to obtain the initial agent model. θ0 represents the parameters corresponding to the initial proxy model;

[0022] The loss function during training is defined as follows:

[0023]

[0024] In the formula, λ curve With λ target These are weighting coefficients used to balance the curve loss and the target loss; n is the number of performance curves. S represents the model's predicted value for the nth performance curve. n(x) represents the nth actual performance curve of the electromagnetic structure obtained by simulation software; O(x) represents the model's predicted value of the target, while O(x) represents the actual target value of the electromagnetic structure.

[0025] Furthermore, the sensitivity analysis of matrix variables using the Sobol index, generating a sample set through Latin hypercube sampling, and calculating the relevant variance and index based on a surrogate model to quantify the contribution ratio of variables, includes:

[0026] Sensitivity analysis of the matrix variables was performed using the Sobol exponent, and Latin hypercube sampling was used to ensure that the values ​​of each variable were uniformly distributed in the sample space; let the expected number of samples be N. pre For each variable x i Each generates a random number of length N. pre A sequence, each sequence containing 0 and One 1; generate n*m ​​N pre The length sequences are combined to obtain a uniformly distributed sample set X for sensitivity analysis. pre ;

[0027] Based on the initial agent model Calculate the total variance Var(Y) and the input variable x. i The contribution variance Var(Y) to the output i );

[0028] Based on the calculated Var(Y) and Var(Y) i ), calculate the first-order Sobol exponent S i , which represents the percentage contribution of a single input variable to the output variance.

[0029] Furthermore, the initial agent model Calculate the total variance Var(Y) and the input variable x. i The contribution variance Var(Y) to the output i ),include:

[0030] Variance decomposition measures the contribution of each input variable to the output. Let... The formula for the total variance decomposition of Y is:

[0031]

[0032] In the formula, Y i It is variable x i The impact on output variance measures the change in x while keeping other variables constant. i The impact of the value on the output; d is the total number of variables, Y ij Indicate variable xi The influence of the interaction between x j on the output;

[0033] Using a proxy model For a sample set X pre Calculate the target value Y k , and calculate the mean value:

[0034]

[0035] Add a weight value ω k to the sample, and the new formula is defined as:

[0036]

[0037] According to the mean value obtained, calculate the total variance of the output:

[0038]

[0039] Calculate the conditional expectation of each variable x i , for the case where x i is fixed at 0, filter out the samples in the high and low samples where x i is 0, and calculate their conditional expectation according to the following formula:

[0040]

[0041] For the case where x i is fixed at 1, calculate the conditional expectation E(Y|x i =1) of x i =1, and further calculate the conditional expectation E(Y|x i ):

[0042] E(Y|x i ) = p i E(Y|x i =0) + (1-p i ) E(Y|x i =1)

[0043] In the formula, p i is the probability of x i =0 in the sample; after completing the calculation of the conditional expectation, the conditional expectation variance of each variable is calculated by the following formula:

[0044]

[0045] The formula for calculating the first-order Sobol index S i is:

[0046]

[0047] Where S i The larger the value of x, the better. i The more significant the impact on the model output.

[0048] Furthermore, the step of extracting key points based on contribution ratio and optimizing the training of the proxy model includes:

[0049] The first-order Sobol exponent S of each input variable was calculated. i Sort the variables from largest to smallest, and select the top 64 variables as key variables, denoted as x. imp ={x i1 ,x i2 ,…,x i64 The proxy model is retrained using dataset D0; however, the input matrix of the trained proxy model is no longer composed of the entire space, but is reconstructed into an 8*8 matrix using only these 64 variables. The resulting new proxy model is denoted as .

[0050] Furthermore, the simulated annealing optimization step includes:

[0051] A1. In dataset D0, except for the key variable x imp Other variables are denoted as x unimp ; Select the sample with the largest target from the dataset D0 calculated through simulation, and let it be x. imp With x unimp The value is assigned as the starting solution for the simulated annealing algorithm. During the optimization process, x... unimp The solution will always remain consistent with the initial solution to effectively reduce the dimensionality of the optimization space; its corresponding objective value is denoted as O. curebt ;

[0052] A2, with To optimize the objective, a simulated annealing algorithm is constructed, and initial parameters are set.

[0053] A3. Randomly select x imp Invert one or more variables in the equation to generate a neighborhood solution x' imp And calculate the target value corresponding to the neighborhood solution based on the surrogate model. If o new <o curent If the solution is positive, then the newly generated neighborhood solution is accepted; otherwise, the solution is accepted with probability P. ac Accept, the expression is:

[0054]

[0055] Iteratively update the current solution and temperature. If this step accepts a new solution, then x imp =x' imp Ocurent =O new ; otherwise, keep the original solution; adjust the temperature by using a preset temperature decay rate alpha:

[0056] T = alpha * T

[0057] A4, iterate the process of step A3 until the temperature T is lower than the minimum temperature T min or the number of iteration rounds reaches the maximum number of iterations; after the iteration is completed, the final sample obtained is taken as the starting solution, and the simulated annealing optimization step is cycled again until the number of iterations reaches a preset value; wherein the performance ranking of the samples generated in the iteration process in the front of the preset bit are all updated to the same sample set x top .

[0058] Further, the simulation verification is performed on part of the samples after the multiple rounds of iterations, and the data is merged and the proxy model is updated, including:

[0059] simulate and verify the Top-K samples in the sample set x top , and merge them into the data set D0, and the newly obtained data set is denoted as D1;

[0060] train the proxy model using the data set D1 to obtain a new proxy model and update the weight value omega k .

[0061] The second technical solution adopted by the application is:

[0062] An electronic device, the electronic device includes a processor and a memory, the memory has at least one instruction, at least one program, a code set or an instruction set stored therein, the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize an online complex high-dimensional electromagnetic structure automatic design method as described above.

[0063] The third technical solution adopted by the application is:

[0064] A computer readable storage medium, the storage medium has at least one instruction, at least one program, a code set or an instruction set stored therein, the at least one instruction, the at least one program, the code set or the instruction set is loaded and executed by the processor to realize an online complex high-dimensional electromagnetic structure automatic design method as described above.

[0065] The fourth technical solution adopted by the application is:

[0066] A computer program product or computer program comprises computer instructions stored in a computer-readable storage medium. A processor of a computer device can read the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions so that the computer device performs the method described above.

[0067] The application has the beneficial effects that: the application proposes a spatial dimension reduction scheme based on sensitivity analysis, calculates the sensitivity of input variables by combining high and low confidence samples obtained through simulation and proxy model prediction to reduce the prediction and optimization space, improve the prediction accuracy and reduce the search cost, and gradually improves the accuracy of sensitivity analysis using an online model updating method, so that the algorithm significantly reduces the calculation and time cost while maintaining high accuracy. BRIEF DESCRIPTION OF DRAWINGS

[0068] In order to more clearly illustrate the technical solutions in the embodiments of the application or the prior art, the following introduces the drawings of the related technical solutions in the embodiments of the application or the prior art. It should be understood that the drawings in the following introduction are only for the convenience of clearly describing some embodiments of the technical solutions in the application, and for those skilled in the art, other drawings can also be obtained without creative labor on the basis of these drawings.

[0069] Figure 1 is a step flow chart of an online complex high-dimensional electromagnetic structure automatic design method in the embodiments of the application;

[0070] Figure 2 is a schematic diagram of an online complex high-dimensional electromagnetic structure automatic design method based on sensitivity analysis in the embodiments of the application. DETAILED DESCRIPTION

[0071] The embodiments of the application are described in detail below, and examples of the embodiments are shown in the drawings, wherein the same or similar reference signs represent the same or similar elements or elements with the same or similar functions throughout. The embodiments described below by referring to the drawings are exemplary and are only used to explain the application, and cannot be understood as a limitation of the application. For the step numbers in the following embodiments, they are only set for the convenience of explanation and description, and the order between the steps is not limited in any way, and the execution order of each step in the embodiments can be adaptively adjusted according to the understanding of those skilled in the art.

[0072] In the description of the present application, it should be understood that the orientation description, such as the orientation or position relationship indicated by the upper, lower, front, rear, left, right and the like, is based on the orientation or position relationship shown in the drawings, and is only for the convenience of describing the present application and simplifying the description, and does not indicate or imply that the device or element referred to must have a particular orientation, be constructed and operated in a particular orientation, and therefore cannot be understood as a limitation on the present application.

[0073] In the description of the present application, one or more is understood as one or more, more than two is understood as more than two, greater than, less than, more than, etc. are understood as not including the number, above, below, etc. are understood as including the number. If the first, second is described, it is only for the purpose of distinguishing technical features, and cannot be understood as indicating or implying relative importance or implicitly indicating the number of indicated technical features or the order of indicated technical features.

[0074] In the description of the present application, unless otherwise explicitly limited, the words such as setting, installing, connecting and the like should be broadly understood, and those skilled in the art can reasonably determine the specific meaning of the above words in the present application in combination with the specific content of the technical scheme.

[0075] With the in-depth application of machine learning technology in the field of automatic design of electromagnetic structures, improving the optimization efficiency in high-dimensional space has become an important research direction. Existing design methods introduce surrogate models to replace time-consuming simulation software and combine heuristic algorithms to explore high-dimensional design space, which makes it difficult for the predictor to fit the global characteristics of the objective function accurately using sparse data, and also wastes a lot of computing resources in redundant dimensions. In order to overcome these limitations, the present application proposes an online complex high-dimensional electromagnetic structure automatic design scheme based on sensitivity analysis, which can effectively utilize the surrogate model to complete the dimensionality reduction of the effective high-dimensional design space without increasing the simulation cost, and significantly improve the efficiency and overall performance of high-dimensional optimization. Specifically, the process of the present application scheme can be divided into seven steps: (a) problem definition and mathematical modeling; (b) training surrogate model; (c) sensitivity analysis of variables; (d) extracting key points and optimizing surrogate model; (e) simulated annealing optimization; (f) simulation verification and online surrogate model updating; (g) re-finding key points and performing optimization.

[0076] Example 1

[0077] As shown in Figure 1 and Figure 2 The present embodiment provides an online complex high-dimensional electromagnetic structure automatic design method based on sensitivity analysis, which can effectively improve the optimization efficiency of electromagnetic structures, reduce the computational complexity and resource consumption, improve the optimization accuracy and reliability, and has wide application prospect in the field of electromagnetic structure design. The method specifically includes the following steps:

[0078] S1, model the electromagnetic structure as a binary variable matrix, set the optimization objective function, and construct an initial dataset.

[0079] In some embodiments, step S1 specifically comprises the following steps:

[0080] S1-1: In this step, the electromagnetic structure is first modeled as a binary variable matrix composed of 01 matrices, where each element in the matrix represents a material unit, with 1 representing a metal unit and 0 representing a non-metal unit. According to the size constraints of the target structure and the minimum processing unit, it is modeled as an n*m matrix.

[0081] S1-2: Set the objective function for quantifying the performance of the electromagnetic structure, common objectives include gain, bandwidth, etc. By using simulation software (such as CST STUDIO, HFSS, etc.), we can simulate each structure and calculate its corresponding performance curve S n (x). Where x represents the structure corresponding matrix, n represents different curve types such as S 21 curve, gain curve, etc. Although electromagnetic structures have various performance curves, only some performances in certain frequency bands are often focused on during the design process, such as the maximum gain in the 5-6 GHz frequency band, so the corresponding target value O n (x) of the structure needs to be further calculated. To simplify multi-objective optimization, we further calculate the total target value O(x) to assist subsequent optimization tasks, which is defined as the target value with the largest difference from the preset target.

[0082] S1-3: Generate N sim matrices through uniform random sampling and use simulation software to calculate their performance curves and target values to form an initial dataset D0. At the same time, due to the high time cost of simulation evaluation, N sim often needs to be set to a small value.

[0083] S2, train the initial dataset to obtain an initial surrogate model.

[0084] As an optional implementation, step S2 comprises the following steps:

[0085] S2-1: In order to speed up the optimization process, the embodiment uses a ResNet50 model as a surrogate model to replace the time-consuming full-wave simulation. In this step, the input data needs to be preprocessed first, and the data set is selected in step S1-3. In order to make the model have strong generalization, the input n*m dimensional matrix will be upsampled to 224*224 dimensions. In terms of output, since the performance curve of the electromagnetic structure is generally composed of 1001 dimensions, in order to enhance the understanding of the model to complex high-dimensional data and improve the representation ability, the model will predict the performance curve and the target value of the structure at the same time. The training of the model is constrained by the mean square error (MSE) loss, and the specific loss function is defined by the following formula:

[0086]

[0087] where λ curve and λ target are weight coefficients for balancing the curve loss and target loss, n is the number of performance curves, is the predicted value of the nth performance curve by the model, is the predicted value of the target by the model. Based on this step, we will get the initial surrogate model θ0 is the parameter corresponding to the initial surrogate model, it is worth noting that in the subsequent steps only the target prediction value is output by default

[0088] S3, Sobol index is used to analyze the sensitivity of the matrix variable, and a sample set is generated by Latin hypercube sampling. Based on the surrogate model, the relevant variance and index are calculated to quantify the contribution of the variable.

[0089] As an optional implementation, step S3 includes the following steps:

[0090] S3-1: The embodiment of the application uses Sobol index to complete the sensitivity analysis of the matrix variable. In order to reduce the calculation complexity and avoid the high cost of complete enumeration, Latin hypercube sampling can be used to ensure that the values of each variable are uniformly distributed in the sample space. Let the expected generated sample be N pre (N pre >>N sim ), for each variable x i (total n*m variables), generate a random sequence of length N pre , each sequence contains 0 and 1. Combine the generated n*m N pre length sequences, and you can get a uniformly distributed sample set X pre for sensitivity analysis.

[0091] S3-2: Measure the contribution of each input variable to the output through variance decomposition. Let... The formula for the total variance decomposition of Y is:

[0092]

[0093] Among them, Y i Refers to variable x i The impact on output variance measures the change in x while keeping other variables constant. i The impact of the value on the output, where d is the total number of variables, and Y... ij Indicate variable x i With x j The interaction between them affects the output. To calculate the first-order Sobol exponent, the key lies in calculating the total variance Var(Y) and the relationship between the input variable x. i The contribution variance Var(Y) to the output i This step uses the proxy model obtained from S2. For sample set X pre Calculate the target value Y k And calculate the mean based on this:

[0094]

[0095] However, due to Y k The samples predicted by the surrogate model are low-confidence samples, while the samples in dataset D0 are high-confidence samples that have been verified through simulation. Therefore, a weight ω needs to be added to each item. k To maximize the accuracy of the sensitivity analysis, where ω corresponds to each high-confidence sample. k All are 1, ω for low confidence samples k The value is determined by the prediction accuracy of the surrogate model and the current optimization round. The prediction accuracy of the surrogate model is determined by the Kendall's Tau correlation coefficient, hereinafter referred to as KTau, which is a nonparametric statistic that measures the correlation between two variables. It is usually used to assess the rank correlation between two variables, and its formula is:

[0096]

[0097] Where set x represents the sample labels of dataset D0, i.e., the actual simulation performance target corresponding to each sample, and set y represents the surrogate model output corresponding to each sample.

[0098] Let the total number of optimization rounds be M. max Round M, current round number, weight ω of low confidence samples k for When the initial optimization round M is 1, ωk for At this point, the expected value is defined by the following new formula:

[0099]

[0100] Using the obtained mean, the total output variance can be further calculated:

[0101]

[0102] S3-3: For calculating Var(Y) i First, we need to calculate each variable x. i The conditional expectation. For a fixed x i If the value is 0, select the high and low samples obtained in the previous steps for x. i For samples with a value of 0, their conditional expectation is calculated using the following formula:

[0103]

[0104] The high and low confidence level sample weights ω k Keeping the same as the previous step. Similarly, for a fixed x... i For the case of location 1, x can be calculated similarly. i The conditional expectation E(Y|x) is 1 i =1), and further calculate the conditional expectation E(Y|x) i ):

[0105] E(Y|x i ) = p i E(Y|x i =0)+(1-p i )E(Y|x i =1)

[0106] Where p i For x in the sample i The probability that x is 0 in the sample i The proportion of samples with a value of zero out of the total sample. After calculating the conditional expectation, the conditional expectation variance of each variable can be calculated using the following formula:

[0107]

[0108] S3-4: Using Var(Y) and Var(Y) calculated in the first two steps respectively i This allows for the calculation of the first-order Sobol exponent S, a key parameter used to measure sensitivity. i The index is calculated to represent the percentage contribution of a single input variable to the output variance, and its mathematical expression is:

[0109]

[0110] By calculating the first-order Sobol index S i of each input variable, we can quantify the independent contribution of each input to the model output. The greater the value of S i , the more significant the influence of variable x i on the model output.

[0111] S4, extract key points according to the contribution ratio, and optimize the training of the surrogate model.

[0112] Exemplarily, the first-order Sobol index S i of each input variable calculated in step S3 is sorted from large to small, and the top 64 variables are taken as x imp ={x i1 ,x i2 ,…,x i64}. The surrogate model training of step S2 is completed again using dataset D0, but the input matrix of the surrogate model trained in this step is no longer composed of the full space, but only the 8*8 matrix input reconstructed from the 64 variables, which effectively reduces the prediction difficulty of the surrogate model for complex high-dimensional space. The new surrogate model obtained by training is denoted as

[0113] S5, simulated annealing optimization step: construct a simulated annealing algorithm based on the optimized surrogate model, and perform iterative optimization through generating neighborhood solutions, calculating target values, and probabilistic acceptance strategy.

[0114] In one embodiment, step S5 specifically includes:

[0115] S5-1: Let the remaining variables be x unimp . From the dataset D0 obtained by simulation calculation, the sample with the maximum target is selected, and x imp and x unimp are assigned values to serve as the starting solution of the simulated annealing algorithm. In the optimization process, x unimp will always remain consistent with the starting solution to effectively reduce the dimension of the optimization space, and the corresponding target value is denoted as O curent .

[0116] S5-2: With O as the optimization target, construct a simulated annealing algorithm, the core mechanism of which is to gradually reduce the "energy" level of the system by introducing a "temperature" parameter to find the global optimal solution. In the initial stage, the parameters of the simulated annealing algorithm need to be set, among which the initial temperature needs to be set to a higher value such as T0=1000, the temperature decay rate a is set to a floating point number in the interval (0.8, 0.99), and the lowest temperature T minSet to 1, the maximum number of iterations max_iter is set to 1000.

[0117] S5-3: Randomly reverse one or more variables in x to generate neighborhood solution x' imp imp , and calculate the objective value corresponding to the neighborhood solution based on the surrogate model If O new < O curent , accept the newly generated neighborhood solution, otherwise accept it with probability P ac , which is defined by the following formula:

[0118]

[0119] Iteratively update the current solution and temperature, if the new solution is accepted at this step, x imp = x' imp , O curent = O new , otherwise keep the original solution, it is worth noting that the top 50 samples generated in the process will be continuously updated and stored in the sample set x top . Then adjust the temperature according to the following formula using the preset temperature decay rate:

[0120] T = aT

[0121] S5-4: Continue iteration of the process of S5-3 until the temperature T is lower than the minimum temperature T min or the number of iterations reaches the maximum number of iterations max_iter. After the iteration is completed, the final sample obtained is used as the starting solution, and the S5 step is cycled again, until a total of 10 rounds of iteration are completed. The top 50 samples generated in the process will be updated to the same sample set x top .

[0122] S6, after multiple iterations, simulate and verify part of the samples, merge the data and update the surrogate model.

[0123] Exemplarily, in order to enable the surrogate model to quickly adapt to environmental changes and gradually improve the prediction accuracy and the reliability of sensitivity analysis, the present application applies an online model updating method in the optimization process, so that it can adjust the strategy in time according to new conditions, enhance the dynamic adaptability of the model, and improve the optimization efficiency. Therefore, in this step, the Top-K samples in x top are simulated and verified, and are merged into the data set D0, and the newly obtained data set is denoted as D1. The surrogate model training in the S2 process is completed again using the D1 data, and a new surrogate model is obtained, and the KTau value of the model is calculated to update the weight value ω k ​The input of the model at this time is also consistent with that described in step S2. When this entire process is completed, the optimization round M plus one.

[0124] S7, recalculate the key points and perform optimization until the preset maximum optimization round is reached, and finally select the sample with the maximum target value as the optimization result.

[0125] Specifically, steps S3 to S6 are executed in a loop, and each loop uses the new agent model updated online and the data set with the added new simulation data, until the optimization round M reaches the preset maximum optimization round M max , and finally selects the sample with the maximum target value from all simulation-verified samples as the final optimization result.

[0126] In general, the existing machine learning-based automatic electromagnetic structure design method usually uses a simple agent model combined with a traditional heuristic algorithm to search the entire design space, resulting in poor global optimization ability and low efficiency. This method is difficult to accurately fit the global characteristics of the objective function when dealing with complex high-dimensional space design problems due to the difficulty in ensuring sample quality and representativeness, and the optimization process wastes a lot of computing resources in redundant dimensions, and cannot quickly find the optimal solution region. In contrast, the present application innovatively proposes a space dimension reduction scheme based on sensitivity analysis, which calculates the sensitivity of the input variables by combining high and low confidence samples obtained by simulation and agent model prediction to reduce the prediction and optimization space, improve prediction accuracy and reduce search cost, and uses an online model updating method to gradually improve the accuracy of sensitivity analysis, so that the algorithm can maintain high accuracy while significantly reducing computing and time costs.

[0127] In summary, the method of the present application has at least the following advantages and beneficial effects:

[0128] (1) By using the sensitivity analysis technology based on the agent model, the present application can significantly improve the optimization efficiency in electromagnetic structure design tasks, especially in complex high-dimensional design tasks. The agent model can further improve the data representation ability, making the inference result more accurate.

[0129] (2) The existing technology usually needs to search the entire huge design space, while the present application uses the prediction data of the agent model to calculate the sensitivity of the variables combined with high and low confidence samples, which can achieve reliable space dimension reduction without increasing additional simulation cost, which greatly reduces the difficulty of model prediction and optimization search, and also reduces the required computing resources and costs.

[0130] (3) The application uses an online model updating method to gradually improve the prediction accuracy of the proxy model for the good sample area in the optimization process, and continuously improve the sensitivity value, further improving the optimization ability of the algorithm.

[0131] (4) The application has wide application adaptability, and is not only suitable for electromagnetic structure design problems, but also can be applied to other structure design problems, such as super surface design, photon design, bearing design, etc., significantly improving the efficiency and performance of these tasks. Therefore, the application scheme has higher technical advantages and wide practical application value.

[0132] Embodiment 2

[0133] The embodiment of the application further provides an electronic device, which comprises a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set, which are loaded and executed by the processor to realize the online complex high-dimensional electromagnetic structure automatic design method as shown in Figure 1

[0134] It can be understood that the memory can include a random access memory (RAM) and a read-only memory (ROM). Optionally, the memory includes a non-transitory computer-readable storage medium. The memory can be used to store instructions, programs, codes, code sets or instruction sets. The memory can include a program storage area and a data storage area, wherein the program storage area can store instructions for implementing an operating system, instructions for at least one function, instructions for implementing various method embodiments described above, etc.; and the data storage area can store data created according to the use of the server, etc.

[0135] ​The processor can include one or more processing cores. The processor connects various parts within the entire server by various interfaces and lines, executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory, and calling data stored in the memory. Optionally, the processor can be implemented in at least one of a hardware form of a digital signal processing (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor can be integrated with one or several combinations of a central processing unit (CPU) and a modem. Among them, the CPU mainly processes operating systems and application programs; the modem is used to process wireless communication. It can be understood that the above-mentioned modem can also not be integrated into the processor, but be implemented by a single chip.

[0136] Since the electronic device is an electronic device corresponding to the online complex high-dimensional electromagnetic structure automatic design method of the embodiments of the application, and the principle of solving problems of the electronic device is similar to that of the method, the implementation of the electronic device can be referred to the implementation process of the above-mentioned method embodiments, and the repeated parts will not be described herein.

[0137] Embodiment 3

[0138] The embodiments of the application also provide a computer readable storage medium, wherein at least one instruction, at least one program, a code set or an instruction set are stored in the storage medium, and the at least one instruction, the at least one program, the code set or the instruction set are loaded and executed by a processor to implement an online complex high-dimensional electromagnetic structure automatic design method as shown in Figure 1

[0139] ​Those skilled in the art can understand that all or part of the steps of various methods in the above embodiments can be completed by instructing the relevant hardware through a program, and the program can be stored in a computer readable storage medium, including Read-Only Memory (ROM), Random Access Memory (RAM), Programmable Read-only Memory (PROM), Erasable Programmable Read Only Memory (EPROM), One-time Programmable Read-Only Memory (OTPROM), Electrically-Erasable Programmable Read-Only Memory (EEPROM), Compact Disc Read-Only Memory (CD-ROM) or other optical disk storage, magnetic disk storage, magnetic tape storage, or any other medium that can be used to carry or store data which can be read by a computer.

[0140] Since the storage medium is a storage medium corresponding to the online automatic design method of complex high-dimensional electromagnetic structure of the embodiments of the present application, and the principle of solving problems of the storage medium is similar to the method, the implementation of the storage medium can refer to the implementation process of the above-mentioned method embodiments, and the repeated parts will not be described again.

[0141] Embodiment 4

[0142] In some possible implementation manners, various aspects of the method of the embodiments of the present application can also be implemented in the form of a program product, which includes program codes for causing a computer device to execute the steps of the online automatic design method of complex high-dimensional electromagnetic structure according to various exemplary embodiments of the present application described above in the specification when the program product is run on the computer device. Wherein, the executable computer program codes or "codes" for executing various embodiments can be written in high-level programming languages such as C, C++, C#, Smalltalk, Java, JavaScript, Visual Basic, Structured Query Language (for example, Transact-SQL), Perl, or in various other programming languages.

[0143] It should be understood that various aspects of the application can be implemented in hardware, software, firmware or a combination of them. In the above embodiments, various steps or methods can be implemented in software or firmware which is stored in memory and executed by a suitable instruction execution system. For example, if implemented in hardware, and in another embodiment, any of the following technologies, known in the art, or their combination, can be used: discrete logic circuitry having logic gates for implementing logic functions upon an application of data signals, application-specific integrated circuits having appropriate combinational logic gates, programmable gate arrays (PGA), field-programmable gate arrays (FPGA), and so forth.

[0144] In the description of the present specification, the description of the terms "one embodiment", "some embodiments", "an example", "a specific example", or "some examples" and the like means that the specific features, structures, materials or characteristics described in connection with the embodiment or example are included in at least one embodiment or example of the present application. In the present specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Also, the specific features, structures, materials or characteristics described can be combined in any appropriate manner in one or more embodiments or examples. In addition, the person skilled in the art can combine and combine the different embodiments or examples described in the present specification and the features of the different embodiments or examples, without contradiction.

[0145] The above embodiments are only for the purpose of illustrating the technical concept and characteristics of the present application, and the purpose is to enable those skilled in the art to understand the content of the present application and to implement it, and cannot limit the protection scope of the present application. Any equivalent changes or modifications made according to the essence of the present application should be covered within the protection scope of the present application.

Claims

1. An online automatic design method for complex high-dimensional electromagnetic structures, characterized in that, The method comprises the following steps: modeling the electromagnetic structure as a binary variable matrix, setting an optimization objective function, and constructing an initial data set; training the surrogate model using the initial data set to obtain an initial surrogate model; performing sensitivity analysis on the matrix variables using Sobol indices, generating a sample set through Latin hypercube sampling, and calculating the relevant variance and indices based on the surrogate model to quantify the contribution proportion of the variables; extracting key points according to the contribution proportion and optimizing the training of the surrogate model; simulated annealing optimization step: constructing a simulated annealing algorithm, based on the optimized surrogate model, generating neighborhood solutions, calculating target values, and iteratively optimizing based on the probability acceptance strategy; after multiple iterations, simulating and verifying part of the samples, merging the data and updating the surrogate model; recomputing the key points and performing optimization until the preset maximum optimization round is reached, and finally selecting the sample with the maximum target value as the optimization result; The electromagnetic structure is modeled as an n*m-dimensional binary variable matrix, wherein each element in the matrix represents a material unit. A target function is set to quantify the performance of the electromagnetic structure. The method comprises the following steps: By uniform random sampling, generate matrices and use simulation software to calculate their performance curves and target values to form an initial data set ; The initial data set is used to train the surrogate model to obtain an initial surrogate model. The first-order Sobol index of each input variable calculated The top 64 variables are sorted from large to small and taken as the key variables, denoted as The proxy model is retrained using the dataset The input matrix of the trained proxy model is no longer composed of the full space, but only of the 64 variables reconstructed into an 8*8 matrix input. The new proxy model obtained by training is denoted as .

2. The method according to claim 1, wherein, The loss function during the training process is defined as: The initial data set is preprocessed, and the preprocessed data set is used to train the surrogate model to obtain an initial surrogate model , is a parameter corresponding to the initial surrogate model The Sobol indices are used to perform sensitivity analysis on the matrix variables, and a sample set is generated through Latin hypercube sampling. In the formula, and is a weight coefficient for balancing the curve loss and the target loss; is the number of performance curves; is the predicted value of the model to the first performance curve, is the first real performance curve of the electromagnetic structure calculated by the simulation software; is the predicted value of the model to the target, is the real target value of the electromagnetic structure.

3. The method of claim 1, wherein the method is characterized by: According to the obtained mean value, the total output variance is calculated as follows: Sobol index is used to complete sensitivity analysis of matrix variables, and Latin hypercube sampling is used to ensure that the values of each variable are uniformly distributed in the sample space; set the expected generated sample as n*m, for each variable , generate a random sequence with a length of ; combine the generated n*m length sequences to obtain a uniformly distributed sample set for sensitivity analysis ; Based on an initial proxy model , compute total variance Contribution variance of input variables to output ; According to the calculation With , the first-order Sobol index , which represents the proportion of the contribution of a single input variable to the output variance.

4. The method according to claim 3, wherein, The initial proxy model , compute total variance of the input variables contribution variance to the output, comprising: Using a proxy model To a set of samples Computing target values Adding weight values to samples The mean is calculated as follows: The simulated annealing optimization step comprises the following steps: Compute the conditional expectation of each variable for the case where the fixed values of the other variables are 0, select the samples in the high and low samples for which the value of the variable is 0, and compute their conditional expectation according to the following formula: For the case of fixed the conditional expectation is calculated to be and the conditional expectation is further calculated to be where is the probability that the sample is 0; after the computation of the conditional expectation, the conditional expectation variance of each variable is computed by the following formula: First order Sobol indices The formula for the calculation is: wherein The greater the value of The more significant the impact of the variable on the model output.

5. The method of claim 1, wherein the method is characterized by: After multiple iterations, part of the samples are simulated and verified, the data is merged, and the surrogate model is updated. A1, in the data set , other variables except the key variable are denoted as ; the target maximum sample is screened out from the data set calculated by simulation, and is assigned to and as the starting solution of the simulated annealing algorithm. In the optimization process, will always be consistent with the starting solution to effectively reduce the optimization space dimension, and the corresponding target value is denoted as ; A2, with To optimize the objective, the simulated annealing algorithm is constructed, and the initial parameters are set; A3, random pair reversing one or more variables in the neighborhood solution to generate a new neighborhood solution and computing the objective value corresponding to the new neighborhood solution based on the surrogate model If , then accept the new neighborhood solution; otherwise, accept with probability , the expression is: Iteratively update the current solution and temperature, if this step accepts the new solution, then , ; otherwise, keep the original solution; adjust the temperature using a pre-set temperature decay rate Adjust the temperature: A4、to step A3 process continues iteration, until the temperature below the minimum temperature or the number of iterations reaches the maximum number of iterations; after the iteration is completed, the final sample obtained is used as the starting solution, and the simulated annealing optimization step is cycled again until the number of iterations reaches the preset value; wherein the samples with the top performance ranking in the iteration process are updated to the same sample set medium.

6. The method of claim 1, wherein the method is an online method for automatic design of complex high-dimensional electromagnetic structures. The electronic device comprises a processor and a memory, and the memory stores at least one instruction, at least one program, a code set or an instruction set. simulate and merge the top-k samples in the sample set into the data set , and the new data set is denoted as ; Utilizing the dataset training the proxy model to obtain a new proxy model and updating the weight values .

7. An electronic device, comprising: The storage medium stores at least one instruction, at least one program, a code set or an instruction set, which is loaded and executed by the processor to realize the method of any one of claims 1 to 6.

8. A computer-readable storage medium, characterized in that, ​

Images