Antenna optimization method and device based on adaptive filling criterion multi-objective evolution and storage medium

Abstract

The application discloses an antenna optimization method and device based on a self-adaptive filling criterion multi-objective evolution and a storage medium. The application directly optimizes design parameter variables based on a target function. A candidate population finally searched contains multiple potential individuals. The individuals are parameter combination schemes with superior design indexes. The application has high automation degree and high antenna optimization design efficiency. By applying the self-adaptive filling criterion, the maximum demand of a current population can be determined, and then a corresponding filling criterion is executed. The application can flexibly cope with problems with various characteristics and improve the utilization benefit of computing resources. The application is widely applied to the technical field of antennas.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of antennas, and in particular to an antenna optimization method, device and storage medium based on adaptive filling criterion multi-objective evolution. BACKGROUND

[0002] In a wireless communication system, an antenna is an indispensable transceiving signal device, and its performance directly affects the quality of the entire system. With the rapid development of wireless communication systems, antennas are required to achieve higher performance indicators, and therefore more complex structures are needed. This inevitably increases the number of design variables and their combinations, making the entire design and debugging process require a large number of manual parameter adjustment attempts, which makes it difficult to optimize the device in a short period of time.

[0003] The traditional manual parameter adjustment optimization design method requires testing the positive and negative correlation between each design parameter of the antenna and the antenna indicator, then fine-tuning each parameter, and verifying as many combinations as possible. However, when there are too many antenna design parameters, this method will consume a lot of time for engineers due to the large number of combination schemes. SUMMARY

[0004] In view of the technical problems of too many combination schemes and time-consuming in determining design parameters in the current antenna design process, the purpose of the present application is to provide an antenna optimization method, device and storage medium based on adaptive filling criterion multi-objective evolution.

[0005] In one aspect, the present application embodiment includes an antenna optimization method based on adaptive filling criterion multi-objective evolution, which includes the following steps:

[0006] Obtaining design parameter variables and objective functions of an antenna to be optimized; the objective function represents the optimization target of the design parameter variables;

[0007] Performing at least one round of a loop process until a loop end condition is met; any round of the loop process includes:

[0008] Training and updating at least one surrogate model; the surrogate model is used to fit the response relationship between the design parameter variables and the objective function;

[0009] Searching for optimization in a prediction space to obtain a candidate population; the prediction space is the surrogate model after being trained in the current round of the loop process, and is obtained by predicting a population with the design parameter variables as individuals;

[0010] Determining the population evolution state according to the candidate population;

[0011] determine an adaptive filling criterion according to the population evolution state;

[0012] perform electromagnetic simulation with the adaptive filling criterion, and add a result of the electromagnetic simulation to the archive data.

[0013] Further, the design parameter variable and the objective function of the antenna to be optimized are obtained, including:

[0014] setting

[0015]

[0016]

[0017]

[0018]

[0019]

[0020] as the objective function;

[0021] wherein represents a gain of the antenna to be optimized at an angle of θ , in dBi, represents a gain flatness in the angle range, represents a return loss of the antenna to be optimized at a frequency of , in dB, represents a maximum value of the input return loss S 11 in the working frequency range; represents an axial ratio of the antenna to be optimized at a frequency of , in dB; represents a minimum value of a main polarization of a 90° plane of the antenna to be optimized, represents a minimum value of a main polarization of a 0° plane of the antenna to be optimized, in dB; represents a maximum value of a cross polarization of the 90° plane of the antenna to be optimized, represents a maximum value of a cross polarization of the 0° plane of the antenna to be optimized, in dB.

[0022] Further, the at least one proxy model is trained and updated, including:

[0023] when the current cycle process is the first cycle process, the proxy model is trained with an initial value of the archive data as training data;

[0024] When the current cycle is a cycle other than the first cycle, the archive data added in the last cycle is used as training data, and the agent model trained in the last cycle is used as the basis for training.

[0025] Further, the population evolution state is determined according to the candidate population, and the method comprises:

[0026] A mixed population is obtained, wherein the mixed population is obtained by mixing the non-dominated frontiers of the candidate population and the non-dominated frontiers of the archive data.

[0027] The non-dominated frontiers of the mixed population are obtained.

[0028] The population evolution state is determined according to the non-dominated frontiers of the mixed population.

[0029] Further, the population evolution state is determined according to the non-dominated frontiers of the mixed population, and the method comprises:

[0030] The first number of individuals and the second number of individuals are obtained according to the non-dominated frontiers of the mixed population, wherein the first number of individuals is the number of individuals from the candidate population in the non-dominated frontiers of the mixed population, and the second number of individuals is the total number of individuals in the non-dominated frontiers of the mixed population.

[0031] The ratio of the first number of individuals to the second number of individuals is obtained.

[0032] When the ratio is greater than or equal to a first threshold value, the population evolution state is determined to be in an early evolution stage.

[0033] When the ratio is less than the first threshold value and greater than a second threshold value, the population evolution state is determined to be in a late evolution stage.

[0034] When the ratio is less than or equal to the second threshold value, the population evolution state is determined to be in a prediction deviation stage.

[0035] Further, the adaptive filling criterion is determined according to the population evolution state, and the method comprises:

[0036] When the population evolution state is in the early evolution stage, the adaptive filling criterion is determined to be a population convergence improvement filling criterion.

[0037] When the population evolution state is in the late evolution stage, the adaptive filling criterion is determined to be a population diversity improvement filling criterion.

[0038] When the population evolution state is in the prediction deviation stage, the adaptive filling criterion is determined to be a population precision enhancement filling criterion.

[0039] Further, the electromagnetic simulation according to the adaptive filling criterion comprises:

[0040] When the adaptive filling criterion is a population convergence promotion filling criterion, convergence promotion of each non-dominated individual in the candidate population is obtained by taking the non-dominated individual in the archive data added in the last cycle as a reference, and the electromagnetic simulation is performed on the individual with the largest convergence promotion in the candidate population.

[0041] When the adaptive filling criterion is a population diversity promotion filling criterion, the angle between each non-dominated individual in the candidate population and the closest non-dominated individual in the archive data added in the last cycle is obtained, and the electromagnetic simulation is performed on the individual with the largest angle in the candidate population.

[0042] When the adaptive filling criterion is a population accuracy enhancement filling criterion, the sum of the uncertainties provided by all the surrogate models for each individual in the candidate population is obtained, and the electromagnetic simulation is performed on the individual with the largest sum of uncertainties in the candidate population.

[0043] Further, the cycle end condition is that the total number of real function calculations reaches a number threshold.

[0044] In another aspect, the embodiments of the present application also include a computer device comprising a memory and a processor, the memory being used to store at least one program, and the processor being used to load the at least one program to execute the antenna optimization method based on adaptive filling criterion multi-objective evolution in the embodiments.

[0045] In another aspect, the embodiments of the present application also include a computer readable storage medium, wherein a program executable by a processor is stored, and the program executable by the processor is used to execute the antenna optimization method based on adaptive filling criterion multi-objective evolution in the embodiments when executed by the processor.

[0046] The antenna optimization method based on adaptive filling criterion multi-objective evolution in the embodiments has the beneficial effects that: the antenna optimization method based on adaptive filling criterion multi-objective evolution in the embodiments optimizes the design parameter variable directly based on the objective function, and finally searches for a candidate population containing multiple potential individuals, which are parameter combination schemes with superior design indicators, has high automation degree, and has high efficiency of antenna optimization design; by applying the adaptive filling criterion, the maximum demand of the current population can be determined first, and then the corresponding filling criterion is executed, which can flexibly cope with problems with various characteristics and improve the utilization benefit of computing resources. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 The step schematic diagram of the antenna optimization method based on adaptive filling criterion multi-objective evolution in the embodiments;

[0048] Figure 2 A schematic diagram of the principle of the adaptive filling criterion in the embodiment;

[0049] Figure 3 A diagram defining the structure and parameters of the patch antenna to be optimized in the embodiment;

[0050] Figure 4 A comparison diagram of the input return loss before and after the optimization of the patch antenna in the embodiment;

[0051] Figure 5 A comparison diagram of the axial ratio before and after the optimization of the patch antenna in the embodiment;

[0052] Figure 6 The final optimization results of the patch antenna by each algorithm in the embodiment. DETAILED DESCRIPTION

[0053] A multi-objective evolutionary algorithm assisted by a proxy model generally consists of three parts: a proxy model, an optimizer, and a filling criterion. The filling criterion is a strategy for selecting potential individuals from the candidate population obtained by the optimizer for real function calculation. Real function calculation refers to performing software simulation or processing measurement on the actual antenna model corresponding to the individual to obtain the performance index of the model. On the one hand, the filling criterion can select individuals with superior convergence or diversity for real function calculation to enhance the convergence or diversity of the population, and on the other hand, it can select individuals with high uncertainty for real function calculation to enhance the prediction accuracy of the proxy model.

[0054] If the filling criterion in the multi-objective evolutionary algorithm is comprehensively considered in terms of the convergence, diversity, and uncertainty of the individuals, and is converted into an index that equally considers the needs of the three aspects, such operation cannot flexibly solve problems with different characteristics, and can cause waste of computing resources. However, the use of intelligent optimization algorithms can effectively accelerate the optimization design process.

[0055] Based on the above principle, in the embodiment, an antenna optimization method based on multi-objective evolution of adaptive filling criterion is provided. Referring to Figure 1 , the antenna optimization method based on multi-objective evolution of adaptive filling criterion comprises the following steps:

[0056] S1. Obtain the design parameter variables and objective functions of the antenna to be optimized;

[0057] Perform at least one round of loop process until the loop end condition is met; any round of loop process comprises the following steps S2-S6:

[0058] S2. Update the training of at least one proxy model; the proxy model is used to fit the response relationship between the design parameter variables and the objective functions;

[0059] S3. Searching for optimization in the prediction space to obtain a candidate population; the prediction space is a surrogate model trained after the current cycle process, and the population with the design parameter variables as individuals is predicted to obtain;

[0060] S4. Determining the population evolution state according to the candidate population;

[0061] S5. Determining the adaptive filling criterion according to the population evolution state;

[0062] S6. Performing electromagnetic simulation according to the adaptive filling criterion, and adding the result of the electromagnetic simulation to the archive data.

[0063] In the embodiment, the target of the antenna optimization method based on the adaptive filling criterion multi-objective evolution is to determine the specific values of the design parameter variables of the antenna to be optimized, so as to design an antenna with such specific values of the parameters and to manufacture the antenna.

[0064] In step S1, for an antenna to be optimized, there can be multiple design parameter variables, including structural parameters and material parameters of the device. These design parameter variables each have a corresponding value range (parameter space), so the number of parameter value combinations formed is large, and the specific values of the multiple design parameter variables need to be determined by performing steps S1-S6 (wherein one continuous step S2-S6 constitutes a cycle process, and multiple cycle processes can be required to be performed).

[0065] In the embodiment, when step S1 is performed, the design parameter variables of the selected antenna to be optimized are selected, and these design parameter variables each have a corresponding objective function. The objective function of a design parameter variable represents the condition satisfied when the design parameter variable obtains the optimal (which can be global optimal or local optimal) value.

[0066] In step S1, the design performance indicators such as gain flatness, input return loss, axial ratio, main polarization, and cross polarization can be used as design standards of the objective function, so as to set the following objective functions:

[0067] (1)

[0068] Wherein represents the gain of the antenna to be optimized at an angle of θ , and is in dBi, wherein θ is a variable, the minimum value of the value range of which is , and the maximum value of the value range of which is ; represents the flatness of the gain in the angle range, and the smaller the value is, the better the gain flat-top characteristic of the antenna is.

[0069] (2)

[0070] wherein represents the magnitude of the return loss of the antenna to be optimized at a frequency of , in dB, wherein is a variable, the minimum value of the range of values of which is and the maximum value of the range of values of which is ; represents the maximum value of the input return loss of the antenna to be optimized in the working frequency band, the smaller of which indicates that the input return loss of the antenna in the entire frequency band is smaller. S 11

[0071] (3)

[0072] wherein represents the axial ratio of the antenna to be optimized at a frequency of , in dB; the smaller of which indicates that the antenna is closer to circular polarization.

[0073] (4)

[0074] wherein represents the minimum value of the principal polarization of the 90° plane of the antenna to be optimized, represents the minimum value of the principal polarization of the 0° plane of the antenna to be optimized, in dB; the smaller of which indicates that the principal polarization of the antenna is greater

[0075] (5)

[0076] wherein represents the maximum value of the cross-polarization of the 90° plane of the antenna to be optimized, represents the maximum value of the cross-polarization of the 0° plane of the antenna to be optimized, in dB; the smaller of which indicates that the cross-polarization of the antenna is smaller.

[0077] After step S1 is performed, the design parameter variables of the antenna to be optimized can be converted into the form of a population. Specifically, the design parameter variables of the antenna to be optimized can be represented as

[0078] (design parameter variable 1, design parameter variable 2, …, design parameter variable D )

[0079] ​​The vector form of the design parameter variable is given, where D is the number of design parameter variables to be optimized. For example, in this embodiment, design parameter variable 1 can be a structural parameter such as "mesh size", design parameter variable 2 can be a structural parameter such as "substrate thickness", and so on. D-1 Specifically, this could be a material parameter such as "dielectric constant," or a design parameter variable. D Specifically, it could be a material parameter such as "dielectric loss tangent".

[0080] In this embodiment, multiple individuals can be set, and all individuals constitute the initial population. Each individual is shaped like...

[0081] (Design parameter variable 1, design parameter variable 2, ..., design parameter variable) D )

[0082] The vector form represents each individual, where each individual is a 1-row, D-column vector. The initial values ​​of the individuals (the specific values ​​of design parameter variables such as '1' in the individuals before steps S2-S6 are executed) can be determined using the Latin hypercube sampling method within the defined value range (parameter space) of each design parameter variable. For example, in this embodiment, by using the Latin hypercube sampling method, 11×D-1 individuals can be generated. These 11×D-1 individuals form an initial population, which can be used as the initial values ​​for the archived data.

[0083] After obtaining multiple individuals and their initial population through step S1, the various cyclic processes begin to be executed. In this embodiment, the overall principle of each cyclic process is the same, with one specific cyclic process (the first...) as the starting point. Let's take the cycle process as an example to illustrate.

[0084] No. The cycle process includes the following steps:

[0085] S2. Train and update at least one proxy model;

[0086] S3. Search for the best candidate population in the prediction space;

[0087] S4. Determine the population evolutionary state based on the candidate populations;

[0088] S5. Determine the adaptive filling criterion based on the population evolutionary state;

[0089] S6. Perform electromagnetic simulation using the adaptive filling criterion and add the results of the electromagnetic simulation to the archived data.

[0090] In step S2, the surrogate model is used to fit the response relationship between each design parameter variable and its corresponding objective function. Machine learning models such as Kriging models, radial basis function networks, and artificial neural networks can be used as surrogate models, but are not limited to these. In this embodiment, a Kriging model is used as the surrogate model.

[0091] In step S2, if =1, which means that in the first round of the loop, since there is no previous round of the loop, the initial value of the archived data can be set (e.g., randomly generated, or parameters obtained from other electromagnetic simulations), and the initial value of the archived data can be used to train the surrogate model.

[0092] In step S2, if If ≠1, then the previous iteration process (the first iteration) can be used. -1 round of iteration) The archived data added in step S6 is used as training data in the previous round of iteration (the -1 round of iteration). -1 round of loop process) Execute step S2 to train the updated agent model and continue training.

[0093] Specifically, the training data (archived data) used to train the agent model in step S2 is from the previous loop process (the... The results are obtained through electromagnetic simulation in a -1 round cycle process. Electromagnetic simulation can be a process of simulating values ​​such as gain flatness based on the specific values ​​of antenna design parameters. Therefore, the training data (archived data) contains the simulation results of design parameters such as grid size, substrate thickness, dielectric constant, dielectric loss tangent, etc., and the correspondence between them and objective functions such as gain flatness. This can be used to train the surrogate model's ability to fit the response relationship between design parameters and their corresponding objective functions.

[0094] In this embodiment, M proxy models can be trained in parallel when performing step S2, where M is the number of objective functions.

[0095] In the During the round of iteration, after completing step S2 to train the surrogate model for this round of iteration, the fitting ability of the surrogate model is trained. The trained surrogate model is then used for fitting, that is, in the process of finding the objective functions that satisfy (1)-(5), multiple candidates can be generated (e.g., design parameter variable 1, design parameter variable 2, ... design parameter variable 1). D (The specific values ​​of etc.) These candidates constitute the prediction space of the surrogate model.

[0096] In step S3, an intelligent optimization algorithm is used as the optimizer to search for and optimize within the prediction space of the surrogate model to obtain candidate populations. Suitable intelligent optimization algorithms include, but are not limited to, genetic algorithms (GA), differential evolution (DE), non-dominated sorting-based genetic algorithms (NSGA-II), decomposition-based multi-objective evolutionary algorithms (MOEA / D), and reference vector-guided multi-objective evolutionary algorithms (RVEA). In this embodiment, a reference vector-guided multi-objective evolutionary algorithm is used as the optimizer.

[0097] In this embodiment, the objective function of the offspring in the optimizer is not calculated using the true function, but is a predicted value given by the surrogate model.

[0098] In this embodiment, by executing step S3 to run the optimizer, multiple individuals can be selected from all individuals in the prediction space of the surrogate model trained in step S2 through search and optimization. These selected individuals constitute the first... The candidate population for the round-robin process. Then proceed to step S4.

[0099] In the During the round-robin process, when executing step S4, which is to determine the population evolutionary state based on the candidate population, the following steps can be performed:

[0100] S401. Obtain the mixed population;

[0101] S402. Obtain the non-dominant frontier of a mixed population;

[0102] S403. Determine the population evolutionary state based on the non-dominant frontier of the mixed population.

[0103] In step S401, the mixed population includes multiple individuals, and some of the individuals in the mixed population come from the first... The candidate population obtained by executing step S3 in the current cycle process (the current cycle process) includes another part of the individuals from the first cycle. -1 round of loop process (previous round of loop process) added archive data (if) If =1, then the initial value of the archived data can be taken.

[0104] Among them, because the archived data is obtained by analyzing the 1st, 2nd... The data, such as gain flatness, was obtained from electromagnetic simulations of relevant individuals in the -1 cycle process. Therefore, the archived data actually contains multiple individuals, namely the 1st, 2nd, ... Individuals that have undergone electromagnetic simulation during the -1 round of the cycle can be selected from the archived data in step S401.

[0105] Specifically, in step S401, individuals belonging to the non-dominant front in the candidate population obtained from step S3 are used as part of the mixed population, and individuals belonging to the non-dominant front in the archived data are used as another part of the mixed population. These two parts of individuals constitute the mixed population.

[0106] In this embodiment, a non-dominated front refers to a front formed by a group of individuals closest to the ideal point (e.g., the one with the largest value of the corresponding objective function). For example, a non-dominated front in a candidate population is formed by a group of individuals in the candidate population that are closest to the ideal point of the candidate population; a non-dominated front in archived data is formed by a group of individuals in the archived data that are closest to the ideal point of the archived data.

[0107] The definition of "dominance" is as follows: taking a minimization problem as an example, if all objective functions of individual A are not greater than those of individual B, and at least one objective function of individual A is less than that of individual B, then individual A is said to dominate individual B. Similarly, the definition of non-dominance is that individual A has an objective function less than that of individual B, and individual B has an objective function less than that of individual A. In short, the non-dominant frontier in a population consists of individuals in that population who do not dominate each other.

[0108] After obtaining the mixed population in step S401, the non-dominated frontier of the mixed population is obtained in step S402. In step S403, the population evolution state is determined based on the non-dominated frontier of the mixed population.

[0109] Specifically, in step S403, the first number of individuals is obtained, which is the number of individuals from the candidate population in the non-dominant front of the mixed population; the second number of individuals is obtained, which is the total number of individuals in the non-dominant front of the mixed population; and then the ratio of the first number of individuals to the second number of individuals is calculated. ,Right now

[0110]

[0111] In this embodiment, a first threshold and a second threshold are set (where the first threshold is greater than the second threshold), and a comparison value is used. The relative size is used to determine the threshold. In this embodiment, the first threshold can be set to 0.75 and the second threshold can be set to 0.25.

[0112] If the ratio Larger, for example, ratio Greater than or equal to the first threshold, that is When the value is ≥0.75, the evolutionary state of the mixed population is in the early stage of evolution, that is, the population evolutionary state is determined to be in the early stage of evolution.

[0113] If the ratio Medium, such as ratio Less than the first threshold and greater than the second threshold, that is, 0.25 < When <0.75, the evolutionary state of the mixed population is in the late stage of evolution, that is, the population evolutionary state is determined to be in the late stage of evolution;

[0114] If the ratio Smaller, such as ratio Less than or equal to the second threshold, that is When the value is ≤0.25, the prediction relationship of the surrogate model is biased, that is, the population evolution state is determined to be a state of prediction bias.

[0115] In the During the cycle, when executing step S5, which is to determine the adaptive filling criterion based on the population evolutionary state, the following steps can be performed:

[0116] S501. When the population is in the early stage of evolution, the adaptive filling criterion is determined to be the population convergence improvement filling criterion.

[0117] S502. When the population is in the late stage of evolution, the adaptive filling criterion is determined to be the population diversity enhancement filling criterion.

[0118] S503. When the population evolutionary state is a prediction bias state, determine the adaptive filling criterion as the population accuracy enhancement filling criterion.

[0119] In step S501, when the population is in the early evolutionary stage, the adaptive filling criterion of "population convergence enhancement filling criterion" is executed. The population convergence enhancement filling criterion focuses on improving population convergence, thereby accelerating population convergence. Accordingly, step S601 is selected to be executed when executing step S6.

[0120] In step S502, when the population is in a late evolutionary stage, the adaptive filling criterion of "population diversity enhancement filling criterion" is executed. The population diversity enhancement filling criterion focuses on enhancing population diversity, thereby maintaining a uniform population distribution. Accordingly, step S602 is selected to be executed when performing step S6.

[0121] In step S503, when the population evolution state is in a prediction bias state, the adaptive filling criterion of "population accuracy enhancement filling criterion" is executed. The population accuracy enhancement filling criterion focuses on enhancing the prediction accuracy of the surrogate model. Accordingly, step S603 is selected to be executed when executing step S6.

[0122] In the During the cycle process, when executing step S6, which is the electromagnetic simulation using the adaptive filling criterion, the following steps can be performed:

[0123] If step S501 (the adaptive filling criterion is the population convergence improvement filling criterion) is performed: S601. Obtain the convergence improvement of each non-dominated individual in the candidate population with reference to the non-dominated individuals in the archive data added in the last cycle process, and perform electromagnetic simulation on the individual with the largest convergence improvement in the candidate population. The convergence improvement (CI) is an index for realizing fast convergence of the population, and the calculation method of the convergence improvement is as follows: for a candidate individual in the candidate population, find the individual (which can be referred to as an archive individual) in the archive data that has the smallest angle with the candidate individual, then project the vector of the candidate individual onto the vector of the archive individual, and subtract the length of the projection from the modulus of the vector of the archive individual, and the obtained value is the convergence improvement of the candidate individual. By performing electromagnetic simulation on the individual with the largest convergence improvement in the candidate population, the individual with the largest improvement in the direction of the vector of the archive individual can be selected for electromagnetic simulation, so that the purpose of fast convergence is achieved.

[0124] If step S502 (the adaptive filling criterion is the population diversity improvement filling criterion) is performed: S602. Obtain the angle between each non-dominated individual in the candidate population and the closest non-dominated individual in the archive data added in the last cycle process (the last cycle process), and perform electromagnetic simulation on the individual with the largest angle in the candidate population.

[0125] If step S503 (the adaptive filling criterion is the population accuracy enhancement filling criterion) is performed: S603. Obtain the sum of the uncertainties provided by all M surrogate models for each individual in the candidate population, and perform electromagnetic simulation on the individual with the largest sum of uncertainties in the candidate population.

[0126] In step S603, for any individual in the candidate population, the individual is provided with a corresponding uncertainty by each of the M surrogate models in step S1, so the sum of the corresponding uncertainties provided by the M surrogate models can be obtained.

[0127] The electromagnetic simulation of the individual in steps S601-S603 can also be referred to as the real function calculation of the individual. By performing electromagnetic simulation on an individual, the specific values of the design performance indicators of the antenna, such as gain flatness, input return loss, axial ratio, main polarization and cross polarization, can be obtained when the antenna has the specific structural parameters and material parameters represented by the individual.

[0128] After the first cycle process is performed, the archive data is updated, and the candidate population is updated. ​After step S6 in the round cycle process, the results of the electromagnetic simulation, i.e. the individuals subjected to the electromagnetic simulation and the correspondence between the individuals and the specific values of the design performance indexes such as the gain flatness obtained by the electromagnetic simulation, are added to the archive data.

[0129] After step S2-S6 in the round cycle process, it is determined whether the round cycle process is the last round cycle process according to the number of the round cycle process. After step S6 in the round cycle process, the archive data contains the individuals and the specific values of the design performance indexes obtained by the electromagnetic simulation in the first, second, …, round cycle process. -1、 The individuals and the specific values of the design performance indexes obtained by the electromagnetic simulation in the round cycle process are added to the archive data. +1, If the first +1round cycle process is executed, the archive data obtained in the first

[0130] round cycle process is used to further correct the prediction relationship of the surrogate model in the first +1round cycle process. Figure 1 After step S2-S6 in the round cycle process, it is determined whether the round cycle process is the last round cycle process according to the number of the round cycle process. If the round cycle process is the last round cycle process, the next round cycle process is not executed, and the first round cycle process is taken as the last round cycle process.

[0131] In the embodiment, a number threshold is set, and the round cycle process is ended when the total number of the real function calculation (electromagnetic simulation) reaches the number threshold . Since only one individual is subjected to the real function calculation in one round cycle process, the total number of the real function calculation (electromagnetic simulation) is equal to the total number of the round cycle process, and the round cycle process is ended when the total number of the round cycle process reaches the number threshold . If ≥ , it is determined that the round cycle process is ended, and if < , it is determined that the round cycle process is not ended.

[0132] If the sampling points are extracted by the Latin hypercube sampling method in the initialization, the number of the real function calculation in the process is also included in the total number of the real function calculation (electromagnetic simulation), and thus the total number of the real function calculation (electromagnetic simulation) is greater than the total number of the round cycle process. For example, if the number threshold For 300, 50 sampling points are extracted by Latin hypercube sampling method at initialization, which consumes 50 real calculation times, and the remaining real calculation times are 300-50=250. Since one cycle consumes one real calculation time, the total number of cycles is 250, which means that the cycle end condition is equivalent to "the total number of cycles executed reaches 250".

[0133] In this embodiment, the candidate population obtained in the last cycle contains at least one individual, which has specific values of design parameter variables such as structural parameters and material parameters. The individual in the candidate population obtained in the last cycle is output, that is, the specific values of the design parameter variables, which can be used for further simulation or experiment, or directly used to guide the design and manufacture of the antenna, for example, directly manufacturing an antenna with such specific structural parameters and material parameters.

[0134] The antenna optimization method based on adaptive filling criterion multi-objective evolution in this embodiment directly optimizes the design parameter variables based on the objective function, and the candidate population finally searched contains multiple potential individuals, which are parameter combination schemes with superior design indicators. The method has high automation degree and high efficiency of antenna optimization design. Compared with other proxy model assisted multi-objective evolution algorithms, the difference lies in that the latter integrates the convergence, diversity and accuracy of the proxy model into one index, and equally considers the three aspects, while the former first judges the evolution state of the population, and then executes the corresponding filling criterion, which saves the calculation resources and greatly improves the optimization efficiency. The adaptive filling criterion applied in the antenna optimization method based on adaptive filling criterion multi-objective evolution in this embodiment can first determine the maximum demand of the current population, and then execute the corresponding filling criterion, which can flexibly cope with problems with various characteristics and improve the utilization benefit of calculation resources. Figure 2

[0135] In this embodiment, it is specifically applied to the optimization design of a circularly polarized patch antenna, that is, the circularly polarized patch antenna is used as the antenna to be optimized in step S1. The structure and parameter definition of the circularly polarized patch antenna are as shown in Figure 3 The value range of the parameters is shown in Table 1. The target function selects the input return loss, axial ratio, main polarization and cross polarization at 2.38~2.42GHz, and the definition of the target function is known from formulas 3~6. The optimizer selects the multi-objective evolution algorithm guided by the reference vector, and the search algebra is 20.

[0136] Table 1 Range of parameter variables

[0137] ​

[0138] The cycle end condition is set to 300 real function calculation times, which is abbreviated as FGSAEA in this embodiment.

[0139] The entire automatic optimization process is expected to take 21 hours, Figure 4 and Figure 5 The input return loss and axial ratio of the circularly polarized patch antenna before and after optimization are shown, in which S11 is less than -10dB in the working frequency band, and AR is less than 3dB in the working frequency band. The performance of the antenna is far superior to that before optimization, which confirms the effectiveness of the method in this embodiment.

[0140] In addition, the method in this embodiment also compares the performance with other advanced intelligent algorithms on this embodiment, in which the algorithms include classification-based proxy model assisted evolutionary algorithm (CSEA), Kriging model assisted reference vector guided evolutionary algorithm (K-RVEA), Kriging model assisted two archive evolutionary algorithm (KTA2), multi-classifier assisted evolutionary algorithm (MCEA / D), and relationship prediction model assisted evolutionary algorithm (REMO). The optimization results are shown in Figure 6 The horizontal coordinate represents four objective functions, and the vertical coordinate represents the normalized results of each algorithm optimization. From the figure, it can be concluded that FGSAEA has achieved the minimum value on each objective function, which reflects its superior convergence. FGSAEA also achieves multiple values on each objective function, which reflects its superior diversity. At the same time, the number of individuals optimized by FGSAEA is much larger than that of other methods, which confirms the high efficiency of the method in this embodiment.

[0141] The computer program for executing the antenna optimization method based on adaptive filling criterion multi-objective evolution in this embodiment can be written into a computer device or a storage medium. When the computer program is read out and run, the antenna optimization method based on adaptive filling criterion multi-objective evolution in this embodiment is executed, thereby achieving the same technical effects as the antenna optimization method based on adaptive filling criterion multi-objective evolution in this embodiment.

[0142] It should be noted that, unless otherwise specified, when a feature is referred to as "fixed" or "connected" to another feature, it can be directly fixed or connected to the other feature, or indirectly fixed or connected to the other feature. Furthermore, the descriptions of "upper," "lower," "left," and "right" used in this disclosure are only relative to the relative positional relationships of the components of this disclosure in the accompanying drawings. The singular forms "a," "an," and "the" used in this disclosure are also intended to include the plural forms, unless the context clearly indicates otherwise. Moreover, unless otherwise defined, all technical and scientific terms used in this embodiment have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this embodiment specification is only for describing particular embodiments and is not intended to limit the invention. The term "and / or" as used in this embodiment includes any combination of one or more of the associated listed items.

[0143] It should be understood that although the terms first, second, third, etc., may be used to describe various elements in this disclosure, these elements should not be limited to these terms. These terms are only used to distinguish elements of the same type from each other. For example, a first element may also be referred to as a second element without departing from the scope of this disclosure, and similarly, a second element may also be referred to as a first element. The use of any and all instances or exemplary language (“e.g.,” “such as,” etc.) provided in this embodiment is intended only to better illustrate embodiments of the invention and, unless otherwise required, does not impose a limitation on the scope of the invention.

[0144] It should be recognized that embodiments of the present invention can be implemented or carried out by computer hardware, a combination of hardware and software, or by computer instructions stored in a non-transitory computer-readable storage medium. The method can be implemented using standard programming techniques—including a non-transitory computer-readable storage medium configured with a computer program, wherein such a storage medium causes the computer to operate in a specific and predefined manner—according to the methods and drawings described in the specific embodiments. Each program can be implemented in a high-level procedural or object-oriented programming language to communicate with the computer system. However, if desired, the program can be implemented in assembly or machine language. In any case, the language can be a compiled or interpreted language. Furthermore, for this purpose, the program can run on a programmed application-specific integrated circuit (ASIC).

[0145] Further, the operations of the processes described in this embodiment can be performed in any suitable order, unless otherwise indicated herein or otherwise clearly contradicted by context. The processes described in this embodiment (or variations and / or combinations thereof) can be implemented under the control of one or more computer systems configured with executable instructions (e.g., computer programs, one or more computer programs, or one or more applications) to perform the functionality described herein, and can be implemented completely or partially using hardware, or a combination thereof. Computer programs include machine readable instructions that can be executed by one or more processors.

[0146] Further, the methods can be implemented in any suitable type of computing platform operably connected to, including but not limited to, a personal computer, mini-computer, mainframe, workstation, networked or distributed computing environment, separate or integrated computer platforms, or in communication with charged particle tools or other imaging devices, and the like. Aspects of the present invention can be implemented in machine readable code stored on a non-transitory storage medium or device, whether removable or integrated into a computing platform, such as a hard disk, optical read and / or write storage media, RAM, ROM, and the like, such that it can be read by a programmable computer to configure and operate the computer to perform the processes described herein when the storage medium or device is read by the computer. Further, the machine readable code, or portions thereof, can be transmitted over wired or wireless networks. The present invention encompasses these and other different types of non-transitory computer readable storage media when the instructions or programs implementing the above steps are included in conjunction with a microprocessor or other data processor. The present invention also encompasses the computer itself when programmed in accordance with the methods and techniques of the present invention.

[0147] The computer programs are capable of applying to input data to perform the functions of the present embodiment, thereby transforming the input data to generate output data that is stored to non-volatile memory. The output information can also be applied to one or more output devices, such as a display. In preferred embodiments of the present invention, the transformed data represents a physical and tangible object, including a particular visual depiction of a physical and tangible object produced on a display.

[0148] The above merely preferred embodiments of the present invention, and the present invention is not limited to only the above-described embodiments, as long as the same means achieve the technical effects of the present invention, any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the scope of protection of the present invention. The technical solutions and / or embodiments of the present invention can have various modifications and changes within the scope of protection.

Claims

1. An antenna optimization method based on adaptive filling criteria and multi-objective evolution, characterized in that, The antenna optimization method based on adaptive filling criterion multi-objective evolution includes: Obtain the design parameter variables and objective function of the antenna to be optimized; the objective function represents the optimization objective of the design parameter variables. Execute at least one round of the loop process until the loop termination condition is met; each round of the loop process includes: At least one surrogate model is trained and updated; the surrogate model is used to fit the response relationship between the design parameter variables and the objective function. The candidate population is obtained by searching and optimizing in the prediction space; the prediction space is the surrogate model trained in this round of iteration, which predicts the population with the design parameter variables as individuals. Determine the population evolutionary state based on the candidate populations; Determine the adaptive filling criterion based on the population evolutionary state; Electromagnetic simulation is performed using the aforementioned adaptive filling criterion, and the results of the electromagnetic simulation are added to the archived data. The process of obtaining the design parameter variables and objective function of the antenna to be optimized includes: set up As the objective function; in This indicates that the antenna to be optimized is at an angle of θ Gain at that time, in dBi. The flatness of the gain within the angular range. Indicates frequency as The return loss of the antenna to be optimized is expressed in dB. This indicates the input return loss within the operating frequency band. S 11 The maximum value; Indicates frequency as The axial ratio of the antenna to be optimized, in dB; This represents the minimum principal polarization value of the 90° plane of the antenna to be optimized. This represents the minimum principal polarization value of the antenna at 0°, expressed in dB. This represents the maximum cross-polarization value of the 90° plane of the antenna to be optimized. This represents the maximum cross-polarization value of the 0° plane of the antenna to be optimized, in dB. for θ The minimum value, for θ The maximum value, for The minimum value, for The maximum value, Indicates the degree of circular polarization of the antenna. Indicates the degree of main polarization of the antenna. This indicates the degree of cross-polarization of the antenna.

2. The antenna optimization method based on adaptive filling criterion and multi-objective evolution according to claim 1, characterized in that, The training and updating of at least one proxy model includes: When the current cycle is the first cycle, the initial value of the archived data is used as the training data to train the proxy model; When the current iteration is an iteration other than the first iteration, the archived data added in the previous iteration is used as training data, and the proxy model trained in the previous iteration is trained.

3. The antenna optimization method based on adaptive filling criterion multi-objective evolution according to claim 1, characterized in that, Determining the population evolutionary state based on the candidate population includes: A mixed population is obtained; the mixed population is obtained by mixing the non-dominated fronts of the candidate population with the non-dominated fronts of the archived data; Obtain the non-dominant front of the mixed population; The evolutionary state of the population is determined based on the non-dominant frontier of the mixed population.

4. The antenna optimization method based on adaptive filling criterion and multi-objective evolution according to claim 3, characterized in that, Determining the evolutionary state of the population based on the non-dominant front of the mixed population includes: Based on the non-dominant front of the mixed population, obtain the first number of individuals and the second number of individuals; wherein, the first number of individuals is the number of individuals from the candidate population in the non-dominant front of the mixed population, and the second number of individuals is the total number of individuals in the non-dominant front of the mixed population; Obtain the ratio of the first number of individuals to the second number of individuals; When the ratio is greater than or equal to the first threshold, the population evolutionary state is determined to be in the early evolutionary stage; When the ratio is less than the first threshold and greater than the second threshold, the population evolutionary state is determined to be late evolutionary stage; When the ratio is less than or equal to the second threshold, the population evolutionary state is determined to be a prediction bias state.

5. The antenna optimization method based on adaptive filling criterion multi-objective evolution according to claim 4, characterized in that, The step of determining the adaptive filling criterion based on the population evolutionary state includes: When the population is in the early stage of evolution, the adaptive filling criterion is determined to be the population convergence improvement filling criterion. When the population is in the late stage of evolution, the adaptive filling criterion is determined to be the population diversity enhancement filling criterion. When the population evolutionary state is a prediction bias state, the adaptive filling criterion is determined to be the population accuracy enhancement filling criterion.

6. The antenna optimization method based on adaptive filling criterion and multi-objective evolution according to claim 5, characterized in that, The electromagnetic simulation based on the adaptive filling criterion includes: When the adaptive filling criterion is the population convergence improvement filling criterion, the non-dominant individuals in the archived data added in the previous round of iteration are used as a reference to obtain the convergence improvement of each non-dominant individual in the candidate population, and electromagnetic simulation is performed on the individual with the largest convergence improvement in the candidate population. When the adaptive filling criterion is the population diversity enhancement filling criterion, the angle between each non-dominant individual in the candidate population and the non-dominant individual closest to the one added in the archived data in the previous cycle is obtained, and electromagnetic simulation is performed on the individual with the largest angle in the candidate population. When the adaptive filling criterion is the population accuracy enhancement filling criterion, the sum of the uncertainties provided by all the proxy models for each individual in the candidate population is obtained, and electromagnetic simulation is performed on the individual with the largest sum of uncertainties in the candidate population.

7. The antenna optimization method based on adaptive filling criterion multi-objective evolution according to any one of claims 1-6, characterized in that, The loop terminates when the total number of calculations of the real function reaches a threshold.

8. A computer device, characterized in that, It includes a memory and a processor, the memory being used to store at least one program, and the processor being used to load at least one program to execute the antenna optimization method based on adaptive filling criteria and multi-objective evolution as described in any one of claims 1-7.

9. A computer-readable storage medium storing a processor-executable program, characterized in that, The processor-executable program, when executed by the processor, is used to perform the antenna optimization method based on adaptive filling criteria and multi-objective evolution as described in any one of claims 1-7.

Images