Antenna Design Method Based on Multi-Adaptive Agent Model-Assisted Evolutionary Algorithm

Abstract

This invention discloses an antenna design method based on a multi-adaptive surrogate model-assisted evolutionary algorithm. Within the framework of traditional surrogate-assisted evolutionary algorithms, this invention adaptively adjusts the selection probability of each mutation operator, constructs a separate surrogate model for each offspring individual, incorporates an offspring solution quality judgment operation, and utilizes dynamic incremental database calculations to determine the number of individuals for electromagnetic simulation. This invention better balances the algorithm's exploration and development capabilities, reduces the probability of getting trapped in local optima, decreases fitting time and the number of simulations with useless solutions, and further improves optimization speed. Simulation results show that the three antenna structures in the embodiments of this invention all achieved the desired antenna performance after optimization design using the multi-adaptive surrogate model-assisted evolutionary algorithm.

Description

Technical Field

[0001] This invention belongs to the field of radar communication technology, and further relates to an antenna design method based on a multi-adaptive surrogate model-assisted evolutionary algorithm in the field of antenna technology. This invention can be used for the optimization of the size parameters of various antennas. Background Technology

[0002] Due to the complexity of antenna structures, traditional electromagnetic analysis theory struggles to obtain near-globally optimal initial solutions when dealing with antenna structures of varying shapes. In this context, traditional parameter scanning methods are not only time-consuming and laborious but also fail to guarantee that the final antenna structure is globally optimal. Some traditional population-based algorithms based on random search, such as genetic algorithms, differential evolution algorithms, and particle swarm optimization, have demonstrated good optimization performance and high potential in such situations. However, these algorithms generate large-scale offspring populations, and performing full-wave simulations for each offspring solution slows down the optimization process. This is because electromagnetic simulations occupy a significant portion of the optimization algorithm's execution time; therefore, reducing the number of simulations for useless solutions becomes an effective improvement method to accelerate the optimization process. Surrogate model-assisted evolutionary algorithms (SMAs) use surrogate models to pre-screen the offspring population, retaining more promising individuals for full-wave simulations, greatly reducing the number of simulations for useless solutions. To further improve optimization speed, many methods have been incorporated into the surrogate model-assisted evolutionary algorithm optimization framework for antenna optimization to further enhance the algorithm's local optimization capabilities. However, this brings problems such as reduced global search capability and susceptibility to local optima.

[0003] Xi'an Jiaotong University disclosed an antenna optimization method based on multi-output Gaussian process Bayesian optimization in its patent application "An Antenna Optimization Method Based on Multi-Output Gaussian Process Bayesian Optimization" (Patent Application No. 202110438497.5, Publication No. CN 113076699 A). This method utilizes a small number of high-precision electromagnetic simulations and a large number of low-precision electromagnetic simulations to construct a surrogate model. Bayesian optimization is then applied to this surrogate model, significantly improving the antenna optimization speed. While replacing traditional evolutionary algorithms with Bayesian optimization improves optimization speed, this method still has shortcomings. For multi-peak antenna optimization problems, since Bayesian optimization optimizes a fitted model built from known data points, it heavily relies on the location of the sampling points and lacks global search randomness. Once the optimization result gets stuck in a local optimum, it is difficult to escape in a short time, leading to poor quality of the final result.

[0004] In his paper "An Efficient Method for Antenna Design Based on a Self-Adaptive Bayesian Neural Network-Assisted Global Optimization Technique" (IEEE Transactions on Antennas and Propagation, 2022), Yushi Liu disclosed a surrogate model optimization method based on a Bayesian neural network for global antenna optimization. This method employs an adaptive confidence lower bound approach to escape local optima, improving the algorithm's global search capability and preventing premature local optima. However, this method still has shortcomings. The increased global search capability leads to a decrease in local optimization capability. In the later stages of optimization, excessive consideration of global search capability results in a decrease in optimization speed. Therefore, the optimization speed advantage of this method compared to existing algorithms is not significant, and it cannot adaptively adjust the bias between global search capability and local optimization capability as the optimization process progresses.

[0005] Jiangsu University of Science and Technology disclosed an antenna design method based on a Kriging surrogate model and a multi-objective genetic algorithm in its patent application "An Antenna Design Method Based on Kriging Surrogate Model and Multi-Objective Genetic Algorithm" (Patent Application No. 202010250633.3, Publication No. CN 111460665A). This method uses particle swarm optimization to sample and construct a training set, uses it to train a Kriging model predictor, and uses the NSGA3 optimizer to perform optimization to obtain an optimized solution set, thus achieving global optimization of the antenna. However, the surrogate model training process of this method is computationally intensive and requires a large sample size, resulting in excessively long training time. Compared with traditional evolutionary algorithms, its time advantage is not significant. Summary of the Invention

[0006] The purpose of this invention is to address the technical problems existing in the prior art by providing an antenna design method based on a multi-adaptive surrogate model-assisted evolutionary algorithm. This method aims to solve the problems that global optimization algorithms are prone to getting stuck in local optima in the early stages of optimization, the global search capability and local optimization capability are difficult to adaptively adjust, the surrogate model training time is too long, and the time advantage compared with traditional evolutionary algorithms is not obvious.

[0007] The underlying principle of this invention is as follows: Addressing the aforementioned problems, this invention adaptively adjusts the selection probability of each mutation operator, providing a suitable mutation scheme for different optimization stages. Since each mutation operator has different tendencies in global search and local optimization, adaptively adjusting the mutation operator can solve the problem of the algorithm easily getting trapped in local optima too early, and the difficulty in adaptively adjusting global search and local optimization capabilities. In this invention, each offspring individual is constructed with a separate surrogate model, reducing the number of sampling points used in each fitting iteration and solving the problem of excessively long surrogate model training time. The offspring solution quality assessment operation can determine the quality of the optimal offspring solution, generating a higher-quality offspring population, further reducing the number of simulations of useless solutions, and solving the problem of the algorithm's lack of significant time advantage compared to traditional evolutionary algorithms. Utilizing dynamic incremental database calculations to determine the number of individuals in electromagnetic simulations, more individuals are added to the database in the early stages of optimization, increasing the uncertainty of each iteration and enhancing the algorithm's global search capability. In the later stages of optimization, the optimal individuals of each generation are retained, improving the optimization speed. This operation further solves the problem of the algorithm easily getting trapped in local optima too early, and the difficulty in effectively adjusting global search and local optimization capabilities. Using differential evolution as the core optimization algorithm, a surrogate model is used to reduce the number of useless solution simulations. Operations such as adaptively adjusting the selection probability of each mutation operator, constructing a surrogate model for each offspring individual, judging the quality of offspring solutions, and dynamically incrementing the database to determine the number of individuals for electromagnetic simulation are added. These operations adaptively adjust the global search capability and local optimization capability of the algorithm, reduce fitting time, and further improve optimization efficiency.

[0008] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0009] Step 1: Construct a parameter database based on the value range of the parameter to be optimized;

[0010] Step 2: Adaptively adjust the probability of each mutation operator being selected, select one mutation operator, perform the mutation and crossover operation of the differential evolution algorithm to generate offspring to be evaluated; construct surrogate models for each offspring individual to obtain the comprehensive performance prediction value, confidence lower bound value and average fitting distance value of each offspring individual.

[0011] Step 3: Using the offspring solution quality judgment mechanism, determine whether the lowest confidence lower bound value among all offspring individuals to be evaluated in the current iteration is less than the minimum comprehensive performance value of all samples in the database. If yes, proceed to step 4; otherwise, proceed to step 2.

[0012] Step 4, Find the optimal sample:

[0013] The confidence lower bound and overall performance values ​​of all individuals are sorted separately, resulting in two corresponding sorts for all individuals. One sort is randomly selected, and electromagnetic simulations are performed on the top N individuals in the selected sort. Each simulated individual and its corresponding overall performance value are added to the database. The sample X corresponding to the best overall performance is selected from the updated parameter database. best As the optimal sample for the current iteration, the value of N is equal to the dynamic increment of the database during the current iteration;

[0014] Step 5: Determine whether the current iteration meets the termination condition. If yes, proceed to step 6 after obtaining the optimal comprehensive performance; otherwise, proceed to step 2.

[0015] Step 6, assign the optimal overall performance to sample X. best The parameters are loaded into the antenna to be designed, and the antenna design is completed.

[0016] Compared with the prior art, the present invention has the following advantages:

[0017] First, because this invention adaptively adjusts the probability of each mutation operator being selected, it adaptively selects the appropriate mutation operator for different optimization problems and different optimization stages, better balancing the algorithm's exploration and development capabilities, and enabling the algorithm to have better performance when facing different problems and different optimization stages.

[0018] Second, this invention constructs a surrogate model for each offspring individual, using the nearest data points to each prediction point to construct the surrogate model, which reduces the number of data points required to fit each prediction point, and reduces the fitting time without significantly reducing the fitting accuracy.

[0019] Third, this invention incorporates a offspring solution quality assessment mechanism. This mechanism judges the quality of the optimal offspring individual; if the quality does not meet the requirements, the mutation and crossover operation of the differential evolution algorithm is re-executed until an offspring individual that meets the quality requirements is generated. This pre-judgment improves the quality of the offspring solution, significantly reduces the number of simulations for useless solutions, and accelerates the optimization process.

[0020] Fourth, this invention utilizes dynamic incremental database calculations to determine the number of individuals in the electromagnetic simulation. In the early stages of optimization, this increases the diversity of solutions, reduces the probability of getting trapped in local optima, and simultaneously increases the probability of retaining individuals in the offspring population generated by each mutation and crossover whose overall performance is lower than the best sample in the current database. In the later stages of optimization, this reduces the number of simulations with useless solutions, improves the optimization speed, and better balances the algorithm's global exploration and local exploitation capabilities. Attached Figure Description

[0021] Figure 1 This is a flowchart of the present invention;

[0022] Figure 2A schematic diagram of an ultra-wideband antenna optimized using the algorithm of this invention;

[0023] Figure 3 To optimize the design using the algorithm of this invention, the frequency response diagram of the reflection coefficient modulus of an ultra-wideband antenna structure is obtained;

[0024] Figure 4 A schematic diagram of a circularly polarized antenna designed using the algorithm of this invention;

[0025] Figure 5 This is a performance diagram of a circularly polarized antenna structure optimized using the algorithm of this invention; wherein, Figure 5 (a) is the frequency response diagram of the reflection coefficient magnitude of a circularly polarized antenna; Figure 5 (b) is the axial ratio frequency response diagram of the circularly polarized antenna;

[0026] Figure 6 A schematic diagram of a filter antenna optimized using the algorithm of this invention;

[0027] Figure 7 This is a performance diagram of a filter antenna structure optimized using the algorithm of this invention; wherein, Figure 7 (a) is the frequency response diagram of the reflection coefficient magnitude of the filtered antenna; Figure 7 (b) is the gain-frequency response diagram of the filtered antenna. Detailed Implementation

[0028] To make the technical problems solved by the present invention, the technical solutions, and the beneficial effects clearer, the following specific embodiments provide a further detailed description of the present invention. It should be understood that the specific embodiments described herein are merely illustrative and are not intended to limit the scope of the invention.

[0029] Embodiment 1 of the present invention is an ultra-wideband antenna. Figure 2 This is a schematic diagram of the antenna structure. Figure 2 The white area represents the dielectric substrate, while the gray area represents the square radiating patch, microstrip feed line, and ground plane. The white rectangles on the square radiating patch and ground plane are rectangular slots.

[0030] Embodiment 2 of the present invention is a circularly polarized antenna. Figure 4 This is a schematic diagram of the antenna structure. Figure 2 The outer square is a dielectric substrate, and the inner small square is a dielectric resonator. One cylindrical metal block, two short rectangular metal blocks, and two long rectangular metal blocks constitute the radiator of the antenna.

[0031] Embodiment 3 of the present invention is a filter antenna. Figure 6 This is a schematic diagram of the antenna structure. Figure 6The light gray area represents the dielectric substrate, while the dark gray area represents the metal radiating patch and the ground plane. The upper layer is the metal radiating patch, and the lower layer is the metal ground plane. The gray area inside the metal radiating patch on the left is a filtering structure achieved by slotting the radiating patch. The white circles are cylindrical metal through holes.

[0032] Reference Figure 1 The steps of the novel antenna optimization method based on Gaussian stochastic process and differential evolution algorithm provided by the present invention are further described below.

[0033] Step 1: Based on antenna design experience, define each dimensional parameter and its corresponding range for the antenna to be optimized. Perform Latin hypercube sampling within the range of each parameter to obtain the set of sampling points for that parameter. Construct a current parameter database containing individual parameters and overall performance. Based on antenna design experience, define each dimensional parameter and its corresponding range (in mm) for the antenna to be optimized.

[0034] In Embodiment 1 of the present invention, the range of optimized parameters for the antenna to be designed is set as follows:

[0035] 0≤L f ≤8

[0036] 2≤L gnd ≤6

[0037] 0≤L p ≤10

[0038] 0≤L s1 ≤2

[0039] 0≤L s2 ≤1

[0040] 16≤L sub ≤20

[0041] 8≤W≤12

[0042] 0≤W1≤8

[0043] 1≤W f ≤3

[0044] 0≤W p ≤1

[0045] 0≤W s1 ≤8

[0046] 0≤W s2 ≤8

[0047] 10≤W sub ≤15

[0048] Where W represents the side length of the square patch, L pW represents the length of the rectangular slot on the square patch. p W represents the width of the rectangular slot on the square patch. f L represents the width of the microstrip feed line. f W represents the distance between the square patch and the dielectric substrate, and W1 represents the distance between the two rectangular slots on the square patch. sub L represents the width of the dielectric substrate. sub W represents the length of the dielectric substrate. s1 L represents the width of the vertical rectangular groove on the floor plane. s1 W represents the length of the vertical rectangular groove on the floor plane. s1 L represents the length of the horizontal rectangular groove on the floor plane. s1 L represents the width of the horizontal rectangular groove on the floor plane. gnd This indicates the width of the floor surface.

[0049] In Embodiment 2 of the present invention, the range of optimized parameters for the antenna to be designed is set as follows:

[0050] 3≤L1≤4

[0051] 8≤L2≤10

[0052] 1≤W≤2

[0053] Where L1 represents the length of the short rectangular block, L2 represents the length of the long rectangular block, and W represents the thickness of the annular block. In Embodiment 3 of the present invention, the optimized parameter ranges for the antenna to be designed are set as follows:

[0054] 7≤m l ≤8

[0055] 1≤m w ≤3

[0056] 10≤f≤14

[0057] 32≤L p ≤35

[0058] 35≤W p ≤38

[0059] Where m1 represents the length of the rectangular slot on the left rectangular patch, m w L represents the width of the rectangular slot on the left rectangular patch, f represents the distance from the rectangular slot on the left rectangular patch to the edge of the patch, and L represents the width of the rectangular slot on the left rectangular patch. p W represents the width of the rectangular patch on the right. p This indicates the length of the rectangular patch on the right.

[0060] Latin hypercube sampling is performed within the range of each parameter to obtain the set of sampling points for that parameter. An unselected sampling point is randomly selected from the set of sampling points for each parameter as a sample. Electromagnetic simulation is performed on each sample to obtain the sample's gain performance, reflection coefficient modulus performance, and axial ratio performance.

[0061] The overall performance of each sample is calculated using the following formula;

[0062] F i =α 1,i (F gn,i )+α 2,i (F S11,i )+α 3,i (F AR,i )

[0063] Among them, F i α represents the overall performance of the i-th sample. 1,i α represents the penalty factor for the gain of the i-th sample. 2,i α represents the penalty factor for the magnitude of the reflectance coefficient of the i-th sample. 3,i α represents the penalty factor for the axis ratio of the i-th sample. 1,i α 2,i α 3,i Selection is made manually based on the designer's experience;

[0064] In Embodiment 1 of the present invention, the overall performance of the i-th sample is:

[0065] F i =F S11,i

[0066] F |S11|,i =max(|S 11 | i +15, 0), 3.1GHz≤f≤10.6GHz

[0067] Among them, |S 11 | i This represents the magnitude of the reflection coefficient of the i-th sample obtained after electromagnetic simulation.

[0068] In Embodiment 2 of the present invention, the overall performance of the i-th sample is:

[0069] F i =F S11,i +5×F AR,i

[0070] F S11,i =max(|S 11 | i +20, 0), 2.3GHz≤f≤2.8GHz

[0071] F AR,i =max(AR) i -3, 0), 2.3GHz≤f≤2.8GHz

[0072] Among them, |S 11 | i AR represents the magnitude of the reflection coefficient of the i-th sample obtained after electromagnetic simulation. i This represents the magnitude of the reflection coefficient of the i-th sample obtained after electromagnetic simulation.

[0073] In Embodiment 3 of the present invention, the overall performance of the i-th sample is

[0074] F i =F S11,i +5×F gn,i

[0075] F S11,i =max(|S 11 | i +10, 0), 2.7GHz≤f≤2.85GHz

[0076] F gn,i =-min(gain i -5, 0), 2.7GHz≤f≤2.85GHz

[0077] Among them, |S 11 | i Gain represents the magnitude of the reflection coefficient of the i-th sample obtained after electromagnetic simulation. i This represents the gain of the i-th sample obtained after electromagnetic simulation.

[0078] A parameter database is created by combining the parameters of all samples with their corresponding overall performance.

[0079] Step 2: In this embodiment of the invention, 50 iterations are performed. In the first 50 iterations, the values ​​of P(1), P(2), and P(3) are set to 1 / 3. The number of individuals NS(i) in the offspring population generated using each operator whose predicted overall performance is less than the minimum overall performance of all samples in the database is updated, and the total number of offspring individuals Nu(i) generated using each operator is updated. After 50 iterations, the probability P(i) of each of the three mutation operators being selected is calculated.

[0080]

[0081] Where i represents the index number corresponding to the three operators, i = 1, 2, 3, and the probabilities of selecting the three mutation operators DE / rand / 1, DE / currentto best / 1 and DE / best / 1 are set to P(1), P(2), and P(3), respectively.

[0082] The proportions of the three operators DE / rand / 1, DE / current to best / 1, and DE / best / 1 in the roulette are set as the probability of each operator being selected. The roulette wheel is used to randomly select a mutation operator, and the mutation and crossover operation of the differential evolution algorithm is performed to generate offspring to be evaluated.

[0083] Sort all samples in the parameter database according to their comprehensive performance value from smallest to largest. The top M individuals form the parent population, where M is 5 to 10 times the total number of parameters to be optimized. Perform mutation and crossover operations using the differential evolution algorithm to generate offspring for evaluation.

[0084] In Embodiment 1 of the present invention, M is set to 65; in Embodiment 2 of the present invention, M is set to 15; and in Embodiment 3 of the present invention, M is set to 30.

[0085] Using the T nearest sample points to each offspring in the parameter database, a surrogate model is constructed separately for each offspring to obtain the comprehensive performance prediction value, confidence lower bound, and average fitting distance value for each offspring.

[0086] In Embodiment 1 of the present invention, T is set to 13; in Embodiment 2 of the present invention, T is set to 3; and in Embodiment 3 of the present invention, T is set to 5.

[0087] Step 3: Using the offspring solution quality judgment mechanism, determine whether the lowest confidence lower bound value among all offspring individuals to be evaluated in the current iteration is less than the minimum comprehensive performance value of all samples in the database. If yes, proceed to step 4; otherwise, proceed to step 2.

[0088] Step 4: Sort all individuals by confidence lower bound and overall performance value in ascending order, obtaining the ranking of all individuals corresponding to the confidence lower bound and overall performance value rankings, respectively. Randomly select one ranking and perform electromagnetic simulation on the top N individuals in the selected ranking. Add each simulated individual and its corresponding overall performance value to the database. Select the sample X corresponding to the optimal overall performance from the updated parameter database. bestAs the optimal sample for the current iteration, N is equal to the dynamic increment of the database at the time of the current iteration. The dynamic increment of the database refers to setting the dynamic increment to 0, and then, based on the sorted results, sequentially determining whether the average fitting distance of each individual is less than a distance limit and incrementing the dynamic increment value by 1, until the average fitting distance is less than the limit distance or the dynamic increment value reaches a set upper limit. The limit distance is 0.5 to 2 times the parameter to be optimized, and the upper limit is set to 1 to 2 times the parameter to be optimized.

[0089] In Embodiment 1 of the present invention, the defined distance is 12, and the maximum number is set to 14.

[0090] In Embodiment 2 of the present invention, the defined distance is 3, and the maximum number is set to 3.

[0091] In Embodiment 3 of the present invention, the defined distance is 5, and the maximum number is set to 5.

[0092] Electromagnetic simulations are performed on the top N individuals after sorting to obtain their comprehensive performance values. The parameters and comprehensive performance values ​​of each individual are then added to the current parameter database. The sample X corresponding to the best comprehensive performance is selected from the updated parameter database. best As the optimal sample for the current iteration, the value of N is equal to the dynamic increment value of the database.

[0093] Step 5: Determine if the current iteration meets the termination condition. If yes, proceed to step 6 after obtaining the optimal synthesis performance; otherwise, proceed to step 2. In these three instances, the termination condition is the satisfaction of one of the following conditions:

[0094] Condition 1: The overall performance value of the best sample in the current database remains unchanged after 50 consecutive generations of optimization;

[0095] Condition 2: The overall performance value of the best sample in the database is less than 0.

[0096] Step 6, assign the optimal overall performance to sample X. best The parameters are loaded into the antenna to be designed, and the antenna design is completed.

[0097] The antenna optimization method based on the multi-adaptive surrogate model-assisted evolutionary algorithm proposed in this invention is used to optimize the antenna. Figure 2 The optimized ultra-wideband antenna structure parameters shown are finally obtained as L. f =5.1, L gnd =2.1, L p =2.7, L s1 =1.6, L s2 =0.5, L sub=16.8, W=8.7, W1=4.7, W f =1.8, W p =0.6, W s1 =2.1, W s2 =4.2, W sub =11.

[0098] Figure 3 To design and obtain the frequency response diagram of the reflection coefficient magnitude of a circularly polarized antenna structure using the algorithm described in this invention, Figure 3 The horizontal axis represents frequency, and the vertical axis represents the magnitude of the antenna's reflection coefficient at each frequency. Figure 3 As can be seen, by using a multi-adaptive surrogate model to assist the evolutionary algorithm in optimizing the parameters of the ultra-wideband antenna given in Case 1, the reflection coefficient modulus of the structure corresponding to the optimal parameters is less than -15dB in the frequency range of 3.1-10.6GHz, which meets the design requirements.

[0099] The antenna optimization method based on the multi-adaptive surrogate model-assisted evolutionary algorithm proposed in this invention is used to optimize the antenna. Figure 4 The optimized circularly polarized antenna structure parameters are as follows: L1 = 3.8, L2 = 9, W = 1.5.

[0100] Figure 5 A performance diagram of a circularly polarized antenna structure designed using the algorithm described in this invention is shown, wherein... Figure 5 (a) is a frequency response diagram of the reflection coefficient magnitude of a circularly polarized antenna. The horizontal axis represents the frequency, and the vertical axis represents the reflection coefficient magnitude of the antenna at each frequency point. Figure 5 (b) shows the axial ratio frequency response of the circularly polarized antenna, where the horizontal axis represents frequency and the vertical axis represents the axial ratio of the antenna at each frequency. The parameters of the circularly polarized antenna given in Embodiment 2 of this invention are optimized using a multi-adaptive surrogate model-assisted evolutionary algorithm. Figure 5 (a) It can be seen that the reflection coefficient modulus of the structure corresponding to the optimal parameters is less than -15dB in the frequency range of 2.3-2.8GHz. Figure 5 (b) It can be seen that the shaft ratio of the structure corresponding to the optimal parameters is less than 3dB in the frequency range of 2.3-2.8GHz, which meets the design requirements.

[0101] The antenna optimization method based on the multi-adaptive surrogate model-assisted evolutionary algorithm proposed in this invention is used to optimize the antenna. Figure 6 The optimized filter antenna structure parameters shown result in the final structure parameter m. l =7.5, m w =2, f=12, L p =34, W p =36.

[0102] Figure 7The diagram illustrates the performance of the filter antenna structure designed using the algorithm described in this invention. Figure 7 (a) is a frequency response diagram of the reflection coefficient magnitude of the filtered antenna. The horizontal axis represents the frequency, and the vertical axis represents the reflection coefficient magnitude of the antenna at each frequency point. Figure 7 (b) shows the gain-frequency response of the filtered antenna, with the horizontal axis representing frequency and the vertical axis representing the antenna gain at each frequency point along the Z-axis. The parameters of the filtered antenna given in Embodiment 3 of this invention are optimized using a multi-adaptive surrogate model-assisted evolutionary algorithm. Figure 7 (a) It can be seen that the reflection coefficient modulus of the structure corresponding to the optimal parameters is less than -10dB in the frequency range of 2.71-2.84GHz. Figure 7 (b) It can be seen that the gain of the structure corresponding to the optimal parameters is greater than 5 dBi in the frequency range of 2.7-2.85 GHz, which meets the design requirements.

[0103] The above embodiments are merely one of the implementation methods for achieving the technical solution of the present invention. The scope of protection claimed by the present invention is not limited to this embodiment, but also includes any variations, substitutions and other implementation methods that can be easily conceived by those skilled in the art within the scope of the technology disclosed in the present invention.

Claims

1. An antenna design method based on a multi-adaptive surrogate model-assisted evolutionary algorithm, characterized in that, The method adaptively adjusts the selection probability of each mutation operator, constructs a separate surrogate model for each offspring individual, implements an offspring solution quality judgment mechanism, and uses dynamic incremental determination of the number of individuals for electromagnetic simulation using a database; the steps of this method include the following: Step 1: Construct a parameter database based on the range of optimized parameter values ​​for the antenna to be designed; Step 2: Adaptively adjust the probability of each mutation operator being selected, select a mutation operator, perform the mutation and crossover operation of the differential evolution algorithm, and generate offspring to be evaluated; By using the nearest data points to each predicted point, a separate surrogate model is constructed for each offspring individual, and the comprehensive performance prediction value, confidence lower bound, and average fitting distance value of each offspring individual are obtained. Step 3: Using the offspring solution quality judgment mechanism, determine whether the lowest confidence lower bound value among all offspring individuals to be evaluated in the current iteration is less than the minimum comprehensive performance value of all samples in the database. If yes, proceed to step 4; otherwise, proceed to step 2. Step 4, Find the optimal sample: The confidence lower bound and overall performance values ​​of all individuals are sorted separately, resulting in two corresponding sorts for all individuals. One sort is randomly selected, and electromagnetic simulations are performed on the top N individuals in the selected sort. Each simulated individual and its corresponding overall performance value are added to the database. The sample X corresponding to the best overall performance is selected from the updated parameter database. best As the optimal sample for the current iteration, the value of N is equal to the dynamic increment of the database during the current iteration; Step 5: Determine whether the current iteration meets the termination condition. If yes, proceed to step 6 after obtaining the optimal comprehensive performance; otherwise, proceed to step 2. Step 6, assign the optimal overall performance to sample X. best The parameters are loaded into the antenna to be designed, and the antenna design is completed.

2. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, The range of optimized parameters for the antenna to be designed mentioned in step 1 refers to setting each size parameter and its corresponding range for the antenna to be optimized based on antenna design experience, and performing Latin hypercube sampling within the range of each parameter to obtain the set of sampling points for that parameter.

3. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, The steps for constructing the parameter database described in step 1 are as follows: The first step is to randomly select an unselected sampling point from the sampling point set for each parameter as a sample; The second step is to perform electromagnetic simulation on each sample to obtain the sample gain performance, sample reflection coefficient magnitude performance, and sample axial ratio performance. The third step is to calculate the overall performance of each sample using the following formula: F i =a 1,i (F gn,i )+a 2,i (F S11,i )+a 3,i (F AR,i ) Among them, F i α represents the overall performance of the i-th sample. 1,i α represents the penalty factor for the gain of the i-th sample. 2,i α represents the penalty factor for the magnitude of the reflectance coefficient of the i-th sample. 3,i This represents the penalty factor for the axis ratio of the i-th sample; The fourth step is to compile a parameter database by combining the parameters of all samples with their corresponding comprehensive performance.

4. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, The adaptive adjustment of the selection probability of each mutation operator in step 2 refers to setting the values ​​of P(1), P(2), and P(3) to 1 / 3; updating the number of individuals NS(i) in the offspring population generated by each operator whose overall performance prediction value is less than the minimum overall performance value of all samples in the database; updating the total number of offspring individuals Nu(i) generated using each operator; and calculating the selection probability P(i) of the three mutation operators after the iteration terminates. Where i represents the index number corresponding to the three operators, i = 1, 2, 3, and DE / rand / 1, DE / current tobest / 1 and DE / best / 1 are the probabilities of the three mutation operators being selected, respectively, P(1), P(2), P(3).

5. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 4, characterized in that, Step 2, which involves selecting a mutation operator, refers to setting the proportions of the three operators DE / rand / 1, DE / current to best / 1, and DE / best / 1 in the roulette wheel to the probability of each operator being selected, and then performing a roulette wheel gamble to randomly select a mutation operator.

6. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, Step 2, which involves generating the offspring population to be evaluated, means: sorting all samples in the parameter database according to their comprehensive performance values ​​from smallest to largest, and forming the parent population from the top M individuals, where M is 5 to 10 times the total number of parameters to be optimized; and performing the mutation and crossover operations of the differential evolution algorithm to generate the offspring to be evaluated.

7. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, Step 2, which involves constructing a separate surrogate model for each offspring individual, refers to using the T sample points closest to each offspring individual in the parameter database to construct a separate surrogate model for each offspring individual, thereby obtaining the comprehensive performance prediction value, confidence lower bound, and mean distance from the fitted point for each offspring individual. The T is set to be 1 to 2 times the parameter to be optimized.

8. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, The sorting of confidence lower bound and overall performance value for all individuals in step 4 refers to sorting the confidence lower bound and overall performance value for all individuals in ascending order, respectively, to obtain the sorting of all individuals corresponding to the confidence lower bound sorting and the sorting of all individuals corresponding to the overall performance value sorting.

9. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, The dynamic increment of the database mentioned in step 4 refers to setting the dynamic increment of the database to 0, and judging whether the average fitting distance of each individual is less than the distance limit according to the sorted results, and incrementing the dynamic increment value by 1, until the average fitting distance is less than the limit distance or the dynamic increment value of the database reaches the set upper limit, thus obtaining the dynamic increment of the database at the current iteration; the limit distance is 0.5 to 2 times the parameter to be optimized, and the upper limit of the number is set to 1 to 2 times the parameter to be optimized.

10. The antenna design method based on the multi-adaptive surrogate model-assisted evolutionary algorithm according to claim 1, characterized in that, The termination condition mentioned in step 5 is a case where one of the following conditions is met: Condition 1: The overall performance value of the best sample in the current database remains unchanged after 50 consecutive generations of optimization; Condition 2: The overall performance value of the best sample in the database is less than 0.

Images