A design optimization system and design method of broadband transparent absorber based on multi-objective genetic algorithm

Abstract

The present invention discloses a design optimization system and a design method of a broadband transparent absorber based on a multi-objective genetic algorithm. The multi-objective genetic algorithm module calls other modules to realize the rapid design of the transparent absorber; the electromagnetic simulation software module calculates the true value of the primary population; the neural network regression module calculates the true value of the same type of parameters; and the database module stores all the true values ​​obtained by simulating with the electromagnetic simulation software module. It is used to solve the problem of optimizing multiple objectives simultaneously to obtain balanced optimization results of each parameter.

Description

Technical Field

[0001] The invention belongs to the field of microwave-band metasurface absorbers, and in particular relates to a design optimization system and a design method of a broadband transparent absorber based on a multi-objective genetic algorithm. Background Art

[0002] Microwave absorbers can convert incident electromagnetic waves into impedance heat, thereby absorbing microwaves. However, detection methods based on microwave radars require receiving return signals to detect targets, and covering important targets with absorbing materials will greatly reduce the return signal amplitude, thereby achieving stealth requirements. Based on the characteristic of electromagnetic wave absorption, transparent absorbers also have a wide range of uses in the civilian field. For example, optical windows in microwave darkrooms, radio frequency systems, and scenes such as preventing electromagnetic radiation near communication base stations all have great demand for absorbers.

[0003] The use of resistive film layers to control the absorption of electromagnetic waves has a long history. Salisbury screens and Jaumann screens based on uniform film layers can achieve perfect absorption in narrow bands. However, uniform film absorbers have no practical application value due to their narrow frequency bands. The application of metasurfaces in the design of absorbers has greatly broadened the broadband performance of absorbers, thereby achieving strong absorption over a wide frequency band.

[0004] Metasurface refers to a multi-layered artificial structure that can modulate the behavior of electromagnetic waves. It has a sub-wavelength scale pattern on the horizontal plane. In electromagnetic theory, metasurfaces use constitutive parameters in three-dimensional space through specific boundary conditions to modulate the behavior of electromagnetic waves. Electromagnetic waves incident on metasurface absorbers often pass through the resonator layer and the reflective floor layer in turn, and the electromagnetic waves are absorbed through the heat loss of the resonant pattern layer.

[0005] The key to the absorption performance of the absorber over a wide frequency lies in the absorption potential of the proposed resonator layer over a wide frequency. The existing analog circuit method and equivalent medium method can design and optimize simple pattern resonators. However, the broadband absorption performance and high grazing angle incidence performance of simple pattern resonators are often poor. Complex pattern resonators have better performance in these two aspects. The existing way to propose complex resonators is to optimize the random parameters of the algorithm and call the simulation software for simulation. This process has extremely high requirements on computing power and time consumption. Therefore, there is an urgent need to design fast broadband absorbers with low computing power requirements. It is also urgent to optimize multiple targets at the same time and obtain balanced optimization results of each parameter. Summary of the invention

[0006] The present invention provides a design optimization system and a design method of a broadband transparent absorber based on a multi-objective genetic algorithm, which are used to solve the problem of optimizing multiple objectives simultaneously to obtain balanced optimization results of various parameters.

[0007] The present invention is achieved through the following technical solutions:

[0008] A design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm, the design optimization system comprising a multi-objective genetic algorithm module, an electromagnetic simulation software module, a neural network regression module and a database module;

[0009] The multi-objective genetic algorithm module calls other modules to realize the rapid design of transparent absorbers;

[0010] The electromagnetic simulation software module calculates the performance of the primary population;

[0011] The neural network regression module predicts the objective function value of the same type of population;

[0012] The database module stores all real values ​​obtained by simulation using the electromagnetic simulation software module.

[0013] A design method for a design optimization system of a broadband transparent absorber based on a multi-objective genetic algorithm, the design optimization method comprising the following steps:

[0014] Step 1: Use the multi-objective genetic algorithm module to establish the absorber population;

[0015] Step 2: Calculate the performance of the absorber population in step 1 or step 3. If there are identical individual results in the database module, proceed to step 4. If the prediction accuracy of the neural network module for the previous generation is higher than 60%, randomly select some individuals and proceed to step 5, and the remaining individuals proceed to step 6. If neither of the above is satisfied, proceed to step 6, and input the results obtained in step 6 into the database module.

[0016] Step 3: Determine whether the performance calculated in step 2 meets the requirements. If so, stop the optimization and output the optimized absorber; if not, generate a new population individual and repeat step 2.

[0017] Step 4: Call the real value in the database module;

[0018] Step 5: Call the neural network regression module to predict individual performance;

[0019] Step 6: Call the electromagnetic simulation software module to calculate individual performance;

[0020] Furthermore, the multi-objective genetic algorithm in step 1 can obtain the Pareto optimal solutions of multiple evaluation indicators, and the performance of the absorber population is evaluated using the obtained Pareto optimal solutions of the multiple evaluation indicators.

[0021] Furthermore, the establishment of the absorber population in step 1 is divided into establishing an initial population and generating a sub-population; the initial population is established by randomly generated parameters; the sub-population is generated based on the evaluation results of the previous population, using non-dominated sorting values ​​and crowding, selection, crossover and mutation operations.

[0022] Furthermore, the electromagnetic simulation software model in step 6 calculates the true value of the primary population and calculates the true value of part of the offspring population.

[0023] Furthermore, the electromagnetic simulation software model calculation in step 6 is specifically to define the relative bandwidth corresponding to the frequency range in which the absorption rate is greater than 0.9 or 0.95 as the calculation range of the electromagnetic simulation software model;

[0024] The formula for absorbance is defined as:

[0025] A=1-|S 11 | 2 -|S 21 | 2

[0026] Where S11 is the reflectivity and S21 is the transmittance;

[0027] The relative bandwidth formula is defined as:

[0028] ffoc=2×(fH-fL) / (fH+fL)

[0029] Where ffoc represents the relative bandwidth, fH represents the highest frequency of the entire frequency band with an absorption rate greater than 0.9, and fL represents the lowest frequency of the entire frequency band with an absorption rate greater than 0.9.

[0030] Furthermore, the neural network regression module of step 5 learns the data in the database, establishes the relationship between the parameters and the results, and calculates the true value of the offspring population;

[0031] When deciding whether to use the neural network to evaluate the nth generation and n>3, the original data in the database module and the true value data of the previous n-2 generations are needed to predict the accuracy of the n-1 generation; when the accuracy of the n-1 generation is greater than 60%, 50% of the random individuals of the nth generation are predicted using the neural network, and the rest are calculated using the electromagnetic simulation software module; where accuracy = 1-|predicted value-true value| / true value.

[0032] Furthermore, the neural network regression module is calculated specifically as follows: the neural network regression is expressed as the function ffoc&Thickness=ANN(Par). Since the relationship between the input parameters and the absorption relative bandwidth and thickness is established, the new parameters are input into the relationship to obtain the predicted value of the relative bandwidth corresponding to the parameters; a multi-layer fully connected neural network is used; the input layer includes all parameters, and the output layer has relative bandwidth and thickness.

[0033] Furthermore, the database module of step 6 stores the real values ​​trained by the electromagnetic simulation software and a large number of performances of each generation of populations.

[0034] Beneficial effects of the present invention:

[0035] The invention can optimize the absorption bandwidth and the thickness of the absorber at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 It is a schematic diagram of the absorber pixel unit of the present invention.

[0037] Figure 2 Schematic diagram of an array of absorbers according to the present invention.

[0038] Figure 3 It is a schematic diagram of the transparent grid medium supporting the conductive film layer of the present invention.

[0039] Figure 4 is a graph of the normal incidence absorptivity of an example of the present invention.

[0040] Figure 5 This is an oblique incidence absorptivity diagram of an example of the present invention.

[0041] Figure 6 Graph showing different polarization absorption rates of an example of the present invention.

[0042] Figure 7 These are illustrations of Embodiment 2 of the present invention, wherein (a) is an upper resonator layer, and (b) is a diagram of a middle resonator layer.

[0043] Figure 8 It is a flow chart of the method of the present invention.

[0044] Fig. 9 It is a functional schematic diagram of the multi-objective genetic algorithm module of the present invention. DETAILED DESCRIPTION

[0045] The technical solutions in the embodiments of the present invention will be described clearly and completely below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0046] Example 1

[0047] The structure of the absorber resonator layer is designed using a neural network-assisted multi-objective optimization algorithm, thereby achieving the design of metasurface absorbers under low computing power conditions. This solves the problem that the design of complex pattern resonator absorbers requires high computer configuration and is time-consuming.

[0048] A design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm, the design optimization system comprising a multi-objective genetic algorithm module, an electromagnetic simulation software module, a neural network regression module and a database module;

[0049] The multi-objective genetic algorithm module calls other modules to realize the rapid design of transparent absorbers;

[0050] The electromagnetic simulation software module calculates the performance of the primary population;

[0051] The neural network regression module predicts the objective function value of the same type of population;

[0052] The database module stores all real values ​​obtained by simulation using the electromagnetic simulation software module.

[0053] A design method for a design optimization system of a broadband transparent absorber based on a multi-objective genetic algorithm, the design optimization method comprising the following steps:

[0054] Step 1: Use the multi-objective genetic algorithm module to establish the absorber population;

[0055] Step 2: Calculate the performance of the absorber population in step 1 or step 3. If there are identical individual results in the database module, proceed to step 4. If the prediction accuracy of the neural network module for the previous generation is higher than 60%, randomly select some individuals and proceed to step 5, and the remaining individuals proceed to step 6. If neither of the above is satisfied, proceed to step 6, and input the results obtained in step 6 into the database module.

[0056] Step 3: Determine whether the performance calculated in step 2 meets the requirements. If so, stop the optimization and output the optimized absorber; if not, generate a new population individual and repeat step 2.

[0057] Step 4: Call the real value in the database module;

[0058] Step 5: Call the neural network regression module to predict individual performance;

[0059] Step 6: Call the electromagnetic simulation software module to calculate individual performance.

[0060] Furthermore, the multi-objective genetic algorithm in step 1 can obtain the Pareto optimal solutions of multiple evaluation indicators, and the performance of the absorber population is evaluated using the obtained Pareto optimal solutions of the multiple evaluation indicators.

[0061] Furthermore, the establishment of the absorber population in step 1 is divided into establishing an initial population and generating a sub-population; the initial population is established by randomly generated parameters; the sub-population is generated based on the evaluation results of the previous population, using non-dominated sorting values ​​and crowding, selection, crossover and mutation operations.

[0062] Furthermore, the electromagnetic simulation software model in step 6 calculates the true value of the primary population and the true value of the offspring population.

[0063] Furthermore, the electromagnetic simulation software model calculation in step 6 is specifically to define the relative bandwidth corresponding to the frequency range in which the absorption rate is greater than 0.9 or 0.95 as the calculation range of the electromagnetic simulation software model;

[0064] The formula for absorbance is defined as:

[0065] A=1-|S 11 | 2 -|S 21 | 2

[0066] Where S11 is the reflectivity and S21 is the transmittance;

[0067] The relative bandwidth formula is defined as:

[0068] ffoc=2×(fH-fL) / (fH+fL)

[0069] Where ffoc represents the relative bandwidth, fH represents the highest frequency of the entire frequency band where the absorption rate is greater than 0.9, and fL represents the lowest frequency of the entire frequency band where the absorption rate is greater than 0.9.

[0070] Furthermore, the neural network regression module of step 5 learns the data in the database, establishes the relationship between the parameters and the results, and calculates the true value of the offspring population;

[0071] When deciding whether to use the neural network to evaluate the nth generation and n>3, the original data in the database module and the true value data of the previous n-2 generations are needed to predict the accuracy of the n-1 generation; when the accuracy of the n-1 generation is greater than 60%, 50% of the random individuals of the nth generation are predicted using the neural network, and the rest are calculated using the electromagnetic simulation software module; where accuracy = 1-|predicted value-true value| / true value.

[0072] The prediction accuracy is specifically:

[0073] Neural network optimization starts from the third generation or higher. The result of the first generation calculated by the simulation software is A1, and the result of the second generation calculated by the simulation software is A2. The original data in the database is AA. The algorithm for predicting accuracy is: use the data of AA and A1 to train the neural network, and use it to predict the result of the second generation to obtain B2, and compare the results obtained by B2 and A2. For example, there are only two individuals in the population, the true value A2 is (1.15, 1.21), and the predicted value B2 is (1, 1.3), then the accuracy is equal to (1-|1.15-1| / 1.15, 1-|1.3-1| / 1.3). That is, accuracy = 1-|predicted value-true value| / true value. The accuracy of the second generation obtained is used to determine how many proportions of individuals in the third generation are optimized by the neural network. When the accuracy is higher than 60%, 50% of the random individuals in the third generation are optimized by the neural network to obtain data B3, and the rest are calculated by the simulation software to obtain A3.

[0074] When we start to decide on the fourth generation, we now have four data points, AA, A1, A2, and A3. We will use AA, A1, and A2 to train the neural network to predict the third generation results and obtain the accuracy of B3 and A3 calculations. The same logic applies. That is, when deciding how to evaluate the nth generation, we need to use the data of the previous n-2 generations to predict the accuracy of the n-1 generation. The accuracy of the n-1 generation is used to determine the random 50% individuals of the nth generation to be calculated using the neural network. The accuracy = 1-|predicted value-true value| / true value.

[0075] Furthermore, the neural network regression module is calculated specifically as follows: the neural network regression is expressed as the function ffoc&Thickness=ANN(Par). Since the relationship between the input parameters and the absorption relative bandwidth and thickness is established, the new parameters are input into the relationship to obtain the predicted value of the relative bandwidth corresponding to the parameters; a multi-layer fully connected neural network is used; the input layer includes all parameters, and the output layer has relative bandwidth and thickness.

[0076] Furthermore, the database module of step 6 stores the real values ​​trained by the electromagnetic simulation software and a large number (more than 20,000) of population performances of each generation.

[0077] Example 2

[0078] In order to optimize the metasurface absorber, the absorber structure needs to be parameterized. Taking the three-layer metasurface structure as an example, the absorber includes a reflective floor layer, a middle resonator layer and an upper resonator layer, and the film layers are air medium. What needs to be parameterized are the shape of the resonator, the surface resistance of the resonator, the unit period length, the interlayer spacing, etc. Parameters expressed by numerical values ​​can be optimized directly, but the shape optimization of the resonator requires other means. The shape data of the resonator is characterized by digital coding. The resonator layer is divided into a chessboard grid, and the number "0" or "1" is filled in each grid of the chessboard. Among them, "0" means that there is no conductive coating at this position, and "1" means that there is a conductive coating at this position. In general, in order for the absorber to have better polarization stability, a four-fold rotational symmetry is adopted, so only the parameters corresponding to the shape of one quarter of the position need to be defined. The use of a double-layer complementary form can achieve low-coupling multi-mode absorption. The encoding parameter x pattern =[0,0,0,1,1,1,0,0,0,0,0,0,0,1,1]

[0079] Other parameters: x parameter =[x1,x2,x3,x4]=[22.7.27,3.46,87.14]. Among them, x1,x2,x3,x4 represent the unit period length, the distance between the upper layer and the middle layer, the distance between the middle layer and the lower layer, and the resistance of the resonator film layer respectively.

[0080] The metamaterial absorber corresponding to this unit has good absorption effects on electromagnetic waves of different polarizations and incident at different angles. Electromagnetic waves of different polarizations can have a high absorption rate of more than 0.9 in the frequency range of 6.1-40GHz, and the thickness is only 6.105mm, achieving a low profile.

[0081] An optical simulation module is arranged in parallel with the electromagnetic simulation software module, the neural network regression module and the database module. The optical simulation module directly calculates the transmittance by the transfer matrix method, so that the system can simultaneously optimize the absorption bandwidth, transmittance and the thickness of the absorber.

[0082] Example 3

[0083] As attached Figure 1 , Attachment Figure 2 , Attachment Figure 3 As shown, the metamaterial absorber structure of the present invention includes a bottom low-impedance conductive film layer 1, a middle patterned conductive film layer 2, and an upper patterned conductive film layer 3. The absorber size parameters include the side length of the x1 unit, the thickness of the upper air cavity x2, and the thickness of the lower air cavity x3. The absorber units are combined into a 10×10 rectangular array, and the conductive film layer is attached to a transparent grid medium such as Figure 2 shown. Figure 3The width of the transparent grid medium is w1.

[0084] First, an electromagnetic simulation model of the absorber unit is established in the electromagnetic simulation software, and the physical parameters of the absorber are input. The film used in the present invention is a PET-ITO conductive film layer. The dielectric constant of PET is set to 2.65, the loss tangent is set to 0.015, and the PET film thickness is set to 0.012mm. The surface resistance of the bottom conductive film layer is set to 6Ω / □, and the parameter element x is set to parameter , the unit side length x1, the thickness of the upper air cavity x2, the thickness of the lower air cavity x3, and the surface resistance x4 of the upper and middle transparent conductive films are variables. In the electromagnetic simulation model, there are only the upper and middle PET substrates and the bottom low-impedance transparent conductive film. The patterns of the upper and middle layers are filled by the program.

[0085] Enter the number of individuals x that evolve in each generation of the genetic algorithm in the program written in Python base , and the selectable range of x1, x2, x3, and x4. Run the program and use random sampling to initialize the pattern element x of the first generation simulation within the selectable range of variables. pattern With parameter element x parameter . Using the electromagnetic simulation software interface in the program, a pattern model is established in the electromagnetic simulation software, and the initial values ​​of the variables are imported for simulation.

[0086] The simulation software can directly obtain the S parameters of the absorber, and the post-processing program converts the S parameters into the required fitness function value through the formula. Through the fitness function value, the individuals of this generation are selected, crossed, and mutated to generate the pattern elements x of the next generation. pattern With parameter element x parameter , and then imported into the electromagnetic simulation software to start the subsequent cycle.

[0087] When the value of the fitness function meets the requirement, the genetic algorithm stops running. The pattern elements and parameter elements corresponding to the absorber with the largest relative bandwidth are obtained. The designed broadband absorber unit is obtained by using them. The units are arranged in a matrix and etched on a uniform conductive film layer. The film layer is attached to a transparent grid to make a sample.

[0088] The simulation results of the absorption unit are shown in the attached figure. Figure 4 , Attachment Figure 5 , Attachment Figure 6 As shown. The absorption rate at 9.28GHz and 31.33GHz reaches 0.99, and the absorption higher than 0.9 covers 6-36GHz, with a relative bandwidth of 1.42. Within the range of less than 50° inclined incidence, the absorption capacity higher than 0.9 can still be guaranteed within the frequency band. When electromagnetic waves of different polarizations are incident, the absorption capacity of the absorber proposed by the present invention is not affected.

Claims

1. A design method for a broadband transparent absorber design optimization system based on a multi-objective genetic algorithm, characterized in that: The design optimization system includes a multi-objective genetic algorithm module, an electromagnetic simulation software module, a neural network regression module and a database module; The multi-objective genetic algorithm module calls other modules to realize the rapid design of transparent absorbers; The electromagnetic simulation software module calculates the performance of the primary population; The neural network regression module predicts the objective function value of the same type of population; The database module stores all real values ​​obtained by simulation using the electromagnetic simulation software module; The design method comprises the following steps: Step 1: Use the multi-objective genetic algorithm module to establish the absorber population; Step 2: Calculate the performance of the absorber population in step 1 or step 3. If there are identical individual results in the database module, proceed to step 4. If the prediction accuracy of the neural network module for the previous generation is higher than 60%, randomly select some individuals and proceed to step 5, and the remaining individuals proceed to step 6. If neither of the above is satisfied, proceed to step 6, and input the results obtained in step 6 into the database module. Step 3: Determine whether the performance calculated in step 2 meets the requirements. If so, stop the optimization and output the optimized absorber; if not, generate a new population individual and repeat step 2. Step 4: Call the real value in the database module; Step 5: Call the neural network regression module to predict individual performance; Step 6: Call the electromagnetic simulation software module to calculate individual performance.

2. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The multi-objective genetic algorithm in step 1 can obtain the Pareto optimal solutions of multiple evaluation indicators, and the performance of the absorber population is evaluated using the obtained Pareto optimal solutions of the multiple evaluation indicators.

3. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The establishment of the absorber population in step 1 is divided into establishing an initial population and generating a sub-population; the initial population is established by randomly generated parameters; the sub-population is generated based on the evaluation results of the previous population, using non-dominated sorting values ​​and crowding, selection, crossover and mutation operations.

4. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The electromagnetic simulation software model in step 6 calculates the true value of the primary population and the true value of part of the offspring population.

5. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The electromagnetic simulation software model calculation in step 6 specifically includes defining the relative bandwidth corresponding to the frequency range in which the absorption rate is greater than 0.9 or 0.95 as the calculation range of the electromagnetic simulation software model; The formula for absorbance is defined as: A=1- | S 11| 2 - | S 21| 2 Where S11 is the reflectivity and S21 is the transmittance; The relative bandwidth formula is defined as: ffoc=2×(fH-fL) / (fH+fL) Where ffoc represents the relative bandwidth, fH represents the highest frequency of the entire frequency band where the absorption rate is greater than 0.9, and fL represents the lowest frequency of the entire frequency band where the absorption rate is greater than 0.

9.

6. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The neural network regression module in step 5 learns the data in the database, establishes the relationship between parameters and results, and calculates the true value of the offspring population; When deciding whether to use the neural network to evaluate the nth generation and n>3, the original data in the database module and the true value data of the previous n-2 generations are needed to predict the accuracy of the n-1 generation; when the accuracy of the n-1 generation is greater than 60%, 50% of the random individuals of the nth generation are predicted using the neural network, and the rest are calculated using the electromagnetic simulation software module; where accuracy = 1-|predicted value-true value| / true value.

7. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The neural network regression module is calculated specifically as follows: the neural network regression is expressed as a function ffoc&Thickness=ANN(Par). Since the relationship between the input parameters and the absorption relative bandwidth and thickness is established, the new parameters are input into the relationship to obtain the predicted value of the relative bandwidth corresponding to the parameters; a multi-layer fully connected neural network is used; the input layer includes all parameters, and the output layer has relative bandwidth and thickness.

8. The design method of a design optimization system for a broadband transparent absorber based on a multi-objective genetic algorithm according to claim 1, characterized in that: The database module in step 6 stores the real values ​​trained by the electromagnetic simulation software and a large number of population performances of each generation.

Images