A method for optimizing the design of infrared stealth material film based on machine learning

Abstract

The present invention discloses an infrared stealth material film optimization design method based on machine learning, comprising: (1) constructing a mathematical optimization model to minimize the error target value of the infrared stealth material film spectral emissivity; (2) using Latin hypercube sampling to generate a population; (3) constructing a Kriging machine learning model combined with a maximum information coefficient with attention information enhancement; (4) generating a first subpopulation and a second subpopulation based on the error target value and a diversity index; (5) performing a dimensional perturbation-driven gradient descent mutation on the first subpopulation, and selecting the first offspring individual based on the prediction error; (6) performing a double-layer differential mutation on the second subpopulation, and selecting the second offspring individual based on the expected improvement value; (7) performing simulation analysis and updating the database, and returning to step (3) until the number of simulation analysis reaches the design cycle. The present invention can improve the convergence accuracy of the infrared stealth material film design problem.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence and swarm intelligence technology, and in particular to a method for optimizing the design of infrared stealth material film layers based on machine learning. Background Art

[0002] Infrared stealth materials have wide application potential in the civilian field, especially in high-tech equipment, where their role is becoming increasingly important. With the increasing demand for stealth technology in the fields of drones, aerospace, etc., the research on infrared stealth materials has gradually become a key technical direction. Traditional infrared stealth materials, such as low-emissivity materials, mainly improve stealth by reducing the infrared radiation characteristics of the target. However, an important problem with this type of material is that they will increase the surface temperature of the target, which may expose the existence of the target. Therefore, how to reduce infrared radiation while avoiding the increase in surface temperature has become a core challenge for optimization design.

[0003] In order to deal with this problem, researchers have begun to focus on the development of spectrally selective infrared stealth materials in recent years. The design concept of this type of material is to maintain a low emissivity within a specific wavelength range, especially the atmospheric window band (band range of 3μm-5μm and 8μm-14μm), and maintain a high emissivity in the non-atmospheric window band (band range of 5μm-8μm). In this way, it is possible to achieve precise control of infrared radiation characteristics, effectively reducing the infrared visibility of the target and avoiding excessive increase in surface temperature, thereby enhancing the stealth effect.

[0004] Designing spectrally selective infrared stealth materials requires comprehensive consideration of factors such as film thickness, material selection, and structural layout to ensure that the emissivity in different wavelength ranges meets the requirements. This process usually involves a lot of calculations, experimental debugging, and time-consuming simulation evaluation. In addition, when the design space is extremely complex, traditional methods such as the control variable method are often inefficient and require a lot of time and resources.

[0005] In order to solve these problems, evolutionary algorithms have been introduced into the optimization design of spectrally selective infrared stealth materials. By simulating the evolutionary process in nature, evolutionary algorithms can automatically find the optimal solution in a huge design space and reduce human intervention. However, although evolutionary algorithms have the ability of global optimization, they require a large number of simulation calculations, so the computational overhead is large, and the complex coupling between material structure and physical properties may not be fully considered during the design process, resulting in unsatisfactory results.

[0006] At the same time, the rapid development of machine learning technology has brought new opportunities for the design of spectrally selective infrared stealth materials. Machine learning can efficiently process complex simulation models and provide accurate prediction results. Especially when facing high-dimensional design spaces, machine learning can greatly improve optimization efficiency and shorten calculation time. However, many current optimization algorithms have not fully integrated the potential of machine learning and have failed to maximize its advantages in complex designs. Summary of the invention

[0007] In view of the limitations of the existing technology or the need for improved technology, the present invention proposes an infrared stealth material film optimization design method based on machine learning. Based on the time-consuming and complex simulation characteristics and the single-objective optimization requirements of reflectivity error involved in the optimization design of infrared stealth material film, an optimization design method based on machine learning combining a Kriging machine learning model and an evolutionary algorithm is studied and designed. The method replaces the time-consuming simulation by constructing a Kriging machine learning model with maximum information coefficient enhanced by attention information, divides the population by the error target value and diversity index, and combines the gradient descent mutation driven by dimensional perturbation and the algorithm based on double-layer differential mutation to perform optimization search, which effectively improves the efficiency of the optimization design of infrared stealth material film.

[0008] The infrared stealth material film layer optimization design method based on machine learning designed in the present invention can obtain high-performance optimization results within a limited design cycle. It can not only be used for the design of infrared stealth material film layers, but also provide a reference for the optimization of other complex structures.

[0009] To achieve the above object, according to one aspect of the present invention, a method for optimizing the design of infrared stealth material film layers based on machine learning is provided, the method comprising:

[0010] Step (1): Taking the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer as the optimization design parameters and determining the design space, taking the error target value of the spectral emissivity of the infrared stealth material film layer as the optimization target, and constructing a mathematical optimization model that minimizes the error target value;

[0011] Step (2): Use the Latin hypercube sampling method to generate a population in the design space, obtain the spectral emissivity of the individuals in the population through optical simulation software and Matrix Laboratory analysis, calculate the error target value, and build a database;

[0012] Step (3): Calculate the mutual information value and maximum information coefficient of each optimized design parameter, fuse the maximum information coefficient correlation function in a linear combination manner to construct the overall correlation function, and derive the Kriging machine learning model combined with the maximum information coefficient for attention information enhancement;

[0013] Step (4): Based on the error target value, the first subpopulation and the temporary population are constructed and the local diversity index is calculated. The temporary population is merged with the database to form a temporary database and the global diversity index is calculated. In the non-dominated sorting method, the second subpopulation is constructed according to the global diversity and local diversity indexes.

[0014] Step (5): For each individual in the first subpopulation, generate an extended subpopulation and the first type of candidate offspring individuals through dimensional perturbation mutation and gradient descent mutation, and select the first offspring individual based on the affine propagation clustering algorithm and the Kriging machine learning model;

[0015] Step (6): For each individual of the second subpopulation, the second offspring individual is obtained by combining the double-layer differential variation and Kriging machine learning model;

[0016] Step (7): Perform simulation analysis on the first and second offspring individuals and update the database; if the number of simulation analyses consumed reaches the design cycle, output the optimal infrared stealth material film thickness parameter value; otherwise, go to step (3) until the design cycle is reached.

[0017] Furthermore, the specific steps of step (1) are as follows:

[0018] In the first step, considering that the infrared stealth material film layer is composed of the thickness of multiple metal layers and dielectric layers stacked in sequence, the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer are used as optimization design parameters;

[0019] The second step is to determine the value range of the optimized design parameters according to the optical transparency of the infrared stealth material film layer and the structural characteristics of applicability, and determine the design space according to the value range;

[0020] The third step is to use the error target value of the spectral emissivity of the infrared stealth material film layer as the optimization target, and construct a mathematical optimization model that minimizes the error target value. Its mathematical expression is as follows:

[0021] ,

[0022] in, It represents the optimized design parameters corresponding to the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer. Indicates the thickness of the first layer of material, Indicates the thickness of the second layer of material, Indicates n The thickness of the layer material, RMSD represents the error target value of the spectral emissivity of the infrared stealth material film layer, that is, the error target value of the spectral emissivity of the designed infrared stealth material film layer in the atmospheric window band and the non-atmospheric window band relative to the ideal spectral emissivity. Indicates the wavelength of the infrared stealth material film layer. The designed infrared stealth material film layer is The spectral emissivity measured at wavelength, The designed infrared stealth material film layer is The ideal spectral emissivity at the wavelength is 0 in the atmospheric window band and 1 in the non-atmospheric window band. W represents the number of wavelengths taken, It represents the design space determined by the thickness parameter range of each metal layer and dielectric layer of the infrared stealth material film layer, Find represents the optimal solution for optimizing the design parameters, Min represents minimizing the error target value, and St represents the constraints that need to be met.

[0023] Furthermore, the specific steps of step (2) are as follows:

[0024] The first step is to determine the number of individuals in the population according to the design cycle of the infrared stealth material film design and the number of design variables in the design space;

[0025] In the second step, the population is generated using the Latin hypercube sampling method in the design space according to the number of individuals in the population;

[0026] The third step is to obtain the spectral emissivity of all individuals in the population through optical simulation software and Matrix Laboratory analysis. The specific steps are as follows:

[0027] For each population individual, start an optical simulation software session and create a new simulation project, set the dimensions of the simulation area, and define the desired mesh accuracy;

[0028] Define the material properties of the substrate and create the substrate geometry within the simulation area;

[0029] Add the designed number of film layers in sequence, set the material properties of each film layer, and set the geometric dimensions and thickness parameters of the film layer;

[0030] Set up a plane wave light source to cover the desired wavelength range and set the incident angle and polarization state;

[0031] Place transmission and reflection monitors, specify the positions and directions of the transmission and reflection monitors to collect reflection and transmission data, and finally, set all six boundaries of the simulation area to perfect matching layer simulation boundary conditions. After the simulation configuration is completed, start the spectral emissivity simulation and start calculating and collecting the spectral emissivity simulation result data corresponding to the individuals in the population.

[0032] The fourth step is to calculate the error target value of the spectral emissivity of each individual in the population, and use all individuals in the population and the corresponding error target values ​​to construct a database.

[0033] Furthermore, the specific steps of step (3) are as follows:

[0034] In the first step, based on the individuals and error target values ​​in the database, the mutual information value of each optimized design parameter in the design space relative to the error target value is calculated. The expression is as follows:

[0035] ,

[0036] in, represents the mutual information value of each optimized design parameter in the design space relative to the error target value, The design space k The optimization design parameters, RMSD represents the error target value of the spectral emissivity of the infrared stealth material film layer, m represents the number of individuals in the database, is the probability density function estimated from the rectangular grid constructed by individuals, Indicates the first i The individual k Optimized design parameters, Indicates the first i The error target value of each individual;

[0037] The second step is to determine the maximum information coefficient based on the mutual information value. The expression is as follows:

[0038] ,

[0039] Among them, MIC represents the maximum information coefficient between the optimization design parameters and the error target value, Indicated in r，s Less than L Take the maximum value under the condition, L=q 0.6 , q represents the number of individuals modeled, r and s Indicates two parameters determined according to the database, log 2 It represents the logarithm with base 2, and min represents the minimum value;

[0040] The third step is to construct the maximum information coefficient correlation function, the expression is:

[0041] ,

[0042] in, Represents the correlation function of the maximum fusion information coefficient of any two individuals, represents two Gaussian random processes with zero mean but non-zero covariance, T i and T j Respectively represent i With j Individuals, represents the exponential function, n represents the number of optimized design parameters in the design space, represents the scale parameter, Indicates k The influence of the optimization design parameters on the error target value is Indicates j The design space of an individual k Optimized design parameters;

[0043] The fourth step is to calculate the attention information correlation function between any two individuals, expressed as:

[0044] ,

[0045] in, Represents the correlation function of the attention information between any two individuals, , , , ;

[0046] The fifth step is to fuse the maximum information coefficient correlation function in a linear combination to construct the overall correlation function. The expression of the overall correlation function is:

[0047] ,

[0048] in, represents the overall correlation function between any two individuals, and Represents two proportional parameters with a value range of (0,1);

[0049] The sixth step is to construct the maximum likelihood estimation function in the process of constructing the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient according to the overall correlation function. The expression is:

[0050] ,

[0051] Among them, L represents the constructed maximum likelihood estimation function, and 1 represents the dimension. The unit vector of The dimension is The correlation coefficient matrix of represents its inverse matrix, represents pi, and They represent the mean and variance of the Gaussian process respectively, and det represents the determinant;

[0052] Step 7: Correlation coefficient matrix in maximum likelihood estimation function The expression is as follows:

[0053] ,

[0054] in, represents the overall correlation function value between the first individual and itself, Indicates the first individual and the n The overall correlation function value of each individual, Indicates n The overall correlation function value of individuals and the first individual, Indicates n The overall correlation function value of each individual with itself;

[0055] In the eighth step, the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient is derived. The corresponding expressions of the predicted value and predicted variance are:

[0056] ,

[0057] ,

[0058] in, , , The Kriging machine learning model combining the maximum information coefficient with attention information enhancement is used for individual T The predicted value of ， The Kriging machine learning model combining the maximum information coefficient with attention information enhancement is used for individual T The prediction variance of Represents the Gaussian process mean corresponding to the Kriging machine learning model with maximum information coefficient combined with attention information enhancement, The Gaussian process variance corresponding to the Kriging machine learning model with maximum information coefficient combined with attention information enhancement;

[0059] In the ninth step, in the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient, the corresponding Gaussian process mean and variance calculation expressions are as follows:

[0060] ,

[0061] ,

[0062] in, Represents the transpose of a unit vector.

[0063] Furthermore, the step (4) specifically includes the following steps:

[0064] The first step is to sort the error target values ​​of all individuals in the database by descending rules, and select the top individuals in the database. p % individuals are used to construct the first subpopulation and temporary population;

[0065] In the second step, the pdist2 function is used to calculate the Euclidean distance between each individual in the temporary population and other individuals in the temporary population to generate a local distance matrix; the minimum distance of each individual in the temporary population is calculated through the local distance matrix, and the negative value of the minimum distance is taken as the local diversity index. Its mathematical expression is:

[0066] ,

[0067] in, T i , T j They represent the temporary population i, j individuals, TPOP represents the temporary population, represents the temporary population i Individual and j The Euclidean distance of individuals;

[0068] The third step is to merge the temporary population with the database to form a temporary database. The pdist2 function is used to calculate the Euclidean distance of each individual in the temporary database from other individuals in the temporary database to generate a global distance matrix. The minimum distance of each individual in the temporary database is calculated through the global distance matrix, and the negative value of the minimum distance is taken as the global diversity index. Its mathematical expression is:

[0069] ,

[0070] in, T i Indicates the temporary database i Individuals, T j Indicates the temporary database j individuals, DB means temporary database, Indicates the temporary database i Individual and j The Euclidean distance of individuals;

[0071] The fourth step is to use the global diversity index in the non-dominated sorting method. Gdiv ( Ti ) and local diversity indicators Ldiv ( T i ) for both targets N individuals to construct the second subpopulation.

[0072] Furthermore, the step (5) specifically includes the following steps:

[0073] The first step is to generate a mutation probability for each optimized design parameter of each individual in the first subpopulation. P select , which is used to decide whether to disturb the optimization design parameter; for each selected optimization design parameter that needs to be disturbed, a random disturbance value is added to the optimization design parameter to generate M The extended subpopulation of candidate individuals, this random perturbation value is randomly drawn from the Gaussian distribution, and the mathematical expression of dimensional perturbation variation is:

[0074] ,

[0075] in, Indicates the design space k The perturbed value of the optimal design parameter, For the design space k The original values ​​of the optimization design parameters, For the design space k Random perturbation values ​​of the optimization design parameters;

[0076] In the second step, the gradient of all individuals in the expanded subpopulation is calculated according to the Kriging machine learning model; the single step length of the gradient descent is calculated according to the standard deviation used in the dimensional perturbation variation, and the formula is as follows:

[0077] ,

[0078] Among them, SS represents the step size of the gradient descent step, The standard deviation set for the dimensional perturbation variation, is the set number of steps;

[0079] The third step is to use gradient descent to generate The candidate offspring individuals with different step lengths are ranked, and the error target value is predicted by the Kriging machine learning model. All candidate offspring individuals are sorted according to the ascending sorting rule of the predicted error target value, and the top u The candidate offspring individuals are taken as the first type of candidate offspring individuals;

[0080] Step 3: Execute the affine propagation clustering algorithm on all first-category candidate offspring individuals to obtain multiple clusters and their centers; select the optimal center according to the following rules: First, sort the first-category candidate offspring individuals in ascending order, and sort the first The first-class candidate offspring individuals are defined as the current excellent individuals; secondly, the cluster with the most excellent individuals determined by the affine propagation clustering algorithm is selected as the optimal cluster; thirdly, the center of the optimal cluster is taken as the optimal center; each first-class candidate offspring individual generates an optimal center, and the Kriging machine learning model is used to predict the optimal center, and the first-class candidate offspring individual with the smallest predicted target error value is selected as the first offspring individual.

[0081] Furthermore, in step (6), the specific steps are as follows:

[0082] In the first step, for each individual of the second subpopulation, the first-level differential mutation uses the DE / current-to-rand / 2 mutation operation and the binomial crossover operation to obtain the candidate offspring population, where the mutation operator of the DE / current-to-rand / 2 mutation operation is expressed as follows:

[0083] ,

[0084] in, It is a mutant individual generated by the DE / current-to-rand / 2 mutation operation. is the current evolutionary individual, , and are three different individuals randomly selected from the second subpopulation. F 1 and F 2 are two scaling factors used to control the vector;

[0085] The second step is to predict the error target value and expected improvement value of each individual in the candidate offspring population through the Kriging machine learning model;

[0086] The third step is to select the top candidates in the non-dominated sorting method using the individual error target value and expected improvement value as two goals. b % excellent candidate offspring population individuals;

[0087] The fourth step is to use the second-level differential mutation to perform DE / current-to-best / 2 mutation operation and binomial crossover operation on each excellent candidate offspring population individual to obtain the second type of candidate offspring individual pool, where the mutation operator of the DE / current-to-best / 2 mutation operation is expressed as follows:

[0088] ,

[0089] in, It is a mutant individual generated by the differential evolution DE / current-to-best / 2 mutation operation. is the current evolutionary individual, is the best vector in the current population, and are two different individuals randomly selected from the current population, F 1 and F 2 are two scaling factors used to control the scaling of the vector;

[0090] The fifth step is to select the best second-class candidate offspring individual from the second-class candidate offspring individual pool as the second offspring individual according to the error target value predicted by the Kriging machine learning model.

[0091] Furthermore, the specific steps of step (7) are as follows:

[0092] The first step is to use optical simulation software to perform simulation analysis on the first and second offspring individuals and calculate the error target value;

[0093] The second step is to store the first offspring individual and the second offspring individual and the corresponding error target value into the database to update the database;

[0094] In the third step, if the number of simulation analyses consumed reaches the design cycle, the optimal infrared stealth material film thickness parameter value is output; otherwise, go to step (3) until the design cycle is reached.

[0095] In a second aspect, the present invention also provides a computer device, comprising a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for optimizing the design of an infrared stealth material film layer based on machine learning.

[0096] In a third aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a method for optimizing the design of infrared stealth material film layers based on machine learning.

[0097] In summary, compared with the prior art, the infrared stealth material film layer optimization design method based on machine learning provided by the present invention has the following improvements over the limitations of the prior art:

[0098] 1. Considering the characteristics of traditional Kriging method that multiple hyperparameters need to be solved, the present invention adopts the maximum information coefficient to estimate the weight of the hyperparameter, and incorporates the weight of the hyperparameter into the correlation function of the maximum information coefficient to reduce the dimension and improve the model solution speed; introduces attention information to measure the correlation between individuals, and constructs a correlation coefficient matrix to enhance the adaptability of the Kriging model to complex problems;

[0099] 2. Through the population division strategy based on the two screening indicators of local diversity index and global diversity index, two sub-populations with high potential in convergence and exploration were divided, and the corresponding dimensional perturbation mutation was performed respectively, which effectively explored the potential of the offspring, thereby providing diversified search guidance and fast convergence speed during the dimensional perturbation mutation process;

[0100] 3. Traditional evolutionary algorithms are often prone to fall into local optimality, and in order to jump out of the local optimality, the convergence speed is usually reduced. The present invention constructs a similar extended population for individuals to generate diverse offspring individuals, and then supplemented with gradient descent mutation operations, which can significantly improve convergence; finally, the affine propagation clustering algorithm is used to remove disturbances and obtain the overall trend; it can ensure diversity in multiple spaces while improving the convergence speed;

[0101] 4. An evolutionary strategy consisting of two layers of differential mutation was designed. The first layer significantly improved diversity through random evolution, and the second layer enhanced the overall quality of the population through evolution in the optimal direction. At the same time, a dual-index population division strategy was added between the evolutions to adaptively and dynamically balance the impact of the two layers of evolution on the overall evolution.

[0102] 5. The present invention optimizes highly time-consuming design problems, reduces the consumption of computing resources, improves the solution rate, and has high use value. BRIEF DESCRIPTION OF THE DRAWINGS

[0103] Figure 1 A simplified flow chart of a method for optimizing the design of infrared stealth material film layers based on machine learning provided by the present invention. DETAILED DESCRIPTION

[0104] In order to more clearly illustrate the objectives, technical solutions and advantages of the present invention, the following will be analyzed in detail with reference to the accompanying drawings and specific embodiments. It should be particularly emphasized that the embodiments described herein are intended to help understand the core content of the present invention and do not impose any form of limitation on the scope of implementation. In addition, as long as the technical features in each embodiment do not conflict, they can be combined with each other and applied together.

[0105] See also Figure 1The present invention provides an infrared stealth material film layer optimization design method based on machine learning, which is applicable to the infrared stealth material film layer optimization emissivity design optimization problem. Specifically, the method includes steps (1) to (7).

[0106] Step (1): Taking the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer as the optimization design parameters and determining the design space, taking the error target value (such as the root mean square error RMSD target value) of the spectral emissivity of the infrared stealth material film layer as the optimization target, and constructing a mathematical optimization model that minimizes the error target value.

[0107] Step (1) specifically includes the following steps:

[0108] In the first step, considering that the infrared stealth material film layer is composed of the thickness of multiple metal layers and dielectric layers stacked in sequence, the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer are used as optimization design parameters;

[0109] The second step is to determine the value range of the optimized design parameters according to the optical transparency of the infrared stealth material film layer and the structural characteristics of applicability, and determine the design space according to the value range;

[0110] The third step is to use the error target value of the spectral emissivity of the infrared stealth material film layer as the optimization target, and construct a mathematical optimization model that minimizes the error target value. Its mathematical expression is as follows:

[0111] ,

[0112] in, It represents the optimized design parameters corresponding to the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer. Indicates the thickness of the first layer of material, Indicates the thickness of the second layer of material, Indicates n The thickness of the layer material, RMSD represents the error target value of the spectral emissivity of the infrared stealth material film layer, that is, the error target value of the spectral emissivity of the designed infrared stealth material film layer in the atmospheric window band and the non-atmospheric window band relative to the ideal spectral emissivity (such as the root mean square error RMSD target value), Indicates the wavelength of the infrared stealth material film layer. The designed infrared stealth material film layer is The spectral emissivity measured at wavelength, The designed infrared stealth material film layer is The ideal spectral emissivity at the wavelength is 0 in the atmospheric window band and 1 in the non-atmospheric window band. W represents the number of wavelengths taken, It represents the design space determined by the thickness parameter range of each metal layer and dielectric layer of the infrared stealth material film layer, Find represents the optimal solution for optimizing the design parameters, Min represents minimizing the error target value, and St represents the constraints that need to be met.

[0113] Step (2): Use the Latin hypercube sampling method to generate a population in the design space, obtain the spectral emissivity of the individuals in the population through optical simulation software (such as Ansys Lumerical FDTD software) and Matrix Laboratory analysis, calculate the error target value, and build a database.

[0114] Step (2) specifically includes the following steps:

[0115] The first step is to determine the number of individuals in the population according to the design cycle of the infrared stealth material film design and the number of design variables in the design space;

[0116] In the second step, the population is generated using the Latin hypercube sampling method in the design space according to the number of individuals in the population;

[0117] The third step is to obtain the spectral emissivity of all individuals in the population through optical simulation software and Matrix Laboratory analysis. The specific steps are as follows:

[0118] For each population individual, start an optical simulation software session and create a new simulation project, set the dimensions of the simulation area, and define the desired mesh accuracy;

[0119] Define the material properties of the substrate and create the substrate geometry within the simulation area;

[0120] Add the designed number of film layers in sequence, set the material properties of each film layer, and set the geometric dimensions and thickness parameters of the film layer;

[0121] Set up a plane wave light source to cover the desired wavelength range and set the incident angle and polarization state;

[0122] Place transmission and reflection monitors, specify the positions and directions of the transmission and reflection monitors to collect reflection and transmission data, and finally, set all six boundaries of the simulation area to perfect matching layer simulation boundary conditions. After the simulation configuration is completed, start the spectral emissivity simulation and start calculating and collecting the spectral emissivity simulation result data corresponding to the individuals in the population.

[0123] The fourth step is to calculate the error target value of the spectral emissivity of each individual in the population, and use all individuals in the population and the corresponding error target values ​​to construct a database.

[0124] Step (3): Calculate the mutual information value and maximum information coefficient of each optimized design parameter, fuse the maximum information coefficient correlation function in a linear combination manner to construct the overall correlation function, and derive the Kriging machine learning model combined with the maximum information coefficient for attention information enhancement.

[0125] Step (3) specifically includes the following steps:

[0126] In the first step, based on the individuals and error target values ​​in the database, the mutual information value of each optimized design parameter in the design space relative to the error target value is calculated. The expression is as follows:

[0127] ,

[0128] in, represents the mutual information value of each optimized design parameter in the design space relative to the error target value, The design space k The optimization design parameters, RMSD represents the error target value of the spectral emissivity of the infrared stealth material film layer, m represents the number of individuals in the database, is the probability density function estimated from the rectangular grid constructed by individuals, Indicates the first i The individual k Optimized design parameters, Indicates the first i The error target value of each individual;

[0129] The second step is to determine the maximum information coefficient based on the mutual information value. The expression is as follows:

[0130] ,

[0131] Among them, MIC represents the maximum information coefficient between the optimization design parameters and the error target value, Indicated in r，s Less than L Take the maximum value under the condition, L=q 0.6 , q represents the number of individuals modeled, r and s Indicates two parameters determined according to the database, log 2 It represents the logarithm with base 2, and min represents the minimum value;

[0132] The third step is to construct the maximum information coefficient correlation function, the expression is:

[0133] ,

[0134] in, Represents the correlation function of the maximum fusion information coefficient of any two individuals, represents two Gaussian random processes with zero mean but non-zero covariance, T i and T j Respectively represent i With j Individuals, represents the exponential function, n represents the number of optimized design parameters in the design space, represents the scale parameter, Indicates k The influence of the optimization design parameters on the error target value is Indicates j The design space of an individual k Optimized design parameters;

[0135] The fourth step is to calculate the attention information correlation function between any two individuals, expressed as:

[0136] ,

[0137] in, Represents the correlation function of the attention information between any two individuals, , , , ;

[0138] The fifth step is to fuse the maximum information coefficient correlation function in a linear combination to construct the overall correlation function. The expression of the overall correlation function is:

[0139] ,

[0140] in, represents the overall correlation function between any two individuals, and Represents two proportional parameters with a value range of (0,1);

[0141] The sixth step is to construct the maximum likelihood estimation function in the process of constructing the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient according to the overall correlation function. The expression is:

[0142] ,

[0143] Among them, L represents the constructed maximum likelihood estimation function, and 1 represents the dimension. The unit vector of The dimension is The correlation coefficient matrix of represents its inverse matrix, represents pi, and They represent the mean and variance of the Gaussian process respectively, and det represents the determinant;

[0144] Step 7: Correlation coefficient matrix in maximum likelihood estimation function The expression is as follows:

[0145] ,

[0146] in, represents the overall correlation function value between the first individual and itself, Indicates the first individual and the n The overall correlation function value of each individual, Indicates n The overall correlation function value of individuals and the first individual, Indicates n The overall correlation function value of each individual with itself;

[0147] In the eighth step, the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient is derived. The corresponding expressions of the predicted value and predicted variance are:

[0148] ,

[0149] ,

[0150] in, , , The Kriging machine learning model combining the maximum information coefficient with attention information enhancement is used for individual T The predicted value of ， The Kriging machine learning model combining the maximum information coefficient with attention information enhancement is used for individual T The prediction variance of Represents the Gaussian process mean corresponding to the Kriging machine learning model with maximum information coefficient combined with attention information enhancement, The Gaussian process variance corresponding to the Kriging machine learning model with maximum information coefficient combined with attention information enhancement;

[0151] In the ninth step, in the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient, the corresponding Gaussian process mean and variance (representing uncertainty) are calculated as follows:

[0152] ,

[0153] ,

[0154] in, Represents the transpose of a unit vector.

[0155] Step (4): Based on the error target value, the first sub-population and the temporary population are constructed and the local diversity index is calculated. The temporary population is merged with the database to form a temporary database and the global diversity index is calculated. In the non-dominated sorting method, the second sub-population is constructed according to the global diversity and local diversity indexes.

[0156] Step (4) specifically includes the following steps:

[0157] The first step is to sort the error target values ​​of all individuals in the database by descending rules, and select the top individuals in the database. p % individuals are used to construct the first subpopulation and temporary population;

[0158] In the second step, the pdist2 function is used to calculate the Euclidean distance between each individual in the temporary population and other individuals in the temporary population to generate a local distance matrix; the minimum distance of each individual in the temporary population is calculated through the local distance matrix, and the negative value of the minimum distance is taken as the local diversity index. Its mathematical expression is:

[0159] ,

[0160] in, T i , T j They represent the temporary population i, j individuals, TPOP represents the temporary population, represents the temporary population i Individual and j The Euclidean distance of individuals;

[0161] The third step is to merge the temporary population with the database to form a temporary database. The pdist2 function is used to calculate the Euclidean distance of each individual in the temporary database from other individuals in the temporary database to generate a global distance matrix. The minimum distance of each individual in the temporary database is calculated through the global distance matrix, and the negative value of the minimum distance is taken as the global diversity index. Its mathematical expression is:

[0162] ,

[0163] in, T i Indicates the temporary database i Individuals, T j Indicates the temporary database jindividuals, DB means temporary database, Indicates the temporary database i Individual and j The Euclidean distance of individuals;

[0164] The fourth step is to use the global diversity index in the non-dominated sorting method. Gdiv ( T i ) and local diversity indicators Ldiv ( T i ) for both targets N individuals to construct the second subpopulation.

[0165] Step (5): For each individual in the first subpopulation, generate an extended subpopulation and the first type of candidate offspring individuals through dimensional perturbation mutation and gradient descent mutation, and select the first offspring individual based on the affine propagation clustering algorithm and the Kriging machine learning model.

[0166] Step (5) specifically includes the following steps:

[0167] The first step is to generate a mutation probability for each optimized design parameter of each individual in the first subpopulation. P select , which is used to decide whether to disturb the optimization design parameter; for each selected optimization design parameter that needs to be disturbed, a random disturbance value is added to the optimization design parameter to generate M The extended subpopulation of candidate individuals, this random perturbation value is randomly drawn from the Gaussian distribution, and the mathematical expression of dimensional perturbation variation is:

[0168] ,

[0169] in, Indicates the design space k The perturbed value of the optimal design parameter, For the design space k The original values ​​of the optimization design parameters, For the design space k Random perturbation values ​​of the optimization design parameters;

[0170] In the second step, the gradient of all individuals in the expanded subpopulation is calculated according to the Kriging machine learning model; the single step length of the gradient descent is calculated according to the standard deviation used in the dimensional perturbation variation, and the formula is as follows:

[0171] ,

[0172] Among them, SS represents the step size of the gradient descent step, The standard deviation set for the dimensional perturbation variation, is the set number of steps;

[0173] The third step is to use gradient descent to generate The candidate offspring individuals with different step lengths are ranked, and the error target value is predicted by the Kriging machine learning model. All candidate offspring individuals are sorted according to the ascending sorting rule of the predicted error target value, and the top u The candidate offspring individuals are taken as the first type of candidate offspring individuals;

[0174] Step 3: Execute the affine propagation clustering algorithm on all first-category candidate offspring individuals to obtain multiple clusters and their centers; select the optimal center according to the following rules: First, sort the first-category candidate offspring individuals in ascending order, and sort the first The first-class candidate offspring individuals are defined as the current excellent individuals; secondly, the cluster with the most excellent individuals determined by the affine propagation clustering algorithm is selected as the optimal cluster; thirdly, the center of the optimal cluster is taken as the optimal center; each first-class candidate offspring individual generates an optimal center, and the Kriging machine learning model is used to predict the optimal center, and the first-class candidate offspring individual with the smallest predicted target error value is selected as the first offspring individual.

[0175] Step (6): For each individual of the second subpopulation, the second offspring individual is obtained by combining the double-layer differential variation and Kriging machine learning model.

[0176] Step (6) specifically includes the following steps:

[0177] In the first step, for each individual of the second subpopulation, the first-level differential mutation uses the DE / current-to-rand / 2 mutation operation and the binomial crossover operation to obtain the candidate offspring population, where the mutation operator of the DE / current-to-rand / 2 mutation operation is expressed as follows:

[0178] ,

[0179] in, It is a mutant individual generated by the DE / current-to-rand / 2 mutation operation. is the current evolutionary individual, , and are three different individuals randomly selected from the second subpopulation. F 1 and F 2 are two scaling factors used to control the vector;

[0180] The second step is to predict the error target value and expected improvement value of each individual in the candidate offspring population through the Kriging machine learning model;

[0181] The third step is to select the top candidates in the non-dominated sorting method using the individual error target value and expected improvement value as two goals. b % excellent candidate offspring population individuals;

[0182] The fourth step is to use the second-level differential mutation to perform DE / current-to-best / 2 mutation operation and binomial crossover operation on each excellent candidate offspring population individual to obtain the second type of candidate offspring individual pool, where the mutation operator of the DE / current-to-best / 2 mutation operation is expressed as follows:

[0183] ,

[0184] in, It is a mutant individual generated by the differential evolution DE / current-to-best / 2 mutation operation. is the current evolutionary individual, is the best vector in the current population, and are two different individuals randomly selected from the current population, F 1 and F 2 are two scaling factors used to control the scaling of the vector;

[0185] The fifth step is to select the best second-class candidate offspring individual from the second-class candidate offspring individual pool as the second offspring individual according to the error target value predicted by the Kriging machine learning model.

[0186] Step (7): Perform simulation analysis on the first and second offspring individuals and update the database; if the number of simulation analyses consumed reaches the design cycle, output the optimal infrared stealth material film thickness parameter value; otherwise, go to step (3) until the design cycle is reached.

[0187] Step (7) specifically includes the following steps:

[0188] The first step is to use optical simulation software to perform simulation analysis on the first and second offspring individuals and calculate the error target value;

[0189] The second step is to store the first offspring individual and the second offspring individual and the corresponding error target value into the database to update the database;

[0190] In the third step, if the number of simulation analyses consumed reaches the design cycle, the optimal infrared stealth material film thickness parameter value is output; otherwise, go to step (3) until the design cycle is reached.

[0191] Example 1

[0192] This embodiment uses the Hybrid Function 6 (N=6) benchmark test function to evaluate the performance of the proposed infrared stealth material film layer optimization design method based on machine learning. The mathematical expression of the benchmark test function is as follows:

[0193] ,

[0194] ,

[0195] in, is the benchmark function expression, n represents the number of dimensions of the design space, is a construction auxiliary function, Indicates i design variables, Indicates i+1 design variables.

[0196] The above-mentioned benchmark test function Hybrid Function 6 (N=6) is processed respectively through steps (1) to (7) of the infrared stealth material film layer optimization design method based on machine learning provided by the present invention to obtain experimental results.

[0197] In order to further illustrate the superiority of the present invention, the present invention is compared with the evolutionary sampling assisted optimization algorithm. The experimental parameters are set as follows: the maximum number of simulation evaluations is 300, and the number of design variables is 10. The average target values ​​of 20 independent runs are compared, and the experimental results are shown in Table 1. The results show that the present invention is significantly better than the evolutionary sampling assisted optimization algorithm under the same number of simulations, proving its excellent performance in solving the optimization design problem of infrared stealth material film layers.

[0198] Table 1: Comparison of optimization results of different algorithms

[0199]

[0200] The present invention provides a method for optimizing the design of infrared stealth material film layers based on machine learning. By constructing a Kriging machine learning agent model with attention information enhancement combined with the maximum information coefficient, a population division strategy based on dual indicators, and a gradient descent mutation driven by dimensional perturbation and a double-layer differential mutation, a comprehensive solution is provided for the optimization design of infrared stealth material film layers.

[0201] Embodiment 2, the embodiment of the present invention provides a computer device, including a memory and a processor, the memory stores a computer program, and when the processor executes the computer program, it implements the steps of a method for optimizing the design of an infrared stealth material film layer based on machine learning in the aforementioned embodiment.

[0202] Embodiment 3, the embodiment of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, the steps of a method for optimizing the design of an infrared stealth material film layer based on machine learning in the aforementioned embodiment are implemented.

[0203] It should be noted that what is described herein is only a preferred embodiment of the method and is not intended to limit the scope of the present invention. Any modifications, equivalent substitutions and improvements made by those skilled in the art without violating the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A method for optimizing the design of infrared stealth material film based on machine learning, characterized in that: The method comprises: Step (1): Taking the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer as the optimization design parameters and determining the design space, taking the error target value of the spectral emissivity of the infrared stealth material film layer as the optimization target, and constructing a mathematical optimization model that minimizes the error target value; Step (2): Use the Latin hypercube sampling method to generate a population in the design space, obtain the spectral emissivity of the individuals in the population through optical simulation software and Matrix Laboratory analysis, calculate the error target value, and build a database; Step (3): Calculate the mutual information value and maximum information coefficient of each optimized design parameter, fuse the maximum information coefficient correlation function in a linear combination manner to construct the overall correlation function, and derive the Kriging machine learning model combined with the maximum information coefficient for attention information enhancement; Step (4): Based on the error target value, the first subpopulation and the temporary population are constructed and the local diversity index is calculated. The temporary population is merged with the database to form a temporary database and the global diversity index is calculated. In the non-dominated sorting method, the second subpopulation is constructed according to the global diversity and local diversity indexes. Step (5): For each individual in the first subpopulation, generate an extended subpopulation and the first type of candidate offspring individuals through dimensional perturbation mutation and gradient descent mutation, and select the first offspring individual based on the affine propagation clustering algorithm and the Kriging machine learning model; Step (6): For each individual of the second subpopulation, the second offspring individual is obtained by combining the double-layer differential variation and Kriging machine learning model; Step (7): Perform simulation analysis on the first and second offspring individuals and update the database; if the number of simulation analyses consumed reaches the design cycle, output the optimal infrared stealth material film thickness parameter value; otherwise, go to step (3) until the design cycle is reached.

2. The method according to claim 1, characterized in that The specific steps of step (1) are as follows: In the first step, considering that the infrared stealth material film layer is composed of multiple metal layers and dielectric layers stacked in sequence, the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer are used as optimization design parameters; The second step is to determine the value range of the optimized design parameters according to the optical transparency of the infrared stealth material film layer and the structural characteristics of applicability, and determine the design space according to the value range; The third step is to use the error target value of the spectral emissivity of the infrared stealth material film layer as the optimization target, and construct a mathematical optimization model that minimizes the error target value. Its mathematical expression is as follows: ， in, It represents the optimized design parameters corresponding to the thickness parameters of each metal layer and dielectric layer of the infrared stealth material film layer. Indicates the thickness of the first layer of material, Indicates the thickness of the second layer of material, Indicates n The thickness of the layer material, RMSD represents the error target value of the spectral emissivity of the infrared stealth material film layer, that is, the error target value of the spectral emissivity of the designed infrared stealth material film layer in the atmospheric window band and the non-atmospheric window band relative to the ideal spectral emissivity. Indicates the wavelength of the infrared stealth material film layer. The designed infrared stealth material film layer is The spectral emissivity measured at wavelength, The designed infrared stealth material film layer is The ideal spectral emissivity at the wavelength is 0 in the atmospheric window band and 1 in the non-atmospheric window band. W represents the number of wavelengths taken, It represents the design space determined by the thickness parameter range of each metal layer and dielectric layer of the infrared stealth material film layer, Find represents the optimal solution for optimizing the design parameters, Min represents minimizing the error target value, and St represents the constraints that need to be met.

3. The method according to claim 1, characterized in that The specific steps of step (2) are as follows: The first step is to determine the number of individuals in the population according to the design cycle of the infrared stealth material film design and the number of design variables in the design space; In the second step, the population is generated using the Latin hypercube sampling method in the design space according to the number of individuals in the population; The third step is to obtain the spectral emissivity of all individuals in the population through optical simulation software and Matrix Laboratory analysis. The specific steps are as follows: For each population individual, start an optical simulation software session and create a new simulation project, set the dimensions of the simulation area, and define the desired mesh accuracy; Define the material properties of the substrate and create the substrate geometry within the simulation area; Add the designed number of film layers in sequence, set the material properties of each film layer, and set the geometric dimensions and thickness parameters of the film layer; Set up a plane wave light source to cover the desired wavelength range and set the incident angle and polarization state; Place transmission and reflection monitors, specify the positions and directions of the transmission and reflection monitors to collect reflection and transmission data, and finally, set all six boundaries of the simulation area to perfect matching layer simulation boundary conditions. After the simulation configuration is completed, start the spectral emissivity simulation and start calculating and collecting the spectral emissivity simulation result data corresponding to the individuals in the population. The fourth step is to calculate the error target value of the spectral emissivity of each individual in the population, and use all individuals in the population and the corresponding error target values ​​to construct a database.

4. The method according to claim 1, characterized in that The specific steps of step (3) are as follows: In the first step, based on the individuals and error target values ​​in the database, the mutual information value of each optimized design parameter in the design space relative to the error target value is calculated. The expression is as follows: ， in, represents the mutual information value of each optimized design parameter in the design space relative to the error target value, The design space k The optimization design parameters, RMSD represents the error target value of the spectral emissivity of the infrared stealth material film layer, m represents the number of individuals in the database, is the probability density function estimated from the rectangular grid constructed by individuals, Indicates the first i The individual k Optimized design parameters, Indicates the first i The error target value of each individual; The second step is to determine the maximum information coefficient based on the mutual information value. The expression is as follows: ， Among them, MIC represents the maximum information coefficient between the optimization design parameters and the error target value, Indicated in r，s Less than L Take the maximum value under the condition, L=q 0.6 , q represents the number of individuals modeled, r and s Indicates two parameters determined according to the database, log2 indicates the logarithm with base 2, and min indicates the minimum value; The third step is to construct the maximum information coefficient correlation function, the expression is: ， in, Represents the correlation function of the maximum fusion information coefficient of any two individuals, represents two Gaussian random processes with zero mean but non-zero covariance, T i and T j Respectively represent i With j Individuals, represents the exponential function, n represents the number of optimized design parameters in the design space, represents the scale parameter, Indicates k The influence of the optimization design parameters on the error target value is Indicates j The design space of an individual k Optimized design parameters; The fourth step is to calculate the attention information correlation function between any two individuals, expressed as: ， in, Represents the correlation function of the attention information between any two individuals, , , , ; The fifth step is to fuse the maximum information coefficient correlation function in a linear combination to construct the overall correlation function. The expression of the overall correlation function is: ， in, represents the overall correlation function between any two individuals, and Represents two proportional parameters with a value range of (0,1); The sixth step is to construct the maximum likelihood estimation function in the process of constructing the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient according to the overall correlation function. The expression is: ， Among them, L represents the constructed maximum likelihood estimation function, and 1 represents the dimension. The unit vector of The dimension is The correlation coefficient matrix of represents its inverse matrix, represents pi, and They represent the mean and variance of the Gaussian process respectively, and det represents the determinant; Step 7: Correlation coefficient matrix in maximum likelihood estimation function The expression is as follows: ， in, represents the overall correlation function value between the first individual and itself, Indicates the first individual and the n The overall correlation function value of each individual, Indicates n The overall correlation function value of individuals and the first individual, Indicates n The overall correlation function value of each individual with itself; In the eighth step, the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient is derived. The corresponding expressions of the predicted value and predicted variance are: ， ， in, , , The Kriging machine learning model combining the maximum information coefficient with attention information enhancement is used for individual T The predicted value of ， The Kriging machine learning model combining the maximum information coefficient with attention information enhancement is used for individual T The prediction variance of Represents the Gaussian process mean corresponding to the Kriging machine learning model with maximum information coefficient combined with attention information enhancement, The Gaussian process variance corresponding to the Kriging machine learning model with maximum information coefficient combined with attention information enhancement; In the ninth step, in the Kriging machine learning model with attention information enhancement combined with the maximum information coefficient, the corresponding Gaussian process mean and variance calculation expressions are as follows: ， ， in, Represents the transpose of a unit vector.

5. The method according to claim 1, characterized in that The step (4) specifically includes the following steps: The first step is to sort the error target values ​​of all individuals in the database by descending rules, and select the top individuals in the database. p % individuals are used to construct the first subpopulation and temporary population; In the second step, the pdist2 function is used to calculate the Euclidean distance between each individual in the temporary population and other individuals in the temporary population to generate a local distance matrix; the minimum distance of each individual in the temporary population is calculated through the local distance matrix, and the negative value of the minimum distance is taken as the local diversity index. Its mathematical expression is: ， in, T i , T j They represent the temporary population i, j individuals, TPOP represents the temporary population, represents the temporary population i Individual and j The Euclidean distance of individuals; The third step is to merge the temporary population with the database to form a temporary database. The pdist2 function is used to calculate the Euclidean distance of each individual in the temporary database from other individuals in the temporary database to generate a global distance matrix. The minimum distance of each individual in the temporary database is calculated through the global distance matrix, and the negative value of the minimum distance is taken as the global diversity index. Its mathematical expression is: ， in, T i Indicates the temporary database i Individuals, T j Indicates the temporary database j individuals, DB means temporary database, Indicates the temporary database i Individual and j The Euclidean distance of individuals; The fourth step is to use the global diversity index in the non-dominated sorting method. Gdiv ( T i ) and local diversity indicators Ldiv ( T i ) for both targets N individuals to construct the second subpopulation.

6. The method according to claim 1, characterized in that The step (5) specifically includes the following steps: The first step is to generate a mutation probability for each optimized design parameter of each individual in the first subpopulation. P select , which is used to decide whether to disturb the optimization design parameter; for each selected optimization design parameter that needs to be disturbed, a random disturbance value is added to the optimization design parameter to generate M The extended subpopulation of candidate individuals, this random perturbation value is randomly drawn from the Gaussian distribution, and the mathematical expression of dimensional perturbation variation is: ， in, Indicates the design space k The perturbed value of the optimal design parameter, For the design space k The original values ​​of the optimization design parameters, For the design space k Random perturbation values ​​of the optimization design parameters; In the second step, the gradient of all individuals in the expanded subpopulation is calculated according to the Kriging machine learning model; the single step length of the gradient descent is calculated according to the standard deviation used in the dimensional perturbation variation, and the formula is as follows: ， Among them, SS represents the step size of the gradient descent step, The standard deviation set for the dimensional perturbation variation, is the set number of steps; The third step is to use gradient descent to generate The candidate offspring individuals with different step lengths are ranked, and the error target value is predicted by the Kriging machine learning model. All candidate offspring individuals are sorted according to the ascending sorting rule of the predicted error target value, and the top u The candidate offspring individuals are taken as the first type of candidate offspring individuals; Step 3: Execute the affine propagation clustering algorithm on all first-category candidate offspring individuals to obtain multiple clusters and their centers; select the optimal center according to the following rules: First, sort the first-category candidate offspring individuals in ascending order, and sort the first The first-class candidate offspring individuals are defined as the current excellent individuals; secondly, the cluster with the most excellent individuals determined by the affine propagation clustering algorithm is selected as the optimal cluster; thirdly, the center of the optimal cluster is taken as the optimal center; each first-class candidate offspring individual generates an optimal center, and the Kriging machine learning model is used to predict the optimal center, and the first-class candidate offspring individual with the smallest predicted target error value is selected as the first offspring individual.

7. The method according to claim 1, characterized in that In step (6), the specific steps are as follows: In the first step, for each individual of the second subpopulation, the first-level differential mutation uses the DE / current-to-rand / 2 mutation operation and the binomial crossover operation to obtain the candidate offspring population, where the mutation operator of the DE / current-to-rand / 2 mutation operation is expressed as follows: ， in, It is a mutant individual generated by the DE / current-to-rand / 2 mutation operation. is the current evolutionary individual, , and are three different individuals randomly selected from the second subpopulation. F 1 and F 2 are the two scaling factors used to control the vector; The second step is to predict the error target value and expected improvement value of each individual in the candidate offspring population through the Kriging machine learning model; The third step is to select the top candidates in the non-dominated sorting method using the individual error target value and expected improvement value as two goals. b % excellent candidate offspring population individuals; The fourth step is to use the second-level differential mutation to perform DE / current-to-best / 2 mutation operation and binomial crossover operation on each excellent candidate offspring population individual to obtain the second type of candidate offspring individual pool, where the mutation operator of the DE / current-to-best / 2 mutation operation is expressed as follows: ， in, It is a mutant individual generated by the differential evolution DE / current-to-best / 2 mutation operation. is the current evolutionary individual, is the best vector in the current population, and are two different individuals randomly selected from the current population, F 1 and F 2 are two scaling factors used to control the scaling of the vector; The fifth step is to select the best second-class candidate offspring individual from the second-class candidate offspring individual pool as the second offspring individual according to the error target value predicted by the Kriging machine learning model.

8. The method according to claim 1, characterized in that The specific steps of step (7) are as follows: The first step is to use optical simulation software to perform simulation analysis on the first and second offspring individuals and calculate the error target value; The second step is to store the first offspring individual and the second offspring individual and the corresponding error target value into the database to update the database; In the third step, if the number of simulation analyses consumed reaches the design cycle, the optimal infrared stealth material film thickness parameter value is output; otherwise, go to step (3) until the design cycle is reached.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images