CN116562124A - Method and device for predicting the influence of high-speed aircraft radome ablation on electromagnetic performance - Google Patents

Abstract

The invention discloses a method and device for predicting the influence of high-speed aircraft radome ablation on electromagnetic performance, including obtaining test data sets of two sub-stages of thermal ablation and thermal wave penetration in the ablation process and a plurality of different Fidelity test data set; based on the associated Gaussian surrogate model method, the lowest fidelity surrogate model is constructed using the above sub-stage level data set; based on the hierarchical kriging method, the whole system is gradually fused according to the fidelity from low to high level data set, and iteratively update the proxy model to obtain the target prediction proxy model; when receiving the corresponding input parameters of the radome ablation, the response prediction is performed through the target prediction proxy model, and the corresponding ablation electromagnetic performance prediction response value and predicted response value of the radome are obtained. error. Therefore, based on the experimental data from various sources, the proxy model of the influence of radome ablation on electromagnetic performance can be constructed more accurately, and then the rapid prediction of the influence of radome ablation on electromagnetic performance can be realized.

Description

Method and device for predicting the effect of high-speed aircraft radome ablation on electromagnetic performance

Technical Field

The present invention relates to the field of antenna technology, and in particular to a method and device for predicting the influence of ablation of a high-speed aircraft radome on electromagnetic performance.

Background Art

During the high-speed flight of the aircraft, due to the extremely high flight speed, the kinetic energy is converted into the internal energy of the air molecules, causing the temperature near the front end to rise, forming a high-temperature shock wave around the antenna cover that is much higher than the pyrolysis temperature of the cover material itself, resulting in the occurrence of thermal ablation; and after the antenna cover material is ablated, its physical structure is destroyed, so that the internal electromagnetic wave forms a thermal transmission wave when it penetrates the cover. At this time, the change in the thickness of the cover material will affect the beam pointing of the antenna, and the change in temperature will also affect the dielectric properties of the cover material. These will have a negative impact on the working performance of the antenna (including antenna impedance mismatch, antenna gain reduction and frequency deviation, etc.), and ultimately cause problems in signal transmission. Since the ablation of the antenna cover cannot be directly avoided, it is particularly important to analyze how the initial factors (such as flight speed, etc.) in the antenna cover ablation process affect the electromagnetic performance, so that the errors caused by ablation can be corrected to ensure the normal operation of the antenna.

However, due to the complexity and high cost of flight test implementation, large-scale implementation is not feasible, so current research generally relies on a large number of simulation tests for analysis. Most of the research focuses on simulating the conditions of the radome ablation physical process, such as calculating the thermal response of the thermal ablation stage through aerodynamic heating software, simulating the antenna array through the high-order moment method in the thermal wave transmission stage to optimize the radome, and analyzing the overall effect of the whole system through thermal and electrical joint simulation.

However, the above scheme only models from the perspective of simulation, relying on large-scale simulation calculations to perform calculations. It is unable to quickly predict the impact of antenna cover ablation on electromagnetic performance, and does not utilize information from actual experiments.

Summary of the invention

The present invention provides a method and device for predicting the influence of antenna cover ablation on electromagnetic performance of high-speed aircraft, which solves the technical problems that the existing scheme only models from the perspective of simulation, relies on large-scale simulation calculation to perform calculations, cannot quickly realize the prediction of the electromagnetic performance caused by antenna cover ablation, and lacks the technical problems of utilizing test data information from multiple sources.

The present invention provides a method for predicting the influence of high-speed aircraft radome ablation on electromagnetic performance, which is characterized by comprising:

Acquire the thermal ablation stage test data set, thermal wave transmission stage test data set and multiple full-system level test data sets of different fidelity during the ablation process of the radome of a high-speed aircraft;

Based on the associated Gaussian proxy model method, the thermal ablation stage test data set and the thermal wave transmission stage test data set are used to construct the lowest fidelity initial prediction proxy model;

Based on the stratified kriging method, the full system level test data set is gradually integrated from low to high according to the fidelity, and the initial prediction proxy model is iteratively updated to generate a target prediction proxy model;

When the ablation input parameters of the radome are received, response prediction is performed through the target prediction agent model to generate electromagnetic performance prediction response values and prediction errors.

Optionally, the step of constructing a minimum fidelity initial prediction proxy model based on the associated Gaussian proxy model method using the thermal ablation stage test data set and the thermal wave transmission stage test data set includes:

When receiving the deterministic test kernel function, a Gaussian process is used in combination with the deterministic test kernel function to respectively construct a thermal ablation stage model and a thermal wave transmission stage model;

According to the thermal ablation stage model and the thermal wave transmission stage model, an uncertainty propagation relation is constructed by using a correlation Gaussian proxy model method;

Substituting the test data set of the thermal ablation stage and the test data set of the thermal wave transmission stage into the uncertainty propagation relationship respectively, and calculating the response variance and the response mean;

The response variance and the response mean are used to generate an initial prediction proxy model with the lowest fidelity.

Optionally, the step of gradually fusing the full system-level test data set according to the fidelity from low to high based on the stratified kriging method, and iteratively updating the initial prediction proxy model with the lowest fidelity to generate the target prediction proxy model includes:

Building a deviation model between the initial prediction proxy model and the actual response based on the initial prediction proxy model;

Selecting a full-system-level test data set corresponding to the current minimum fidelity as a training data set;

Calculate multiple predicted response values corresponding to the training data set through the initial prediction agent model and construct a prediction vector;

Calculating the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the true response vector corresponding to the full system level test data set;

Update the initial prediction proxy model according to the optimal linear unbiased prediction value of the deviation model and the deviation coefficient to generate an intermediate prediction proxy model;

Determining whether there is any unselected system-level test data set;

If yes, the intermediate prediction proxy model is used as a new initial prediction proxy model, and the step of selecting the full-system-level test data set corresponding to the current minimum fidelity as the training data set is executed;

If not, the intermediate prediction proxy model at the current moment is determined as the target prediction proxy model.

Optionally, the step of calculating the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the true response vector corresponding to the full system-level test data set includes:

Constructing a correlation matrix of the full system level test data set and transforming it into an inverse matrix;

Transforming the prediction vector into a transposed matrix;

Determine a corresponding deviation coefficient using the inverse matrix, the transposed matrix, the true response vector and the predicted vector;

Calculating the multiplication value between the deviation coefficient and the prediction vector;

Calculating the difference between the true response vector and the product value corresponding to the full system level test data set;

The difference, the correlation matrix and the inverse matrix are used to determine the optimal linear unbiased prediction value of the deviation model.

Optionally, the step of updating the initial prediction proxy model according to the optimal linear unbiased prediction value of the deviation model and the deviation coefficient to generate an intermediate prediction proxy model comprises:

The deviation coefficient is combined with the initial prediction proxy model, and the optimal linear unbiased prediction value of the deviation model is added to generate an intermediate prediction proxy model;

Calculating the mean square error corresponding to the intermediate prediction proxy model;

The intermediate prediction proxy model is:

The mean square error is:

Among them, x is the input of the proxy model, is the predicted response value of the intermediate prediction agent model, ρ _n+1 is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n+1 (x) is the mean square error, is the correlation vector between the unknown full-system-level test data set and the known full-system-level test data set, R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the known full-system-level test data set, is the regression matrix.

The present invention provides a prediction device for the influence of ablation of a high-speed aircraft radome on electromagnetic performance, which is characterized by comprising:

The data set acquisition module is used to acquire the test data set of the thermal ablation stage, the thermal wave transmission stage and multiple full-system level test data sets of different fidelity during the ablation process of the radome of the high-speed aircraft;

A model building module, used to build a minimum fidelity initial prediction proxy model based on a correlated Gaussian proxy model method using the thermal ablation stage test data set and the thermal wave transmission stage test data set;

A model updating module, for gradually fusing the full system-level test data set from low to high fidelity based on a stratified kriging method, and iteratively updating the initial prediction proxy model to generate a target prediction proxy model;

The prediction module is used to perform response prediction through the target prediction agent model when receiving the ablation input parameters of the radome, and generate the electromagnetic performance prediction response value and prediction error.

Optionally, the model building module is specifically used to:

When receiving the deterministic test kernel function, a Gaussian process is used in combination with the deterministic test kernel function to respectively construct a thermal ablation stage model and a thermal wave transmission stage model;

According to the thermal ablation stage model and the thermal wave transmission stage model, an uncertainty propagation relation is constructed by using a correlation Gaussian proxy model method;

Substituting the test data set of the thermal ablation stage and the test data set of the thermal wave transmission stage into the uncertainty propagation relationship respectively, and calculating the response variance and the response mean;

The response variance and the response mean are used to generate an initial prediction proxy model with the lowest fidelity.

Optionally, the model updating module includes:

A deviation model building submodule, used for building a deviation model between the initial prediction proxy model and the actual response based on the initial prediction proxy model;

A data set selection submodule, used to select a full-system level test data set corresponding to the current minimum fidelity as a training data set;

A prediction vector calculation submodule, used to calculate a plurality of prediction response values corresponding to the training data set through the initial prediction proxy model and construct a prediction vector;

A parameter calculation submodule, used to calculate the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the real response vector corresponding to the full system level test data set;

A model updating submodule, used for updating the initial prediction proxy model according to the optimal linear unbiased prediction value of the deviation model and the deviation coefficient to generate an intermediate prediction proxy model;

A judgment submodule, used for judging whether there is any unselected system-level test data set;

A loop submodule, for: if yes, using the intermediate prediction proxy model as a new initial prediction proxy model, and jumping to the step of selecting the full-system-level test data set corresponding to the current minimum fidelity as a training data set;

The model determination submodule is used to determine the intermediate prediction proxy model at the current moment as the target prediction proxy model if not.

Optionally, the parameter calculation submodule is specifically used for:

Constructing a correlation matrix of the full system level test data set and transforming it into an inverse matrix;

Transforming the prediction vector into a transposed matrix;

Determine a corresponding deviation coefficient using the inverse matrix, the transposed matrix, the true response vector and the predicted vector;

Calculating the multiplication value between the deviation coefficient and the prediction vector;

Calculating the difference between the true response vector and the product value corresponding to the full system level test data set;

The difference, the correlation matrix and the inverse matrix are used to determine the optimal linear unbiased prediction value of the deviation model.

Optionally, the model updating submodule is specifically used for:

The deviation coefficient is combined with the initial prediction proxy model, and the optimal linear unbiased prediction value of the deviation model is added to generate an intermediate prediction proxy model;

Calculating the mean square error corresponding to the intermediate prediction proxy model;

The intermediate prediction proxy model is:

The mean square error is:

Among them, x is the input of the proxy model, is the predicted response value of the intermediate prediction agent model, ρ _n+1 is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n+1 (x) is the mean square error, is the correlation vector between the unknown full-system-level test data set and the known full-system-level test data set, R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the known full-system-level test data set, is the regression matrix.

It can be seen from the above technical solutions that the present invention has the following advantages:

The present invention obtains the test data sets of the two sub-stages of thermal ablation and thermal wave transmission in the ablation process and multiple test data sets of different fidelity at the whole system level; based on the associated Gaussian proxy model method, the minimum fidelity proxy model is constructed using the above sub-stage level data sets; based on the stratified kriging method, the whole system level data sets are gradually integrated from low to high fidelity, and the proxy model is iteratively updated to obtain the target prediction proxy model; when the corresponding input parameters of the radome ablation are received, the response prediction is performed through the target prediction proxy model to obtain the corresponding ablation electromagnetic performance prediction response value and prediction error of the radome. Therefore, based on the test data from multiple sources, a proxy model of the influence of radome ablation on electromagnetic performance is more accurately constructed, thereby realizing the rapid prediction of the influence of radome ablation on electromagnetic performance.

BRIEF DESCRIPTION OF THE DRAWINGS

In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

FIG1 is a flowchart of the steps of a method for predicting electromagnetic performance of radome ablation of an aircraft provided in Embodiment 1 of the present invention;

FIG2 is a flowchart of the steps of a method for predicting electromagnetic performance of radome ablation of an aircraft provided in a second embodiment of the present invention;

FIG3 is a flowchart of the steps of a method for constructing a consistency proxy model according to an embodiment of the present invention;

FIG4 is a flowchart of the steps of predicting electromagnetic performance of radome ablation of an aircraft according to an embodiment of the present invention;

FIG5 is a structural block diagram of a device for predicting electromagnetic performance of radome ablation of an aircraft provided in a third embodiment of the present invention.

DETAILED DESCRIPTION

The radome is a structural protection device located at the front radar of an aircraft. It is usually used to reduce the impact of various external factors on the radar antenna. Its main purpose is to protect the antenna from the influence of the external environment and has good electromagnetic wave penetration characteristics. The analysis process of radome ablation can be divided into two stages, namely the thermal ablation stage and the thermal wave penetration stage. Its complexity makes it difficult to directly model the whole body. It is necessary to consider not only the thermal and mechanical properties of the radome during the whole process, but also the dynamically changing electromagnetic properties. If the whole body is subjected to full-wave thermal-electrical-mechanical analysis and simulation, although the data accuracy is high, the workload is huge and a lot of computing resources are required; and although the staged simulation test has relatively less calculation, there is also a lack of some methods to link the two stages of simulation tests. Therefore, it is very necessary to seek a method that can accurately calculate this process at a lower cost and integrate the existing test data from various sources.

A proxy model is an approximate mathematical model established using the test data of a complex system to describe the relationship between the input and output of the complex system. It plays a very important role in many scientific and engineering fields. The traditional proxy model construction method is called the "All-In-One (AIO)" method. The AIO method regards the entire complex system as a black box, directly tests the entire system and builds a proxy model based on the test data. Traditional AIO modeling methods mainly include polynomial response surface method, support vector machine method, neural network, Kriging, etc. Among them, Kriging (also known as Gaussian process regression model) is suitable for nonlinear and small sample modeling. However, for the above-mentioned radome ablation system, the full system test is very expensive, so the amount of test samples that can be obtained is relatively small, but its fidelity is relatively high. In addition, the AIO method emphasizes that the whole is a black box function and ignores the correlation between each subsystem. Since the radome ablation analysis system is composed of two sub-stages (thermal ablation stage and thermal wave transmission stage) connected in series, they are only unidirectionally related through input and output, so it is possible to consider building a proxy model through the simulation test data of each stage that is relatively easy to obtain. This kind of agent model that couples the agent models of each stage through analytical methods to construct an agent model for the entire system is called an associated agent model.

The embodiments of the present invention provide a method and device for predicting the influence of antenna cover ablation on electromagnetic performance of a high-speed aircraft, which is used to solve the problem that the existing solutions only perform modeling from the perspective of simulation, rely on large-scale simulation calculations for calculation, cannot quickly realize the prediction of the electromagnetic performance caused by antenna cover ablation, and lack the technical problem of fusing test data information from multiple sources.

In order to make the purpose, features and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below 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.

Please refer to FIG. 1 , which is a flowchart of the steps of a method for predicting electromagnetic performance of radome ablation of an aircraft provided in a first embodiment of the present invention.

The present invention provides a method for predicting electromagnetic performance of ablation of a radome of an aircraft, comprising:

Step 101, obtaining a test data set of a thermal ablation phase, a test data set of a thermal wave transmission phase, and a plurality of full-system level test data sets of different fidelity during the ablation process of a radome of a high-speed aircraft;

In the specific implementation, the factors affecting the electromagnetic performance of the radome ablation at each stage can be analyzed first. For example, in the thermal ablation stage, that is, the process in which the aircraft radome is ablated during high-speed flight, the input variables considered are the incoming flow velocity x ₁ , the incoming flow static temperature, the incoming flow static pressure x ₃ , the surface roughness x ₄ , the material density x ₅ , the material specific heat x ₆ , the surface radiation coefficient x ₇ , the thermal conductivity x ₈ , and the liquid layer viscosity coefficient x _9. These variables all act on this thermal ablation process. Therefore, the output variables are the temperature distribution _w1 and the thickness after ablation _w2 ; in the thermal wave transmission stage, that is, the process in which various parameters of the antenna cover after ablation affect its electromagnetic properties, the input variables include external variables - frequency z and the output of the thermal ablation stage, among which the temperature distribution _w1 affects the dielectric constant g( _w1 ) through a known mapping g, and the thickness _w2 after ablation. These three variables act together on this thermal wave transmission process Finally, the output variables, reflection coefficient _y1 and transmission coefficient _y2, which can reflect the electrical properties, are obtained.

In an embodiment of the present invention, in order to obtain a data basis for predicting the electromagnetic performance of the radome ablation, a sub-stage test data set and a plurality of full-system level test data sets with different fidelity may be first obtained.

The sub-stage test data set includes the sub-stage test data set corresponding to the thermal ablation stage and the sub-stage test data set corresponding to the thermal wave transmission stage, that is, in The experimental data set representing the thermal ablation phase, The test data set refers to the test data set of the whole system. The fidelity is sorted from low to high, with the highest fidelity being 100%, which is the real test data at the full system level.

Step 102, based on the associated Gaussian proxy model method, using the thermal ablation stage test data set and the thermal wave transmission stage test data set to construct a minimum fidelity initial prediction proxy model;

In an embodiment of the present invention, after obtaining the sub-stage test data set, a Gaussian process model of the thermal ablation stage and a Gaussian process model of the thermal transmission wave stage can be established for the data set. After analyzing the uncertainty propagation relationship using the associated Gaussian proxy model method, the response variance and response mean are calculated in combination with the sub-stage test data set, thereby obtaining the lowest fidelity prediction proxy model of the entire system through Gaussian process modeling.

It should be noted that the thermal ablation stage model refers to a mathematical model that reflects the thermal ablation process of the antenna cover of a high-speed aircraft during high-speed flight. The thermal wave transmission stage model refers to a mathematical model that reflects the process in which various parameters of the antenna cover after ablation affect its electromagnetic properties. The initial prediction proxy model refers to a proxy model that describes the relationship between the input and output of the entire system and is established by calculating the response (output) of the original model through a finite number of points (input).

Step 103, based on the stratified kriging method, gradually integrate the whole system level test data set from low to high fidelity, and iteratively update the initial prediction proxy model to generate a target prediction proxy model;

After the initial prediction proxy model is established, the full-system test datasets from various sources can be integrated on its basis to update the constructed proxy model and further improve its prediction performance. At this time, each full-system-level test dataset can be integrated in sequence from low to high in terms of fidelity, and the prediction proxy model can be iteratively updated using the full-system-level test dataset to generate a target prediction proxy model.

Step 104 , when receiving the ablation input parameters of the radome, a response prediction is performed through the target prediction agent model to generate an electromagnetic performance prediction response value and a prediction error.

After obtaining the target prediction proxy model, if the corresponding input parameters of antenna cover ablation are received at this time, the target prediction proxy model can be used to perform response prediction to determine the predicted response value of the influence of antenna cover ablation on electromagnetic performance under the input, and the prediction error can be calculated to determine the uncertainty of the predicted response value of the electromagnetic performance.

In the embodiment of the present invention, the test data sets of the two sub-stages of thermal ablation and thermal wave transmission in the ablation process and the test data sets of multiple different fidelity at the whole system level are obtained; based on the associated Gaussian proxy model method, the minimum fidelity proxy model is constructed using the above sub-stage level data sets; based on the stratified kriging method, the whole system level data sets are gradually integrated from low to high fidelity, and the proxy model is iteratively updated to obtain the target prediction proxy model; when the corresponding input parameters of the radome ablation are received, the response prediction is performed through the target prediction proxy model to obtain the corresponding ablation electromagnetic performance prediction response value and prediction error of the radome. Therefore, based on the test data from multiple sources, a proxy model of the influence of radome ablation on electromagnetic performance is more accurately constructed, thereby realizing the rapid prediction of the influence of radome ablation on electromagnetic performance.

Please refer to FIG. 2 , which is a flow chart showing the steps of a method for predicting electromagnetic performance of radome ablation of an aircraft provided in a second embodiment of the present invention.

The present invention provides a method for predicting electromagnetic performance of ablation of a radome of an aircraft, comprising:

Step 201, obtaining a test data set of a thermal ablation phase, a test data set of a thermal wave transmission phase, and a plurality of full-system level test data sets of different fidelity during the ablation process of a radome of a high-speed aircraft;

In the embodiment of the present invention, since the radome will experience a thermal ablation stage and a thermal wave transmission stage during high-speed flight, a thermal ablation stage test data set and a thermal wave transmission stage test data set of the radome of the high-speed aircraft during the ablation process can be obtained. The thermal ablation stage test data set includes input variables in the thermal ablation stage, that is, incoming flow velocity x ₁ , incoming flow static temperature x ₂ , incoming flow static pressure x ₃ , surface roughness x ₄ , material density x ₅ , material specific heat x ₆ , surface radiation coefficient x ₇ , thermal conductivity x ₈ , liquid layer viscosity coefficient x ₉ , and output variables temperature distribution w ₁ and ablation thickness w ₂ obtained after the thermal ablation stage is completed; the thermal ablation stage test data set includes input variables and output variables of the thermal wave transmission stage, that is, the input variable is the dielectric constant g(w ₁ ) generated by the temperature distribution w ₁ through a known mapping g, the external input frequency z and the ablation thickness w ₂ , and the output variables are the reflection coefficient y ₁ and the transmission coefficient y ₂ that reflect the electromagnetic properties of the radome. .

Step 202, when the deterministic test kernel function is received, a Gaussian process is used in combination with the deterministic test kernel function to respectively construct a thermal ablation stage model and a thermal wave transmission stage model;

In the embodiment of the present invention, when a deterministic test kernel function is received, such as a normal kernel function or a Laplace function, a Gaussian process combined with a deterministic test kernel function can be used to construct a thermal ablation stage model and a thermal wave transmission stage model. The thermal ablation stage model is a model constructed to reflect the process of ablation of the aircraft antenna cover during high-speed flight. In the case of considering the observation error e _i and the truncation error u _i , the relationship between the input variable and the output variable can be modeled as:

Where x = [x ₁ ,…,x ₉ ], μ _i (x) is the mean function, is the model error that follows a normal distribution.

The thermal wave transmission stage model is a model that reflects the process in which various parameters of the radome after ablation affect its electromagnetic properties. The relationship between the input variables and the output variables can be modeled as follows:

where μ _j (x) is the mean function, is the model error that follows a normal distribution g(w ₁ ) is the dielectric constant.

The above sub-stage models can be tested through the sub-stage test dataset The solution is obtained.

Step 203, constructing an uncertainty propagation relational expression based on the thermal ablation stage model and the thermal wave transmission stage model by using a correlation Gaussian proxy model method;

The uncertainty propagation relationship is:

y～N(E(y),D(y)),y={y ₁ ,y ₂ }

There is no coupling relationship between the input variables of the two stages. At this time, the above uncertainty propagation relationship is inferred as:

Thus, the relationship between the output y = [y ₁ , y ₂ ] of the second layer of the system, the input x = [x ₁ , ..., x ₉ ] of the first layer, and the external input z of the second layer is: in It means that the expectation of the output variable _yj is a function determined only by the input variables (x, z). It means that the variance of the output variable _yj is a function determined only by the input variables (x, z).

Step 204, using the test data set of the thermal ablation stage and the test data set of the thermal wave transmission stage to substitute into the uncertainty propagation relationship respectively, and calculating the response variance and the response mean;

Step 205 , using the response variance and the response mean, generates an initial prediction proxy model with the lowest fidelity.

Use the response variance and response mean to generate the lowest fidelity initial prediction proxy model Among them, X = {x, z} is the input of the whole system, E (X) is the response mean, and D (X) is the response variance. In the specific implementation, according to the expression form of the kernel function selected in the Gaussian process and whether the experiment is a deterministic experiment, the calculation method is as follows:

a) In the deterministic test, the kernel function is selected as the normal kernel, Laplace kernel or exponential kernel and the smoothness index in the kernel function is 1 or 2, and the expressions of E(y) and D(y) can be directly calculated.

b) In other cases, numerical simulation methods such as Monte Carlo can be used for calculation.

Step 206, based on the stratified kriging method, gradually fuse the whole system level test data set from low to high fidelity, and iteratively update the initial prediction proxy model to generate a target prediction proxy model;

Optionally, step 206 may include the following sub-steps S11-S18:

S11, constructing a deviation model between the initial prediction proxy model and the actual response;

In the embodiment of the present invention, in order to further improve the prediction accuracy of the initial prediction proxy model, the full system level data set D _m , m = 1, ..., n can be gradually integrated from low to high fidelity, and the proxy model can be iteratively updated to be the target prediction proxy model. First, the full system level test data set D ₁ corresponding to the minimum fidelity at the current moment can be selected as the training data set.

Among them, the proxy model _M1 and _M0 obey the following deviation model:

in represents the value of model _M1 at variable x, represents the value of the initial prediction agent model M ₀ at the variable x, ρ ₁ is the bias factor, z ₁ (x)~GP(0,k ₁ (.,.)) is a zero-mean Gaussian process, and

S12, selecting the full system-level test data set corresponding to the current minimum fidelity as the training data set;

The full-system test data set corresponding to the current minimum fidelity is selected as the training data set. The training data set is input into the initial prediction agent model to calculate the corresponding multiple prediction response values and construct the prediction vector in Refers to the initial prediction proxy model in the variable The predicted response value under .

S13, calculating multiple predicted response values corresponding to the training data set through the initial prediction agent model and constructing a prediction vector;

S14, calculating the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the actual response vector corresponding to the full system-level test data set;

Further, step S14 may include the following sub-steps:

Construct the correlation matrix of the full system-level test data set and transform it into an inverse matrix;

Transform the prediction vector into a transposed matrix;

Using the inverse matrix, transposed matrix, true response vector, and predicted vector, determine the corresponding deviation coefficient;

Calculate the multiplication between the deviation coefficient and the prediction vector;

Calculate the difference between the true response vector and the product value corresponding to the full system level test data set;

Using differences, correlation matrices, and inverse matrices, determine the optimal linear unbiased prediction value of the bias model;

In the embodiment of the present invention, the correlation matrix R ₁ corresponding to the full system-level test data set is constructed, which is an N ₁ ×N ₁ correlation matrix R( _xi , _xj ), and is further transformed into an inverse matrix. At the same time, the prediction vector is transformed into a transposed matrix Use the inverse matrix, transposed matrix, true response vector and predicted vector to determine the corresponding deviation coefficient in, is the true response vector for full system level testing.

Furthermore, the multiplication value between the deviation coefficient and the prediction vector is calculated, and the deviation model between the actual response vector and the multiplication value corresponding to the full system-level test data set is calculated to obtain Combine the correlation matrix and the inverse matrix to determine the optimal linear unbiased predictor of the bias model Among them, is the correlation vector between the unknown sample and the training sample.

The intermediate prediction agent model is:

The mean square error is:

in, is the predicted response value of the intermediate prediction agent model, _ρn is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n (x) is the mean square error, is the correlation vector between the unknown sample and the training sample, _Rn is the _Nn × _Nn correlation matrix R( _xi , _xj ) of the training sample, is the regression matrix.

S15, updating the initial prediction proxy model according to the optimal linear unbiased prediction value and the deviation coefficient of the deviation model to generate an intermediate prediction proxy model;

Furthermore, S15 may include the following sub-steps:

The bias coefficient is combined with the initial prediction proxy model, and the optimal linear unbiased prediction value of the bias model is added to generate the intermediate prediction proxy model;

Calculate the mean square error corresponding to the intermediate prediction proxy model;

The intermediate prediction agent model is:

The mean square error is:

Among them, x is the input of the proxy model, is the predicted response value of the intermediate prediction agent model, ρ _n+1 is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n+1 (x) is the mean square error, is the correlation vector between the unknown full-system-level test data set and the known full-system-level test data set, R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the known full-system-level test data set, is the regression matrix.

S16, determining whether there is an unselected full system level test data set;

S17, if yes, use the intermediate prediction proxy model as a new initial prediction proxy model, and jump to the step of selecting the full system-level test data set corresponding to the current minimum fidelity as the training data set;

S17: If not, the intermediate prediction proxy model at the current moment is determined as the target prediction proxy model.

In an embodiment of the present invention, after obtaining the intermediate prediction proxy model, it is possible to further determine whether there is a full-system-level test data set that has not been selected. If so, the selection is performed again for iterative updating until the real test data, that is, the full-system-level test data set with a fidelity of 100%, is selected. If not, it indicates that the intermediate prediction proxy model at this time has completed training, and the intermediate prediction proxy model at the current moment can be determined as the target prediction proxy model.

In the specific implementation, the construction of the target prediction agent model is realized through the following examples:

Based on the full-system simulation test data D _m , m＝1,…,n of the radome ablation electrical performance analysis system with different fidelity, the proxy model M _n is gradually established.

Assume that there are n full-system level simulation experiments with different fidelity (ie, antenna cover ablation electrical performance simulation analysis systems with different precisions), and the test fidelity is from low to high, and D ₁ ,D ₂ ,...,D _n are the corresponding data sets.

1) Establish a fusion framework of proxy models _M1 and _M0 ;

Assume that the proxy models _M1 and _M0 obey the following deviation model:

in represents the value of model _M1 at variable x, represents the value of the model M ₀ established in step 205 at the variable x, ρ ₁ is the deviation factor, z ₁ (x) ～ GP (0, k ₁ (.,.)) is a zero-mean Gaussian process, and

Based on the established surrogate model M ₀ , calculate the predicted value of the input data in D ₁

Remember the vector Then the best linear unbiased prediction (BLUP) value of the deviation model z ₁ (x) is obtained as coefficient in R ₁ is the N ₁ ×N ₁ correlation matrix R( _xi , _xj ) of the training samples. is the regression matrix, is the response value of the full system level test.

Finally, we get the proxy model M ₁ : in:

2) Integrate the whole system level experimental data D ₂ with the proxy model M ₁ to establish the proxy model M ₂ : The specific process is similar to 1).

After gradually integrating data D ₃ , D ₄ , ..., D _n with different test accuracies, the proxy model is obtained:

Based on the real test data D _n+1 of the radome ablation electrical performance analysis system, a proxy model M _n+1 based on multi-source test data is established.

3) Establish a fusion framework of models M _n+1 and M _n ;

Assume that the model _Mn+1 and _Mn obey the following deviation model: Among them: z _n+1 (x)～GP(0,k _n+1 (.,.)) and Using the trained proxy model M _n , the prediction data set is calculated.

Remember the vector Then the best linear unbiased prediction (BLUP) value of the deviation model z _n+1 (x) is obtained as Deviation Factor in R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the training samples. is the regression matrix, is the response value of the real test. Finally, the consistency proxy model of multi-source test data based on the antenna cover ablation electrical performance analysis system is obtained.

Step 207 , when receiving the ablation input parameters of the radome, performing response prediction through the target prediction agent model, and generating electromagnetic performance prediction response value and prediction error.

In the embodiment of the present invention, after obtaining the input parameters corresponding to the radome ablation, that is, the new experimental design point x ₀ , the estimated value thereof can be predicted by the proxy model calculation: Its prediction error (i.e. uncertainty) is

The present invention uses mean absolute error (MAE) and root mean square error (RMSE) to measure the error of the proxy model, which are specifically defined as follows:

First, for the thermal ablation stage, the input variables are the incoming flow velocity, incoming flow static temperature, incoming flow static pressure, surface roughness, material density, material specific heat, surface emissivity, thermal conductivity and liquid layer viscosity coefficient, and the output response is the ablation thickness and temperature value. 1000 test points are obtained for this stage through simulation software, and the proxy model is constructed using Gaussian process. Finally, another 400 test points are used to verify the proxy model for this stage, and the error table between the predicted value of the proxy model and the true value is shown in Table 1.

Secondly, for the thermal wave transmission stage, the input variables are the output variables of the thermal ablation stage, the ablation thickness and temperature value, plus the external variables that jointly affect the reflection coefficient and the transmission coefficient. Through the simulation software, 600 test points were obtained for this stage, and the Gaussian process was used to build the proxy model. Finally, another 200 test points were used to verify the proxy model of this stage, and the error table between the predicted value of the proxy model and the true value was obtained as shown in Table 1.

Table 1 Error table of the associated proxy model constructed based on sub-stage test data

Temperature distribution w ₁ Thickness after ablation w ₂ Reflection coefficient y ₁ Transmission coefficient y ₂ MAE 0.0075 0.1704 0.0128 0.0142 RMSE 0.0174 0.3754 0.0193 0.0222

Finally, the above proxy model is integrated with a set of simulation test data of the whole system. The error table between the predicted value of the proxy model and the true value is shown in Table 2:

Table 2 Error table of multi-fidelity proxy model based on full system level and sub-stage associated proxy model

Reflection coefficient y ₁ Transmission coefficient y ₂ MAE 0.0010 9.3691e-4 RMSE 0.0015 0.0013

Judging from the error of the proxy model, the associated proxy model constructed based on the sub-stage test data has a sufficiently high accuracy, reflecting the accuracy of the associated proxy model method for constructing the antenna cover ablation system of the present invention; in addition, it can be seen that the error of the multi-fidelity proxy model constructed based on test data from different sources is relatively smaller, indicating the accuracy and reliability of the multi-fidelity proxy model method that integrates multi-source test data.

Please refer to FIG. 3 , which shows a flowchart of the steps of a method for constructing a consistency proxy model according to an embodiment of the present invention.

Assume that there are n full-system-level simulation experiments with different fidelity (ie, antenna cover ablation electrical performance simulation analysis systems with different precisions), and the precision of the test data is from low to high, and D ₁ ,D ₂ ,...,D _n are the corresponding data sets.

1) Establish a fusion framework of proxy models _M1 and _M0 ;

Assume that the proxy models _M1 and _M0 obey the following deviation model:

in represents the value of model _M1 at variable x, represents the value of the model M ₀ established in step 301 at the variable x, ρ ₁ is the deviation factor, z ₁ (x) ～ GP (0, k ₁ (.,.)) is a zero-mean Gaussian process, and

Based on the proxy model M ₀ established in step 205 , the predicted value of the input data in D ₁ is calculated

Remember the vector Then the best linear unbiased prediction (BLUP) value of the deviation model z ₁ (x) is obtained as coefficient in R ₁ is the N ₁ ×N ₁ correlation matrix of the training samples. is the regression matrix, is the response value of the full system level test.

Finally, we get the proxy model M ₁ : in:

,

2) Integrate the whole system level experimental data D ₂ with the proxy model M ₁ to establish the proxy model M ₂ : The specific process is similar to 1).

After gradually integrating data D ₃ , D ₄ , ..., D _n with different test accuracies, the proxy model is obtained:

1) Establish the fusion framework of model R( _xi , _xj ).Mn ₊₁ . and _Mn ;

Assume that the model _Mn+1 and _Mn obey the following deviation model: Among them: z _n+1 (x)～GP(0,k _n+1 (.,.)) and

Using the trained proxy model M _n , calculate the prediction data set

Remember the vector Then the best linear unbiased prediction (BLUP) value of the deviation model z _n+1 (x) is obtained as Deviation Factor in R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the training samples. is the regression matrix, is the response value of the real test. Finally, the consistency proxy model of multi-source test data based on the antenna cover ablation electrical performance analysis system is obtained. in

,

Please refer to FIG. 4 , which shows a flowchart of the steps of predicting electromagnetic performance of radome ablation of an aircraft according to an embodiment of the present invention.

Step 1: Analysis of system influencing factors: Analyze the influencing factors at each stage of the radome ablation electromagnetic performance analysis system.

Step 2: Obtaining test data: The present invention considers the following three types of test data from different sources. The first is the test data generated by simulation tests in different sub-stages. in Y _l ⁰ represents the input and output of the system as a whole, W _l ⁰ represents the output of the middle layer; secondly, the simulation test data of different fidelity of the whole system Finally, the real test data of the whole system (If actually permitted, such as a full flight test).

Step 3: Construction of proxy model

Step 301: constructing a proxy model based on sub-stage test data of the radome ablation electrical performance analysis system;

Step 302: gradually establish a proxy model _Mn based on the whole system different fidelity simulation test data _Dm , m=1, ..., n of the radome ablation electrical performance analysis system.

1) Establish a fusion framework of proxy models _M1 and _M0 ;

2) Integrate the whole system level experimental data D ₂ with the proxy model M ₁ to establish the proxy model M ₂ : The specific process is similar to 1).

Step 303: Based on the real test data D _n+1 of the radome ablation electrical performance analysis system, a proxy model M _n+1 based on multi-source test data is established.

Step 4: Use the established proxy model to predict the electrical performance of the radome ablation system

The electrical performance proxy model Mn ₊₁ of the radome ablation system based on multi-source test data established in step 3 is used to perform calculations and predictions, and the uncertainty of the prediction is given. For the new test design point _x0 , the estimated value can be calculated and predicted by the proxy model as Its prediction error (i.e. uncertainty) is

In the embodiment of the present invention, the test data sets of the two sub-stages of thermal ablation and thermal wave transmission in the ablation process and the test data sets of multiple different fidelity at the whole system level are obtained; based on the associated Gaussian proxy model method, the minimum fidelity proxy model is constructed using the above sub-stage level data sets; based on the stratified kriging method, the whole system level data sets are gradually integrated from low to high fidelity, and the proxy model is iteratively updated to obtain the target prediction proxy model; when the corresponding input parameters of the radome ablation are received, the response prediction is performed through the target prediction proxy model to obtain the corresponding ablation electromagnetic performance prediction response value and prediction error of the radome. Therefore, based on the test data from multiple sources, a proxy model of the influence of radome ablation on electromagnetic performance is more accurately constructed, thereby realizing the rapid prediction of the influence of radome ablation on electromagnetic performance.

Please refer to FIG. 5 , which is a structural block diagram of a device for predicting electromagnetic performance of radome ablation of an aircraft according to a third embodiment of the present invention.

The present invention also provides a device for predicting the influence of high-speed aircraft radome ablation on electromagnetic performance, comprising:

The data set acquisition module 501 is used to acquire a test data set of a thermal ablation phase, a test data set of a thermal wave transmission phase, and a plurality of full-system level test data sets of different fidelity during the ablation process of the radome of the high-speed aircraft;

A model building module 502 is used to build a minimum fidelity initial prediction proxy model based on a correlation Gaussian proxy model method using a thermal ablation phase test data set and a thermal wave transmission phase test data set;

The model updating module 503 is used to gradually fuse the whole system level test data set from low to high fidelity based on the stratified kriging method, and iteratively update the initial prediction proxy model to generate a target prediction proxy model;

The prediction module 504 is used to perform response prediction through the target prediction agent model when receiving the ablation input parameters of the radome, and generate the electromagnetic performance prediction response value and prediction error.

Optionally, the model building module 502 is specifically used for:

When the deterministic test kernel function is received, a Gaussian process is used in combination with the deterministic test kernel function to respectively construct a thermal ablation stage model and a thermal wave transmission stage model;

According to the thermal ablation stage model and the thermal wave transmission stage model, the uncertainty propagation relationship is constructed by using the associated Gaussian proxy model method.

The test data sets of thermal ablation stage and thermal wave penetration stage were substituted into the uncertainty propagation equation to calculate the response variance and response mean.

The response variance and response mean are used to generate the lowest fidelity initial prediction surrogate model.

Optionally, the model updating module 503 includes:

A deviation model building submodule is used to build a deviation model between the initial prediction proxy model and the actual response based on the initial prediction proxy model;

The data set selection submodule is used to select the full system-level test data set corresponding to the current minimum fidelity as the training data set;

A prediction vector calculation submodule is used to calculate multiple prediction response values corresponding to the training data set through the initial prediction agent model and construct a prediction vector;

A parameter calculation submodule is used to calculate the optimal linear unbiased prediction value and deviation coefficient of the deviation model based on the prediction vector and the real response vector corresponding to the full system level test data set;

The model updating submodule is used to update the initial prediction proxy model according to the optimal linear unbiased prediction value and the deviation coefficient of the deviation model to generate an intermediate prediction proxy model;

A judgment submodule, used to judge whether there is an unselected full system level test data set;

a loop submodule, for, if yes, using the intermediate prediction proxy model as a new initial prediction proxy model, and jumping to the step of selecting a full-system-level test data set corresponding to the current minimum fidelity as a training data set;

The model determination submodule is used to determine the intermediate prediction proxy model at the current moment as the target prediction proxy model if not.

Optionally, the parameter calculation submodule is specifically used for:

Construct the correlation matrix of the full system-level test data set and transform it into an inverse matrix;

Transform the prediction vector into a transposed matrix;

Using the inverse matrix, transposed matrix, true response vector, and predicted vector, determine the corresponding deviation coefficient;

Calculate the multiplication between the deviation coefficient and the prediction vector;

Calculate the difference between the true response vector and the product value corresponding to the full system level test data set;

The optimal linear unbiased prediction value of the bias model is determined using the difference, correlation matrix and inverse matrix.

Optionally, the model updating submodule is specifically used for:

The deviation coefficient is combined with the initial prediction proxy model, and the optimal linear unbiased prediction value of the deviation model is added to generate an intermediate prediction proxy model; the mean square error corresponding to the intermediate prediction proxy model is calculated;

The intermediate prediction agent model is:

The mean square error is:

Among them, x is the input of the proxy model, is the predicted response value of the intermediate prediction agent model, ρ _n+1 is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n+1 (x) is the mean square error, is the correlation vector between the unknown full-system-level test data set and the known full-system-level test data set, R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the known full-system-level test data set, is the regression matrix.

Those skilled in the art can clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described devices, modules and sub-modules can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

In the several embodiments provided by the present invention, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some interfaces, indirect coupling or communication connection of devices or units, which can be electrical, mechanical or other forms.

The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

As described above, the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that the technical solutions described in the aforementioned embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims (10)

1. A method for predicting the effect of high-speed aircraft radome ablation on electromagnetic performance, comprising: Acquire the thermal ablation stage test data set, thermal wave transmission stage test data set and multiple full-system level test data sets of different fidelity during the ablation process of the radome of a high-speed aircraft; Based on the associated Gaussian proxy model method, the thermal ablation stage test data set and the thermal wave transmission stage test data set are used to construct the lowest fidelity initial prediction proxy model; Based on the stratified kriging method, the full system level test data set is gradually integrated from low to high according to the fidelity, and the initial prediction proxy model is iteratively updated to generate a target prediction proxy model; When the ablation input parameters of the radome are received, response prediction is performed through the target prediction agent model to generate electromagnetic performance prediction response values and prediction errors. 2. The method according to claim 1, characterized in that the step of constructing the minimum fidelity initial prediction proxy model based on the correlation Gaussian proxy model method using the thermal ablation stage test data set and the thermal wave transmission stage test data set comprises: When receiving the deterministic test kernel function, a Gaussian process is used in combination with the deterministic test kernel function to respectively construct a thermal ablation stage model and a thermal wave transmission stage model; According to the thermal ablation stage model and the thermal wave transmission stage model, an uncertainty propagation relation is constructed by using a correlation Gaussian proxy model method; Substituting the test data set of the thermal ablation stage and the test data set of the thermal wave transmission stage into the uncertainty propagation relationship respectively, and calculating the response variance and the response mean; The response variance and the response mean are used to generate an initial prediction proxy model with the lowest fidelity. 3. The method according to claim 1, characterized in that the step of gradually fusing the full system-level test data set according to the fidelity from low to high based on the stratified kriging method, and iteratively updating the initial prediction proxy model with the lowest fidelity to generate the target prediction proxy model comprises: Building a deviation model between the initial prediction proxy model and the actual response based on the initial prediction proxy model; Selecting a full-system-level test data set corresponding to the current minimum fidelity as a training data set; Calculate multiple predicted response values corresponding to the training data set through the initial prediction agent model and construct a prediction vector; Calculating the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the true response vector corresponding to the full system level test data set; Update the initial prediction proxy model according to the optimal linear unbiased prediction value of the deviation model and the deviation coefficient to generate an intermediate prediction proxy model; Determining whether there is any unselected system-level test data set; If yes, the intermediate prediction proxy model is used as a new initial prediction proxy model, and the step of selecting the full-system-level test data set corresponding to the current minimum fidelity as the training data set is executed; If not, the intermediate prediction proxy model at the current moment is determined as the target prediction proxy model. 4. The method according to claim 3, characterized in that the step of calculating the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the true response vector corresponding to the full system level test data set comprises: Constructing a correlation matrix of the full system level test data set and transforming it into an inverse matrix; Transforming the prediction vector into a transposed matrix; Determine a corresponding deviation coefficient using the inverse matrix, the transposed matrix, the true response vector and the predicted vector; Calculating the multiplication value between the deviation coefficient and the prediction vector; Calculating the difference between the true response vector and the product value corresponding to the full system level test data set; The difference, the correlation matrix and the inverse matrix are used to determine the optimal linear unbiased prediction value of the deviation model. 5. The method according to claim 3, characterized in that the step of updating the initial prediction proxy model according to the optimal linear unbiased prediction value of the deviation model and the deviation coefficient to generate an intermediate prediction proxy model comprises: The deviation coefficient is combined with the initial prediction proxy model, and the optimal linear unbiased prediction value of the deviation model is added to generate an intermediate prediction proxy model; Calculating the mean square error corresponding to the intermediate prediction proxy model; The intermediate prediction proxy model is: The mean square error is: Among them, x is the input of the proxy model, is the predicted response value of the intermediate prediction agent model, ρ _n+1 is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n+1 (x) is the mean square error, is the correlation vector between the unknown full-system-level test data set and the known full-system-level test data set, R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the known full-system-level test data set, is the regression matrix. 6. A device for predicting the effect of high-speed aircraft radome ablation on electromagnetic performance, comprising: The data set acquisition module is used to acquire the test data set of the thermal ablation stage, the thermal wave transmission stage and multiple full-system level test data sets of different fidelity during the ablation process of the radome of the high-speed aircraft; A model building module, used to build a minimum fidelity initial prediction proxy model based on a correlated Gaussian proxy model method using the thermal ablation stage test data set and the thermal wave transmission stage test data set; A model updating module, for gradually fusing the full system-level test data set from low to high fidelity based on a stratified kriging method, and iteratively updating the initial prediction proxy model to generate a target prediction proxy model; The prediction module is used to perform response prediction through the target prediction agent model when receiving the ablation input parameters of the radome, and generate the electromagnetic performance prediction response value and prediction error. 7. The device according to claim 6, characterized in that the model building module is specifically used for: When receiving the deterministic test kernel function, a Gaussian process is used in combination with the deterministic test kernel function to respectively construct a thermal ablation stage model and a thermal wave transmission stage model; According to the thermal ablation stage model and the thermal wave transmission stage model, an uncertainty propagation relation is constructed by using a correlation Gaussian proxy model method; Substituting the test data set of the thermal ablation stage and the test data set of the thermal wave transmission stage into the uncertainty propagation relationship respectively, and calculating the response variance and the response mean; The response variance and the response mean are used to generate an initial prediction proxy model with the lowest fidelity. 8. The device according to claim 6, characterized in that the model updating module comprises: A deviation model building submodule, used for building a deviation model between the initial prediction proxy model and the actual response based on the initial prediction proxy model; A data set selection submodule, used to select a full-system level test data set corresponding to the current minimum fidelity as a training data set; A prediction vector calculation submodule, used to calculate a plurality of prediction response values corresponding to the training data set through the initial prediction proxy model and construct a prediction vector; A parameter calculation submodule, used to calculate the optimal linear unbiased prediction value and the deviation coefficient of the deviation model based on the prediction vector and the real response vector corresponding to the full system level test data set; A model updating submodule, used for updating the initial prediction proxy model according to the optimal linear unbiased prediction value of the deviation model and the deviation coefficient to generate an intermediate prediction proxy model; A judgment submodule, used for judging whether there is any unselected system-level test data set; A loop submodule, for: if yes, using the intermediate prediction proxy model as a new initial prediction proxy model, and jumping to the step of selecting the full-system-level test data set corresponding to the current minimum fidelity as a training data set; The model determination submodule is used to determine the intermediate prediction proxy model at the current moment as the target prediction proxy model if not. 9. The device according to claim 8, characterized in that the parameter calculation submodule is specifically used for: Constructing a correlation matrix of the full system level test data set and transforming it into an inverse matrix; Transforming the prediction vector into a transposed matrix; Determine a corresponding deviation coefficient using the inverse matrix, the transposed matrix, the true response vector and the predicted vector; Calculating the multiplication value between the deviation coefficient and the prediction vector; Calculating the difference between the true response vector and the product value corresponding to the full system level test data set; The difference, the correlation matrix and the inverse matrix are used to determine the optimal linear unbiased prediction value of the deviation model. 10. The device according to claim 8, characterized in that the model updating submodule is specifically used for: The deviation coefficient is combined with the initial prediction proxy model, and the optimal linear unbiased prediction value of the deviation model is added to generate an intermediate prediction proxy model; Calculating the mean square error corresponding to the intermediate prediction proxy model; The intermediate prediction proxy model is: The mean square error is: Among them, x is the input of the proxy model, is the predicted response value of the intermediate prediction agent model, ρ _n+1 is the deviation coefficient, is the predicted response value of the initial prediction agent model, is the optimal linear unbiased prediction value of the bias model, MSE _n+1 (x) is the mean square error, is the correlation vector between the unknown full-system-level test data set and the known full-system-level test data set, R _n+1 is the N _n+1 ×N _n+1 correlation matrix R( _xi , _xj ) of the known full-system-level test data set, is the regression matrix.