A method, device, equipment and storage medium for characterizing the earth-space correlation of an aircraft

Abstract

The present application discloses a method, device, equipment and storage medium for characterizing the earth-space correlation of an aircraft, which relates to the field of aerodynamics technology, including: determining the current latent variable characterization based on the preset latent variable data of the aircraft, and constructing the current full-factor sample set using the current latent variable characterization, the preset aerodynamic performance index of the aircraft and the aircraft state data; the preset latent variable data includes the sky flight test environment data, the wind tunnel test environment data and the aircraft appearance feature data; using the preset deep learning model and the current full-factor sample set to perform model training and generalization error evaluation; resampling the preset latent variable data based on the evaluation result, and using the new current quantity characterization to jump back to the step of constructing the current full-factor sample set to construct a target second-order optimization surface; taking minimizing the generalization error as the goal, determining the target latent variable characterization and the target earth-space correlation prediction model based on the target second-order optimization surface. Improved interpretability.

Description

Technical Field

[0001] The present invention relates to the field of aerodynamics technology, and in particular to a method, device, equipment and storage medium for characterizing the earth-sky correlation of an aircraft. Background Art

[0002] The study of space-ground correlation refers to the study of how to establish the correlation between the aerodynamic characteristics of aircraft obtained from ground wind tunnel tests and the aerodynamic characteristics of aircraft under real flight conditions during aircraft design and testing. The prediction target of space-ground correlation is a function of explicit variables and implicit variables, and most of the existing related work is based on analysis and modeling of explicit variables, which has become an important challenge that restricts further improvement and optimization of this problem.

[0003] Although the current CFD (Computational Fluid Dynamics) method can perform numerical simulations on various physical flow problems, on the one hand, the reliability of its data needs to be verified by the numerical method itself, and on the other hand, there is still a lack of experimental data confirmation for complex flows under complex physical phenomena. In addition, the current ground-to-air correlation analysis method does not systematically incorporate potential influencing factors such as experimental environment and aircraft shape into the modeling and analysis vision, which to a large extent leads to poor results in the correlation model when predicting aerodynamic performance, and has great limitations. Summary of the invention

[0004] In view of this, the purpose of the present invention is to provide a method, device, equipment and storage medium for characterizing the earth-space correlation of an aircraft, which can effectively improve the interpretability and accuracy of the earth-space correlation characterization and improve the applicability. The specific scheme is as follows:

[0005] In a first aspect, the present application provides a method for characterizing the earth-space correlation of an aircraft, comprising:

[0006] Determine the corresponding current latent variable representation based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and construct the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft shape feature data;

[0007] Use the preset deep learning model and the current full-factor sample set to train the model, and perform generalization error evaluation on the sky-ground correlation prediction model obtained after training to obtain the corresponding evaluation results;

[0008] Resampling the preset latent variable data based on the evaluation result, and using a new current quantity representation to jump back to the step of constructing a current full-factor sample set using the current latent variable representation, preset aerodynamic performance indicators corresponding to the aircraft, and aircraft state data, so as to construct a target second-order optimization surface related to latent variables and generalization errors;

[0009] Taking minimizing the generalization error as the goal, the latent variable values ​​and latent variable dimensions are learned and optimized based on the target second-order optimization surface, and the corresponding target latent variable representation and target sky-ground correlation prediction model are determined.

[0010] Optionally, the determining of the corresponding current latent variable representation based on the autoencoder algorithm and preset latent variable data corresponding to the aircraft includes:

[0011] Using a quasi-randomization method to process the preset latent variable data corresponding to the aircraft to obtain corresponding processed data;

[0012] The processed data is sampled based on an autoencoder algorithm to determine a corresponding current latent variable representation.

[0013] Optionally, the constructing the current full-factor sample set using the current latent variable representation, preset aerodynamic performance indicators corresponding to the aircraft, and aircraft state data includes:

[0014] The preset aerodynamic performance index is regarded as a function of the aircraft state and the hidden variable, and the Y, X, and Z coordinate axes are constructed based on the preset aerodynamic performance index corresponding to the aircraft, the aircraft state data, and the preset hidden variable data, respectively, and the wind tunnel test data set and the sky flight test data set corresponding to the aircraft are used to determine the full-factor multidimensional space corresponding to the aircraft;

[0015] The data points in the full-factor multidimensional space are classified based on the latent variable values ​​to obtain a current full-factor sample set corresponding to the current latent variable representation.

[0016] Optionally, the using a preset deep learning model and a current full-factor sample set to perform model training includes:

[0017] Divide the current full-factor sample set based on the preset division rule to obtain the corresponding current training sample set and current test sample set;

[0018] After completing preliminary training using the current training sample set and the multi-layer perceptron in the preset deep learning model, the current training sample set is learned based on the gene expression programming algorithm in the preset deep learning model to complete the corresponding model training operation and obtain the trained heaven-earth correlation prediction model.

[0019] Optionally, the generalization error evaluation of the heaven-earth correlation prediction model obtained after training to obtain a corresponding evaluation result includes:

[0020] The heaven-earth correlation prediction model is tested based on the current test sample set, and a generalization error evaluation is performed on the obtained test results using a preset optimization function to obtain a corresponding evaluation result.

[0021] Optionally, resampling the preset latent variable data based on the evaluation result includes:

[0022] Based on the evaluation result and the autoencoder algorithm, sampling is performed in the neighborhood space of the current latent variable representation to complete the corresponding latent variable representation update operation and obtain a new latent variable representation.

[0023] Optionally, the minimization of generalization error is taken as the goal, latent variable values ​​and latent variable dimensions are learned and optimized based on the target second-order optimization surface, and the corresponding target latent variable representation and target sky-ground correlation prediction model are determined, including:

[0024] Comparing the generalization error value of each data point in the target second-order optimization surface to obtain corresponding comparison results;

[0025] Minimizing the generalization error is the goal, and based on the comparison results, a corresponding target latent variable representation and a target sky-ground correlation prediction model are determined;

[0026] By analyzing the change of the generalization error value in the target second-order optimization curve, it is determined whether there is a target dimension value that meets the preset conditions, so as to complete the corresponding latent variable dimension learning operation and obtain the corresponding dimension value judgment result; wherein, the preset condition is that the generalization error value does not change when the latent variable dimension is not less than the target dimension value.

[0027] In a second aspect, the present application provides an aircraft ground-to-space correlation characterization device, comprising:

[0028] A sample set construction module is used to determine the corresponding current latent variable representation based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and to construct the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft shape feature data;

[0029] The error evaluation module is used to use the preset deep learning model and the current full-factor sample set to perform model training, and to perform generalization error evaluation on the earth-sky correlation prediction model obtained after training to obtain corresponding evaluation results;

[0030] A surface construction module, configured to resample the preset latent variable data based on the evaluation result, and use the new latent variable representation to jump back to the step of constructing the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data, so as to construct a target second-order optimization surface related to the latent variables and the generalization error;

[0031] The characterization determination module is used to take minimizing the generalization error as the goal, learn and optimize the latent variable values ​​and latent variable dimensions based on the target second-order optimization surface, and determine the corresponding target latent variable characterization and the target sky-earth correlation prediction model.

[0032] In a third aspect, the present application provides an electronic device, including:

[0033] Memory, used to store computer programs;

[0034] The processor is used to execute the computer program to implement the steps of the aforementioned method for characterizing the space-ground correlation of an aircraft.

[0035] In a fourth aspect, the present application provides a computer-readable storage medium for storing a computer program, which, when executed by a processor, implements the steps of the aforementioned method for characterizing the space-ground correlation of an aircraft.

[0036] It can be seen that in the present application, the corresponding current latent variable representation is determined based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and the current full-factor sample set is constructed using the current latent variable representation, the preset aerodynamic performance indicators corresponding to the aircraft, and the aircraft status data; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft appearance feature data; the model is trained using the preset deep learning model and the current full-factor sample set, and the generalization error of the sky-ground correlation prediction model obtained after training is evaluated to obtain the corresponding evaluation result; based on the evaluation result, the preset latent variable data is resampled, and the new latent variable representation is used to jump back to the step of constructing the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance indicators corresponding to the aircraft, and the aircraft status data to construct a target second-order optimization surface related to the latent variables and generalization error; with minimizing the generalization error as the goal, the latent variable values ​​and the latent variable dimensions are learned and optimized based on the target second-order optimization surface, and the corresponding target latent variable representation and the target sky-ground correlation prediction model are determined. That is, in this application, the sky flight test and the wind tunnel test environment are combined, and the current full-factor sample set is constructed based on preset hidden variable data including sky flight test environment data, wind tunnel test environment data, aircraft shape feature data, and preset aerodynamic performance indicators and aircraft status data corresponding to the aircraft. Then, the sample set is learned using a preset deep learning model and the generalization error of the model is evaluated. Then, the corresponding current full-factor sample set is re-determined in combination with the evaluation results and the model is trained to construct the target second-order optimization surface. After that, the target second-order optimization surface is used for learning to obtain the final target hidden variable representation and the target earth-sky correlation prediction model, so as to predict aerodynamic performance based on the target earth-sky correlation prediction model and the target hidden variable representation, so as to solve the problem of correlation analysis of hidden variable factors in general aerodynamic scenarios. In this way, the interpretability and accuracy of the earth-sky correlation representation can be effectively improved, and the applicability can be improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] 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 embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying creative work.

[0038] Figure 1 A flow chart of a method for characterizing the earth-space correlation of an aircraft provided in this application;

[0039] Figure 2A schematic diagram of a specific method for characterizing the earth-space correlation of an aircraft provided in this application;

[0040] Figure 3 A schematic diagram of determining the current latent variable representation based on the autoencoder algorithm provided in this application;

[0041] Figure 4 A specific full-factor multi-dimensional space schematic diagram provided for this application;

[0042] Figure 5 A specific current full-factor sample set schematic diagram provided for this application;

[0043] Figure 6 A schematic diagram of constructing a target second-order optimization surface provided in this application;

[0044] Figure 7 A schematic diagram of the structure of a device for characterizing the earth-space correlation of an aircraft provided in this application;

[0045] Figure 8 A structural diagram of an electronic device provided for this application. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Although the current CFD method can perform numerical simulations on various physical flow problems, on the one hand, the reliability of its data needs to be verified by the numerical method itself, and on the other hand, there is still a lack of experimental data confirmation for complex flows under complex physical phenomena. In addition, the current ground-to-air correlation analysis method does not systematically incorporate potential influencing factors such as the experimental environment and aircraft shape into the modeling and analysis field of view, which to a large extent leads to poor results in the correlation model when predicting aerodynamic performance, and has great limitations. To this end, the present application provides an aircraft ground-to-air correlation characterization scheme that can effectively improve the interpretability and accuracy of the ground-to-air correlation characterization and improve its applicability.

[0048] See also Figure 1 As shown, an embodiment of the present invention discloses a method for characterizing the earth-sky correlation of an aircraft, comprising:

[0049] Step S11, based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, determine the corresponding current latent variable representation, and use the current latent variable representation, the preset aerodynamic performance indicators corresponding to the aircraft and the aircraft status data to construct the current full-factor sample set; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft appearance feature data.

[0050] In this embodiment, in the scenario of the earth-sky correlation problem, the factors affecting the aerodynamic performance indicators of the aircraft include not only the flight conditions (Mach number, angle of attack, rudder deflection, etc.), but also the test environment (air: real flight test; ground: different wind tunnel test conditions), aircraft shape characteristics and other potential influencing factors, all of which will affect the earth-sky correlation law. Therefore, the test environment and aircraft shape characteristics and other potential influencing factors are used as hidden variables to study the earth-sky correlation.

[0051] Specific, combined Figure 2 It can be seen that the sample set needs to be constructed first. In this process, the generation of latent variable representation, that is, the determination of the corresponding current latent variable representation based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, includes: using a quasi-randomization method to process the preset latent variable data corresponding to the aircraft to obtain the corresponding processed data; sampling the processed data based on the autoencoder algorithm to determine the corresponding current latent variable representation. That is, considering the simplicity of rapid prototyping and technical implementation, before using the autoencoder algorithm, the data is first processed by a quasi-randomization method, that is, considering the random selection of the values ​​of the latent variables within a certain range, and the final latent variable discovery is completed by the generalization error minimization-oriented method. After that, the value optimization technology of the latent variables based on the autoencoder method (Auto Encoder, i.e. AE) is used to improve the effect and obtain the current latent variable representation.

[0052] It should be understood that since the aerodynamic test samples of the aircraft are distributed in the sky and the ground, a mapping that satisfies the following properties can be obtained through learning: / -1:

[0053] ;

[0054] ;

[0055] ;

[0056] The above formula is the mathematical abstraction of the encoding / decoding process of the autoencoder algorithm, and this method is used to solve the accurate representation of latent variables. is the encoder; mapping -1 is the decoder; is the flight state / operating condition of the aircraft; Y is the preset aerodynamic performance index; is a hidden variable. Therefore, the designed self-learning network is constructed as follows Figure 3 As shown, the latent variable representation is completed The current latent variable representation is obtained by initializing . In addition, in a specific implementation, the latent variable representation can also be determined only by a quasi-randomization algorithm.

[0057] Further, regarding the construction of the sample set, in this embodiment, the current full-factor sample set is constructed using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data, including: considering the preset aerodynamic performance index as a function of the aircraft state and the latent variable, constructing the Y, X, and Z coordinate axes based on the preset aerodynamic performance index, aircraft state data, and the preset latent variable data corresponding to the aircraft, respectively, and using the wind tunnel test data set and the sky flight test data set corresponding to the aircraft to determine the full-factor multidimensional space corresponding to the aircraft; classifying the data points in the full-factor multidimensional space based on the latent variable values ​​to obtain the current full-factor sample set corresponding to the current latent variable representation. Among them, the preset aerodynamic performance index may include but is not limited to aerodynamic force, aerodynamic pressure, wind resistance, and lift. That is, this embodiment first establishes a mathematical model of the earth-sky correlation scene containing the full-factor space. Specifically, the preset aerodynamic performance indicators as prediction targets in the scene are regarded as functions of aircraft working conditions and hidden variables, and then the ground-earth correlation problem is transformed into a function discovery and surface fitting problem in high-dimensional space with aerodynamic performance as the solution target. The full-factor multidimensional space obtained by modeling is as follows: Figure 4 shown.

[0058] Said Figure 4 In the figure, the coordinate axis Y represents the preset aerodynamic performance index, and may also represent the vector dimension composed of the preset aerodynamic performance index; the coordinate axis X represents the flight state / operating condition of the aircraft, which may include lower-level indicators such as Mach number, angle of attack, and rudder deflection, and is a vector dimension; Z represents the vector dimension composed of implicit variable factors such as environmental factors and shape; the shaded part in the figure represents the test data set obtained from a relatively large number of ground wind tunnel tests, and the data set corresponding to the actual flight test is relatively sparse in space, as described Figure 4 Middle Z R The corresponding curve. Then the problem of finding the law of correlation between heaven and earth can be transformed into a function The problem of solving Z0, Z1, …, Z m represents the latent variable factor obtained based on the sampled ground wind tunnel test data; Z RRepresents the hidden variable factors obtained based on sampling of real flight test data. Since X represents the flight conditions, a relatively dense shadow surface (such as multiple sets of ground wind tunnel test data sets) and a relatively sparse real flight curve can be constructed based on the existing data in space, and they can be combined to contribute to the learning of the hypersurface of the earth-sky correlation in the physical world, that is, Y is related to , The hypersurface of the flight state / operating condition parameter vector and the aerodynamic characteristic parameter prediction of the specific flight condition are completed.

[0059] It is further to be understood that, in this embodiment, based on the current latent variable representation, a full-factor sample space is constructed to further learn the law of heaven-earth correlation. The construction method is as follows: Figure 5 As shown. Figure 4 The data points on the relatively dense shadow surface and the relatively sparse real flight curve in the multidimensional space shown are classified and marked as sub-sample sets according to their latent variable values. , , , …, and Each sub-sample set contains a different number of data samples, and each data sample is determined by the latent variable , working conditions , aerodynamic performance Y. Among them, for each sub-dataset D i (i takes the values ​​E0, E2, ..., E m With E R ) samples, latent variables The representation is the same. After summarizing, the full sample data set D of the full factor space is obtained W , assuming the data set of the real flight test Contains a samples, then Figure 5 Yes, the total number of samples in the sample set , where N1 is the number of samples corresponding to the ground wind tunnel test; N2 is the number of samples corresponding to the actual flight test; n i for , , , …, The number of samples in each sub-dataset.

[0060] Step S12: Use the preset deep learning model and the current full-factor sample set to perform model training, and perform generalization error evaluation on the heaven-earth correlation prediction model obtained after training to obtain corresponding evaluation results.

[0061] In this embodiment, Figure 6As shown, the rules of heaven-earth correlation based on deep neural network are then learned, that is, the preset deep learning model and the current full-factor sample set are used for model training, including: dividing the current full-factor sample set based on preset division rules to obtain the corresponding current training sample set and current test sample set; after completing preliminary training using the current training sample set and the multi-layer perceptron in the preset deep learning model, the current training sample set is learned based on the gene expression programming algorithm in the preset deep learning model to complete the corresponding model training operation and obtain the trained heaven-earth correlation prediction model. Figure 6 M and b in it represent the model parameters of the preset deep learning model. In addition, the generalization error evaluation is performed on the heaven-earth correlation prediction model obtained after training to obtain the corresponding evaluation results, including: testing the heaven-earth correlation prediction model based on the current test sample set, and using the preset optimization function to perform generalization error evaluation on the obtained test results to obtain the corresponding evaluation results. That is, a multi-layer perceptron can be used for preliminary model training, and later consider using the gene expression programming algorithm for function discovery to further explore the interpretability of the heaven-earth correlation law, divide each sub-sample set in the sample set separately, to train the model and detect the model training error separately, and then estimate the generalization error and record it in the robustness evaluation space.

[0062] Step S13, resampling the preset latent variable data based on the evaluation result, and using the new current quantity representation to jump back to the step of constructing the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance indicators corresponding to the aircraft, and the aircraft state data, so as to construct a target second-order optimization surface related to the latent variables and the generalization error.

[0063] Specifically, in this embodiment, due to the hidden variable The accurate characterization of the hidden variables is the cornerstone for ensuring the accuracy of the laws of heaven-earth correlation. That is, the current latent variable representation needs to be continuously optimized, and the preset latent variable data is resampled based on the evaluation result, including: sampling in the neighborhood space of the current latent variable representation based on the evaluation result and the autoencoder algorithm to complete the corresponding latent variable representation update operation and obtain a new latent variable representation. Combined with Figure 2 As shown in the figure, with the goal of minimizing the generalization error, heuristic algorithms or reinforcement learning methods can be used to optimize the latent variable representation space, that is, sampling, constructing sample sets, model training, and generalization error evaluation are performed in the neighborhood space of the current latent variable representation using a certain strategy, and the changes in the generalization error are examined and repeated to construct a target second-order optimization surface related to the latent variable and generalization error. Each data point in the surface has a corresponding generalization error value, latent variable representation, and latent variable dimension value.

[0064] Step S14, taking minimization of generalization error as the goal, learning and optimizing latent variable values ​​and latent variable dimensions based on the target second-order optimization surface, and determining the corresponding target latent variable representation and target sky-earth correlation prediction model.

[0065] Specifically, in this embodiment, minimizing the generalization error is taken as the goal, and based on the target second-order optimization surface, the latent variable values ​​and latent variable dimensions are learned and optimized (the optimization techniques include reinforcement learning, gradient descent and other methods), and the corresponding target latent variable representation and the target heaven-earth correlation prediction model are determined, including: comparing the generalization error value of each data point in the target second-order optimization surface to obtain the corresponding comparison result; minimizing the generalization error is taken as the goal and the corresponding target latent variable representation and the target heaven-earth correlation prediction model are determined based on the comparison result; by analyzing the change of the generalization error value in the target second-order optimization curve, it is determined whether there is a target dimension value that meets the preset conditions, so as to complete the corresponding latent variable dimension learning operation and obtain the corresponding dimension value judgment result; wherein the preset condition is that the generalization error value does not change when the latent variable dimension is not less than the target dimension value. That is, combined with the generalization error value of each data point on the target second-order optimization surface, the change and distribution of the model generalization error are analyzed, and then the optimized latent variable is obtained. The representation of the target latent variable is determined, and the target sky-land correlation prediction model with the smallest generalization error is determined.

[0066] It can be seen that in the embodiment of the present application, the corresponding current latent variable representation is determined based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and the current full-factor sample set is constructed using the current latent variable representation, the preset aerodynamic performance indicators corresponding to the aircraft, and the aircraft status data; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft appearance feature data; the model is trained using the preset deep learning model and the current full-factor sample set, and the generalization error of the sky-ground correlation prediction model obtained after training is evaluated to obtain the corresponding evaluation result; based on the evaluation result, the preset latent variable data is resampled, and the new latent variable representation is used to jump back to the step of constructing the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance indicators corresponding to the aircraft, and the aircraft status data to construct a target second-order optimization surface related to the latent variables and generalization error; with minimizing the generalization error as the goal, the latent variable values ​​and the latent variable dimensions are learned and optimized based on the target second-order optimization surface, and the corresponding target latent variable representation and the target sky-ground correlation prediction model are determined. That is, in this application, the sky flight test and the wind tunnel test environment are combined, and the current full-factor sample set is constructed based on preset hidden variable data including sky flight test environment data, wind tunnel test environment data, aircraft shape feature data, and preset aerodynamic performance indicators and aircraft status data corresponding to the aircraft. Then, the sample set is learned using a preset deep learning model and the generalization error of the model is evaluated. Then, the corresponding current full-factor sample set is re-determined in combination with the evaluation results and the model is trained to construct the target second-order optimization surface. After that, the target second-order optimization surface is used for learning to obtain the final target hidden variable representation and the target earth-sky correlation prediction model, so as to predict aerodynamic performance based on the target earth-sky correlation prediction model and the target hidden variable representation, so as to solve the problem of correlation analysis of hidden variable factors in general aerodynamic scenarios. In this way, the interpretability and accuracy of the earth-sky correlation representation can be effectively improved, and the applicability can be improved.

[0067] Combine the following Figure 2-6 The schematic diagram disclosed in the figure specifically illustrates the technical solution of the embodiment of the present application.

[0068] In this embodiment, the vector composed of implicit variables such as the test environment and the aircraft shape is regarded as a hidden variable, and the aerodynamic performance index is regarded as a function of the aircraft working environment hidden variable, thereby transforming the earth-sky correlation problem into a function discovery and surface fitting problem in a high-dimensional space with aerodynamic performance as the solution target. In this process, the generalization error is used as a criterion for judging whether it is a real hidden variable. If it is a real hidden variable, the generalization error must be the smallest.

[0069] Specifically, we first conduct scene abstraction and mathematical modeling of the earth-sky correlation in all-element space to obtain Figure 4 The full-factor multidimensional space is shown. Then the preset latent variable data is sampled to determine Figure 5 The full-factor sample set is shown in Figure 1. Then, the preset deep learning model and the full-factor sample set are used for model training and generalization error evaluation. Figure 6 The generalization error related expression shown in is used to determine the error value, and the sample set is updated based on the error value. The model is trained again, and the generalization error is evaluated. This cycle is repeated, and the error value is used to construct a target second-order optimization surface related to the latent variables and generalization error. After that, the target second-order optimization surface is analyzed with minimizing the generalization error as the goal, and the corresponding target latent variable representation and the target earth-sky correlation prediction model are determined, and it is determined whether there is an error equilibrium point. In this way, the aerodynamic performance can be predicted based on the target earth-sky correlation prediction model and the target latent variable representation to solve the problem of correlation analysis of latent variable factors in general aerodynamic scenarios.

[0070] See also Figure 7 As shown, the embodiment of the present application also discloses a device for characterizing the earth-sky correlation of an aircraft, including:

[0071] The sample set construction module 11 is used to determine the corresponding current latent variable representation based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and construct the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data; wherein the preset latent variable data includes air flight test environment data, wind tunnel test environment data, and aircraft shape feature data;

[0072] The error evaluation module 12 is used to perform model training using a preset deep learning model and a current full-factor sample set, and to perform a generalization error evaluation on the earth-sky correlation prediction model obtained after the training to obtain a corresponding evaluation result;

[0073] A surface construction module 13 is used to resample the preset latent variable data based on the evaluation result, and use the new latent variable representation to jump back to the step of constructing the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data, so as to construct a target second-order optimization surface related to the latent variable and the generalization error;

[0074] The characterization determination module 14 is used to take minimization of generalization error as the goal, learn and optimize the latent variable values ​​and latent variable dimensions based on the target second-order optimization surface, and determine the corresponding target latent variable characterization and the target sky-earth correlation prediction model.

[0075] Among them, for more specific working processes of the above-mentioned modules, please refer to the corresponding contents disclosed in the aforementioned embodiments, which will not be repeated here.

[0076] It can be seen that in this application, the sky flight test and the wind tunnel test environment are combined, and the preset hidden variable data including the sky flight test environment data, the wind tunnel test environment data, the aircraft shape feature data, and the preset aerodynamic performance indicators and aircraft status data corresponding to the aircraft are used to construct the current full-factor sample set. Then, the preset deep learning model is used to learn the sample set and evaluate the generalization error of the model. Then, the corresponding current full-factor sample set is re-determined in combination with the evaluation results and the model is trained to construct the target second-order optimization surface. After that, the target second-order optimization surface is used for learning to obtain the final target hidden variable representation and the target earth-sky correlation prediction model, so as to predict the aerodynamic performance based on the target earth-sky correlation prediction model and the target hidden variable representation, so as to solve the problem of correlation analysis of hidden variable factors in general aerodynamic scenarios. In this way, the interpretability and accuracy of the earth-sky correlation representation can be effectively improved, and the applicability can be improved.

[0077] In some specific embodiments, the sample set construction module 11 can be specifically used to: process the preset latent variable data corresponding to the aircraft using a quasi-randomization method to obtain corresponding processed data; sample the processed data based on the autoencoder algorithm to determine the corresponding current latent variable representation.

[0078] In some specific embodiments, the sample set construction module 11 can be specifically used to: regard the preset aerodynamic performance indicators as functions of the aircraft state and latent variables, and construct the Y, X, and Z coordinate axes based on the preset aerodynamic performance indicators corresponding to the aircraft, the aircraft state data, and the preset latent variable data, respectively, and use the wind tunnel test data set and the sky flight test data set corresponding to the aircraft to determine the full-factor multidimensional space corresponding to the aircraft; classify the data points in the full-factor multidimensional space based on the latent variable values ​​to obtain the current full-factor sample set corresponding to the current latent variable representation.

[0079] In some specific embodiments, the error assessment module 12 can be specifically used to: divide the current full-factor sample set based on preset division rules to obtain the corresponding current training sample set and the current test sample set; after completing the preliminary training using the current training sample set and the multi-layer perceptron in the preset deep learning model, learn the current training sample set based on the gene expression programming algorithm in the preset deep learning model to complete the corresponding model training operation and obtain the trained heaven-earth correlation prediction model.

[0080] In some specific embodiments, the error assessment module 12 can be specifically used to: test the heaven-earth correlation prediction model based on the current test sample set, and use a preset optimization function to perform a generalization error assessment on the obtained test results to obtain corresponding evaluation results.

[0081] In some specific embodiments, the surface construction module 13 can be specifically used to: sample in the neighborhood space of the current latent variable representation based on the evaluation result and the autoencoder algorithm to complete the corresponding latent variable representation update operation and obtain a new latent variable representation.

[0082] In some specific embodiments, the representation determination module 14 can be specifically used to: compare the generalization error value of each data point in the target second-order optimization surface to obtain a corresponding comparison result; determine the corresponding target latent variable representation and the target sky-ground correlation prediction model based on the comparison result with the goal of minimizing the generalization error; determine whether there is a target dimension value that meets the preset conditions by analyzing the change of the generalization error value in the target second-order optimization curve, so as to complete the corresponding latent variable dimension learning operation and obtain the corresponding dimension value judgment result; wherein, the preset condition is that the generalization error value does not change when the latent variable dimension is not less than the target dimension value.

[0083] Furthermore, the present application also discloses an electronic device. Figure 8 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content in the diagram cannot be considered as any limitation on the scope of use of the present application.

[0084] Figure 8 A schematic diagram of the structure of an electronic device 20 provided in an embodiment of the present application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 is used to store a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the method for characterizing the space-aircraft correlation of an aircraft disclosed in any of the aforementioned embodiments. In addition, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0085] In this embodiment, the power supply 23 is used to provide working voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and the external device, and the communication protocol it follows is any communication protocol that can be applied to the technical solution of the present application, and is not specifically limited here; the input and output interface 25 is used to obtain external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs and is not specifically limited here.

[0086] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a disk or an optical disk, etc. The resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0087] The operating system 221 is used to manage and control the hardware devices and computer programs 222 on the electronic device 20, and may be Windows Server, Netware, Unix, Linux, etc. In addition to the computer program that can be used to complete the method for characterizing the space-aircraft correlation performed by the electronic device 20 disclosed in any of the aforementioned embodiments, the computer program 222 may further include computer programs that can be used to complete other specific tasks.

[0088] Furthermore, the present application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, the above-disclosed method for characterizing the space-aircraft correlation is implemented. The specific steps of the method can be referred to the corresponding contents disclosed in the above-disclosed embodiments, and will not be repeated here.

[0089] In this specification, each embodiment is described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the embodiments can be referred to each other. For the device disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the method part.

[0090] Professionals may further appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in the above description according to function. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professionals and technicians may use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of this application.

[0091] The steps of the method or algorithm described in conjunction with the embodiments disclosed herein may be implemented directly using hardware, a software module executed by a processor, or a combination of the two. The software module may be placed in a random access memory (RAM), a memory, a read-only memory (ROM), an electrically programmable ROM, an electrically erasable programmable ROM, a register, a hard disk, a removable disk, a CD-ROM, or any other form of storage medium known in the art.

[0092] Finally, it should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0093] The technical solution provided by the present application is introduced in detail above. Specific examples are used in this article to illustrate the principles and implementation methods of the present application. The description of the above embodiments is only used to help understand the method of the present application and its core idea. At the same time, for general technicians in this field, according to the idea of ​​the present application, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as a limitation on the present application.

Claims

1. A method for characterizing the earth-space correlation of an aircraft, characterized in that: include: Determine the corresponding current latent variable representation based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and construct the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft shape feature data; Use the preset deep learning model and the current full-factor sample set to train the model, and perform generalization error evaluation on the sky-ground correlation prediction model obtained after training to obtain the corresponding evaluation results; Resampling the preset latent variable data based on the evaluation result, and using a new current quantity representation to jump back to the step of constructing a current full-factor sample set using the current latent variable representation, preset aerodynamic performance indicators corresponding to the aircraft, and aircraft state data, so as to construct a target second-order optimization surface related to latent variables and generalization errors; Taking minimizing the generalization error as the goal, the latent variable values ​​and latent variable dimensions are learned and optimized based on the target second-order optimization surface, and the corresponding target latent variable representation and target sky-ground correlation prediction model are determined.

2. The method for characterizing the space-ground correlation of an aircraft according to claim 1, characterized in that: The determining of the corresponding current latent variable representation based on the autoencoder algorithm and preset latent variable data corresponding to the aircraft includes: Using a quasi-randomization method to process the preset latent variable data corresponding to the aircraft to obtain corresponding processed data; The processed data is sampled based on an autoencoder algorithm to determine a corresponding current latent variable representation.

3. The method for characterizing the space-ground correlation of an aircraft according to claim 1, characterized in that: The method of constructing a current full-factor sample set using the current latent variable representation, preset aerodynamic performance indicators corresponding to the aircraft, and aircraft status data includes: The preset aerodynamic performance index is regarded as a function of the aircraft state and the hidden variable, and the Y, X, and Z coordinate axes are constructed based on the preset aerodynamic performance index corresponding to the aircraft, the aircraft state data, and the preset hidden variable data, respectively, and the wind tunnel test data set and the sky flight test data set corresponding to the aircraft are used to determine the full-factor multidimensional space corresponding to the aircraft; The data points in the full-factor multidimensional space are classified based on the latent variable values ​​to obtain a current full-factor sample set corresponding to the current latent variable representation.

4. The method for characterizing the space-ground correlation of an aircraft according to claim 1, characterized in that: The method of using a preset deep learning model and a current full-factor sample set to perform model training includes: Divide the current full-factor sample set based on the preset division rule to obtain the corresponding current training sample set and current test sample set; After completing preliminary training using the current training sample set and the multi-layer perceptron in the preset deep learning model, the current training sample set is learned based on the gene expression programming algorithm in the preset deep learning model to complete the corresponding model training operation and obtain the trained heaven-earth correlation prediction model.

5. The method for characterizing the space-ground correlation of an aircraft according to claim 4, characterized in that: The generalization error evaluation of the land-earth correlation prediction model obtained after training is performed to obtain corresponding evaluation results, including: The heaven-earth correlation prediction model is tested based on the current test sample set, and a generalization error evaluation is performed on the obtained test results using a preset optimization function to obtain a corresponding evaluation result.

6. The method for characterizing the space-ground correlation of an aircraft according to claim 1, characterized in that: The resampling of the preset latent variable data based on the evaluation result includes: Based on the evaluation result and the autoencoder algorithm, sampling is performed in the neighborhood space of the current latent variable representation to complete the corresponding latent variable representation update operation and obtain a new latent variable representation.

7. The method for characterizing the space-ground correlation of an aircraft according to any one of claims 1 to 6, characterized in that: The minimization of generalization error is taken as the goal, and the latent variable values ​​and latent variable dimensions are learned and optimized based on the target second-order optimization surface, and the corresponding target latent variable representation and target sky-ground correlation prediction model are determined, including: Comparing the generalization error value of each data point in the target second-order optimization surface to obtain corresponding comparison results; Minimizing the generalization error is the goal, and based on the comparison results, a corresponding target latent variable representation and a target sky-ground correlation prediction model are determined; By analyzing the change of the generalization error value in the target second-order optimization curve, it is determined whether there is a target dimension value that meets the preset conditions, so as to complete the corresponding latent variable dimension learning operation and obtain the corresponding dimension value judgment result; wherein, the preset condition is that the generalization error value does not change when the latent variable dimension is not less than the target dimension value.

8. An aircraft earth-sky correlation characterization device, characterized in that: include: A sample set construction module is used to determine the corresponding current latent variable representation based on the autoencoder algorithm and the preset latent variable data corresponding to the aircraft, and to construct the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data; wherein the preset latent variable data includes sky flight test environment data, wind tunnel test environment data, and aircraft shape feature data; The error evaluation module is used to use the preset deep learning model and the current full-factor sample set to perform model training, and to perform generalization error evaluation on the earth-sky correlation prediction model obtained after training to obtain corresponding evaluation results; A surface construction module, configured to resample the preset latent variable data based on the evaluation result, and use the new latent variable representation to jump back to the step of constructing the current full-factor sample set using the current latent variable representation, the preset aerodynamic performance index corresponding to the aircraft, and the aircraft state data, so as to construct a target second-order optimization surface related to the latent variables and the generalization error; The characterization determination module is used to take minimizing the generalization error as the goal, learn and optimize the latent variable values ​​and latent variable dimensions based on the target second-order optimization surface, and determine the corresponding target latent variable characterization and the target sky-earth correlation prediction model.

9. An electronic device, characterized in that: include: Memory, used to store computer programs; A processor, configured to execute the computer program to implement the method for characterizing the space-ground correlation of an aircraft as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: Used to store a computer program, which, when executed by a processor, implements the method for characterizing the space-ground correlation of an aircraft as described in any one of claims 1 to 7.

Images