Aircraft aerodynamic data generation method, device, equipment and medium

Abstract

The invention discloses an aircraft aerodynamic data generation method and device, equipment and a medium, and relates to the technical field of aerodynamics, and the method comprises the steps: obtaining a preset aerodynamic data generation model based on a preset low-credibility training sample and a preset high-credibility training sample through Gaussian process modeling and entropy adaptive sampling method fusion training; obtaining aerodynamic design parameters of the target aircraft; inputting the aerodynamic design parameters into a preset aerodynamic data generation model to calculate a kernel function vector of the aerodynamic design parameters, and then carrying out fusion operation on the kernel function vector, scaling factors between different credibility levels and a covariance matrix representing the correlation between aerodynamic data to obtain a fusion result; generating a multi-output pneumatic data prediction mean vector corresponding to the pneumatic design parameters; and outputting pneumatic data corresponding to the pneumatic data prediction mean vector to obtain multi-output pneumatic data of the target aircraft. And high-confidence and multi-output aircraft aerodynamic data can be generated at low cost.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of aerodynamics, in particular to an aircraft aerodynamic data generation method, device, equipment and medium. BACKGROUND

[0002] The sources of aircraft aerodynamic database include semi-empirical data table method, linear flow solver, nonlinear flow solver, small wind tunnel test, full-scale wind tunnel test and flight test (arranged in order of increasing fidelity). With the increase of fidelity, the calculation time and cost will also increase. How to quickly, efficiently and economically obtain aerodynamic data to provide analysis and decision support for researchers is one of the current research difficulties and hotspots. In recent years, the rapid development of proxy model technology has brought new research ideas for the establishment of aircraft aerodynamic database. Usually, a small amount or limited amount of aerodynamic data is obtained based on CFD (Computational Fluid Dynamics), wind tunnel test and flight test, and a suitable proxy model technology is used to quickly generate high-reliability aerodynamic data required by the aerodynamic database, which is an effective means to support aircraft design, reduce cost and shorten development cycle.

[0003] Single-output Gaussian process can only simulate output individually in the case of multi-output, but in complex physical processes, there is often correlation between multi-dimensional outputs. The multi-output Gaussian process model is used to process the problem of multi-dimensional correlated output. It models the output at the same time and can transfer useful information across outputs. However, directly using a multi-output model to fit requires a large number of high-confidence sample points to ensure the accuracy of the proxy model, and the acquisition of high-confidence sample points requires more computing resources and cost. Therefore, the aircraft aerodynamic data generated on the basis of considering low computing resources and cost is incomplete, resulting in the loss of some useful information.

[0004] In summary, how to use the correlation between multi-output to generate high-confidence, multi-output aircraft aerodynamic data on the basis of considering low computing resources and cost is a technical problem to be solved in the field. SUMMARY

[0005] Therefore, the purpose of the present application is to provide an aircraft aerodynamic data generation method, device, equipment and medium, which can use the correlation between multi-output to generate high-confidence, multi-output aircraft aerodynamic data on the basis of considering low computing resources and cost. The specific scheme is as follows: In a first aspect, the present application discloses an aircraft aerodynamic data generation method, comprising: obtaining aerodynamic design parameter training data and corresponding aerodynamic output data at different confidence levels to construct a preset low-confidence sample pool and a preset high-confidence sample pool, respectively; constructing a preset low-confidence Gaussian process model with multiple outputs by using preset low-confidence training samples randomly selected from the preset low-confidence sample pool; selecting new training samples from the preset low-confidence sample pool by an entropy-based adaptive sampling method to add to the preset low-confidence training samples, reiteratively training the preset low-confidence Gaussian process model, and obtaining an optimized preset low-confidence Gaussian process model; constructing a preset multi-confidence Gaussian process model with multiple outputs based on the optimized preset low-confidence Gaussian process model and preset high-confidence training samples, wherein the preset high-confidence training samples are training samples randomly selected from the preset high-confidence sample pool; selecting new training samples from the preset high-confidence sample pool by an entropy-based adaptive sampling method to add to the preset high-confidence training samples, reiteratively training the preset multi-confidence Gaussian process model, and obtaining a preset aerodynamic data generation model; obtaining aerodynamic design parameters of a target aircraft; inputting the aerodynamic design parameters into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameters, and then performs fusion operation on the kernel function vector, scaling factors between different confidence levels, and a covariance matrix representing correlations between aerodynamic data, to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameters; outputting aerodynamic data corresponding to the aerodynamic data prediction mean vector to obtain multi-output aerodynamic data of the target aircraft.

[0006] Optionally, the constructing a preset low-confidence Gaussian process model with multiple outputs by using preset low-confidence training samples randomly selected from the preset low-confidence sample pool comprises: calculating similarities between aerodynamic design parameter training data in the preset low-confidence training samples by using a preset kernel function, to construct a first kernel matrix; constructing a first covariance matrix representing correlations between aerodynamic data; combining the first kernel matrix and the first covariance matrix by Kronecker product, to construct a first total covariance structure of the preset low-confidence Gaussian process model; determining parameters of the first kernel matrix and parameters of the first covariance matrix based on aerodynamic output data in the preset low-confidence training samples and by maximum likelihood estimation, to determine initial structure parameters of the first total covariance structure, and constructing a preset low-confidence Gaussian process model with multiple outputs.

[0007] Optionally, the entropy-based adaptive sampling method selects to adaptively select a new training sample from the preset low-confidence sample pool to join the preset low-confidence training sample, and iteratively re-train the preset low-confidence Gaussian process model to obtain an optimized preset low-confidence Gaussian process model, comprising: calculating a first covariance matrix of the joint probability distribution at each first candidate sample point in the preset low-confidence sample pool using the preset low-confidence Gaussian process model of the current iteration; calculating a first predicted entropy corresponding to each first candidate sample point based on the first covariance matrix; selecting the candidate sample point corresponding to the maximum predicted entropy in all first predicted entropies as a new training sample to join the preset low-confidence training sample; using the preset low-confidence training sample, jumping to execute the step of calculating the similarity between the aerodynamic design parameter training data in the preset low-confidence training sample using the preset kernel function to construct the first kernel matrix, until the initial structure parameters of the preset low-confidence Gaussian process model reach the preset low-confidence iteration termination condition, to obtain an optimized preset low-confidence Gaussian process model containing target structure parameters.

[0008] Optionally, the preset multi-confidence Gaussian process model is constructed based on the optimized preset low-confidence Gaussian process model and the preset high-confidence training sample, comprising: constructing a hybrid training sample set according to the preset high-confidence training sample and the preset low-confidence training sample; calculating the similarity between the aerodynamic design parameter training data of the training samples in the hybrid training sample set using the same preset kernel function as the optimized preset low-confidence Gaussian process model to construct a second kernel matrix; using the same covariance matrix as the optimized preset low-confidence Gaussian process model to represent the correlation between the aerodynamic data as a second covariance matrix; fusing the second kernel matrix, the second covariance matrix, and a scaling factor vector representing the association relationship between different confidence levels to construct a second total covariance structure of the preset multi-confidence Gaussian process model; determining the vector value of the scaling factor vector, the parameters of the second kernel matrix, and the parameters of the second covariance matrix according to the aerodynamic output data in the preset high-confidence training sample and through maximum likelihood estimation, to determine the initial structure parameters of the second total covariance structure, and construct a multi-output preset multi-confidence Gaussian process model.

[0009] Optionally, the entropy-based adaptive sampling method adaptively selects new training samples from the preset high-confidence sample pool to join the preset high-confidence training samples, and iteratively trains the preset multi-confidence Gaussian process model again to obtain a preset aerodynamic data generation model, comprising: calculating a second covariance matrix of the joint probability distribution at each second candidate sample point in the preset high-confidence sample pool by using the preset multi-confidence Gaussian process model of the current iteration; calculating a second predicted entropy corresponding to each second candidate sample point based on the second covariance matrix; selecting the candidate sample point corresponding to the maximum predicted entropy in all second predicted entropies as a new training sample to join the preset high-confidence training samples; using the preset high-confidence training samples, jumping to execute the step of constructing a hybrid training sample set according to the preset high-confidence training samples and the preset low-confidence training samples until the initial structure parameters of the preset multi-confidence Gaussian process model reach a preset multi-confidence iteration termination condition, and obtaining a preset aerodynamic data generation model containing target structure parameters.

[0010] In a second aspect, the application discloses an aircraft aerodynamic data generation device, comprising: a sample acquisition module, configured to acquire aerodynamic design parameter training data and corresponding aerodynamic output data under different confidence levels to construct a preset low-confidence sample pool and a preset high-confidence sample pool, respectively; a first model construction module, configured to construct a multi-output preset low-confidence Gaussian process model by using preset low-confidence training samples randomly selected from the preset low-confidence sample pool; a first model optimization module, configured to select new training samples from the preset low-confidence sample pool to join the preset low-confidence training samples by using an entropy-based adaptive sampling method, and iteratively train the preset low-confidence Gaussian process model again to obtain an optimized preset low-confidence Gaussian process model; a second model construction module, configured to construct a multi-output preset multi-confidence Gaussian process model based on the optimized preset low-confidence Gaussian process model and preset high-confidence training samples; wherein the preset high-confidence training samples are training samples randomly selected from the preset high-confidence sample pool; a second model optimization module, configured to select new training samples from the preset high-confidence sample pool to join the preset high-confidence training samples by using an entropy-based adaptive sampling method, and iteratively train the preset multi-confidence Gaussian process model again to obtain a preset aerodynamic data generation model; a parameter acquisition module, configured to acquire aerodynamic design parameters of a target aircraft; a multi-output prediction module, configured to input the aerodynamic design parameter into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameter, and then performs fusion operation on the kernel function vector, scaling factors between different credibility levels, and a covariance matrix representing correlation between aerodynamic data, to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameter; a multi-output data generation module, configured to output aerodynamic data corresponding to the aerodynamic data prediction mean vector, to obtain multi-output aerodynamic data of the target aircraft.

[0011] In a third aspect, the present application discloses an electronic device, comprising: a memory, configured to save a computer program; a processor, configured to execute the computer program to implement the steps of the aircraft aerodynamic data generation method disclosed above.

[0012] In a fourth aspect, the present application discloses a computer readable storage medium, configured to store a computer program; wherein the computer program is executed by a processor to implement the steps of the aircraft aerodynamic data generation method disclosed above.

[0013] It can be seen that the application discloses obtaining aerodynamic design parameter training data and corresponding aerodynamic output data under different credibility levels to construct a preset low-credibility sample pool and a preset high-credibility sample pool; a preset low-credibility Gaussian process model with multiple outputs is constructed by using preset low-credibility training samples randomly selected from the preset low-credibility sample pool; new training samples are adaptively selected from the preset low-credibility sample pool by using an entropy-based adaptive sampling method to add the preset low-credibility training samples, and the preset low-credibility Gaussian process model is re-iteratively trained to obtain an optimized preset low-credibility Gaussian process model; a preset multi-credibility Gaussian process model with multiple outputs is constructed based on the optimized preset low-credibility Gaussian process model and preset high-credibility training samples; wherein the preset high-credibility training samples are training samples randomly selected from the preset high-credibility sample pool; new training samples are adaptively selected from the preset high-credibility sample pool by using an entropy-based adaptive sampling method to add the preset high-credibility training samples, and the preset multi-credibility Gaussian process model is re-iteratively trained to obtain a preset aerodynamic data generation model; aerodynamic design parameters of a target aircraft are obtained; the aerodynamic design parameters are input into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameters, and then the kernel function vector, a scaling factor between different credibility levels, and a covariance matrix representing the correlation between aerodynamic data are fused to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameters; the aerodynamic data corresponding to the aerodynamic data prediction mean vector is output to obtain multi-output aerodynamic data of the target aircraft. It can be seen that, when the preset aerodynamic data generation model is used for prediction and output, other output information can be used for fusion operation by using the covariance matrix and the scaling factor, so that useful information is transmitted across outputs, it is ensured that a set of physically coordinated aerodynamic data is finally generated, physical contradictions and information loss caused by separate prediction of each output are avoided, the problem of large demand for high-credibility sample quantity is solved by multi-credibility fusion, meanwhile, the information is fully utilized under the condition of limited samples by multi-output modeling, that is, the shortage of high-credibility sample quantity is made up by introducing a large amount of low-credibility data and using the correlation between outputs. The model can ensure that the multi-output aerodynamic data output reaches high prediction precision and reliability even if only a small amount of high-credibility samples are used. BRIEF DESCRIPTION OF DRAWINGS

[0014] In order to make the technical solutions in the embodiments of the present application or the prior art clearer, the accompanying drawings needed in the embodiments or prior art description will be briefly introduced below. Obviously, the accompanying drawings in the following description are only some embodiments of the present application, and other drawings can be obtained by those skilled in the art without creative effort on the basis of the provided drawings.

[0015] Figure 1 A flow chart of a method for generating aircraft aerodynamic data disclosed in the present application is shown in FIG. 1. Figure 2 A flow chart of a method for constructing and training a preset aerodynamic data generation model disclosed in the present application is shown in FIG. 2. Figure 3 A structural schematic diagram of an aircraft aerodynamic data generation device disclosed in the present application is shown in FIG. 3. Figure 4 A structural diagram of an electronic device disclosed in the present application is shown in FIG. 4. DETAILED DESCRIPTION

[0016] The technical solutions in the embodiments of the present application will be described clearly and completely below with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only some of the embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without creative effort belong to the scope of protection of the present application.

[0017] The sources of the aircraft aerodynamic database include semi-empirical data table method, linear flow solver, nonlinear flow solver, small wind tunnel test, full-scale wind tunnel test and flight test (arranged in order of increasing fidelity). As the fidelity increases, the calculation time and cost also increase. How to quickly, efficiently and economically obtain aerodynamic data to provide analysis and decision support for researchers is one of the current research difficulties and hotspots. In recent years, the rapid development of proxy model technology has brought new research ideas for the establishment of the aircraft aerodynamic database. Usually, a small amount or limited amount of aerodynamic data is obtained based on CFD / wind tunnel test / flight test and other means, and a suitable proxy model technology is used to quickly generate high-reliability aerodynamic data required by the aerodynamic database, which is an effective means to support aircraft design, reduce cost and shorten development cycle.

[0018] The single-output Gaussian process can only simulate the output individually in the multi-output case, but in the complex physical process, there is often correlation between multi-dimensional outputs, and the multi-output Gaussian process model is used to process the problem of multi-dimensional correlated output, which models the output and can transmit useful information across the output. However, directly using the multi-output model to fit requires a large number of high-confidence sample points to ensure the accuracy of the surrogate model, and the acquisition of high-confidence sample points requires more computing resources and cost. Therefore, the aircraft aerodynamic data information generated on the basis of considering the computing resources and cost is incomplete, resulting in the loss of some useful information.

[0019] Therefore, the present application provides an aircraft aerodynamic data generation scheme, which can utilize the correlation between multi-outputs to generate high-confidence, multi-output aircraft aerodynamic data on the basis of considering low computing resources and computing cost.

[0020] Referring to Figure 1 The present application provides an aircraft aerodynamic data generation method, comprising: Step S11: Obtain aerodynamic design parameter training data and corresponding aerodynamic output data at different confidence levels to construct a preset low-confidence sample pool and a preset high-confidence sample pool.

[0021] In this embodiment, the aerodynamic design parameter training data and the corresponding aerodynamic output data are obtained by linear flow solver calculation or small wind tunnel test, and are used as preset low-confidence samples. The aerodynamic design parameter training data and the corresponding aerodynamic output data are obtained by nonlinear flow solver calculation, full-size wind tunnel test or flight test, and are used as preset high-confidence samples. The physical fidelity and computing test cost of the linear flow solver calculation or small wind tunnel test are much lower than those of the nonlinear flow solver calculation, full-size wind tunnel test or flight test. Therefore, the sample number of the preset low-confidence samples is much larger than that of the preset high-confidence samples. The aerodynamic output data is the dependent variable to be obtained by the above CFD calculation or wind tunnel test, and is the core index for measuring the performance of the aircraft, and includes lift coefficient, drag coefficient, pitch moment coefficient, yaw moment coefficient, roll moment coefficient, etc. without specific limitation. Then, the corresponding preset low-confidence sample pool and preset high-confidence sample pool are constructed based on the obtained different confidence samples.

[0022] Step S12: Construct a multi-output preset low-confidence Gaussian process model using preset low-confidence training samples randomly selected from the preset low-confidence sample pool.

[0023] In this embodiment, some preset low-confidence training samples are randomly selected from a preset low-confidence sample pool, and then a preset low-confidence Gaussian process model is constructed and trained by using the aerodynamic design parameter training data of the randomly selected preset low-confidence training samples, a selected kernel function, aerodynamic output data corresponding to the aerodynamic design parameter training data, and a selected covariance matrix parameterization method.

[0024] Specifically, the similarity between the aerodynamic design parameter training data in the preset low-confidence training samples is calculated by using a preset kernel function to construct a first kernel matrix; a first covariance matrix representing the correlation between the aerodynamic data is constructed; and the first kernel matrix and the first covariance matrix are combined by Kronecker product to construct a first total covariance structure of the preset low-confidence Gaussian process model; and the parameters of the first kernel matrix and the parameters of the first covariance matrix are determined according to the aerodynamic output data in the preset low-confidence training samples and by maximum likelihood estimation to determine the initial structure parameters of the first total covariance structure, thereby constructing a multi-output preset low-confidence Gaussian process model. It can be understood that the preset kernel function can be any one of a Gaussian kernel function, a Matérn function ( ), and a Matérn function ( ). The smoothness parameter is represented by s; after selecting the kernel function, the similarity between each pair of aerodynamic design parameter training data is calculated by the kernel function, and these similarities collectively fill a first kernel matrix, which represents the spatial correlation structure between all aerodynamic design parameter training data in the low-confidence training samples and serves as the covariance structure basis of the preset low-confidence Gaussian process model. Then, according to the first kernel matrix, a first covariance matrix is further constructed based on the correlation between the corresponding aerodynamic output data, the first covariance matrix adopts the free-form parameterization method proposed by Bonilla et al., and is parameterized by Cholesky decomposition; then, the parameters of the first kernel matrix and the parameters of the first covariance matrix are determined by maximum likelihood estimation and based on the aerodynamic output data corresponding to the randomly selected low-confidence aerodynamic design parameter training data in the corresponding low-confidence training samples to determine the initial structure parameters of the first total covariance structure, thereby constructing a multi-output preset low-confidence Gaussian process model.

[0025] Step S13: A new training sample is adaptively selected from the preset low-confidence sample pool by an entropy-based adaptive sampling method to join the preset low-confidence training samples, and the preset low-confidence Gaussian process model is re-iterated and trained to obtain an optimized preset low-confidence Gaussian process model.

[0026] In this embodiment, the first covariance matrix of the joint probability distribution at each first candidate sample point in the preset low-confidence sample pool is calculated using the preset low-confidence Gaussian process model of the current iteration; the first prediction entropy corresponding to each first candidate sample point is calculated based on the first covariance matrix; the candidate sample point corresponding to the maximum prediction entropy among all first prediction entropies is selected as a new training sample to be added to the preset low-confidence training sample; the preset kernel function is used to calculate the similarity between the aerodynamic design parameter training data in the preset low-confidence training sample to construct the first kernel matrix, and the initial structure parameter of the preset low-confidence Gaussian process model is iterated until the preset low-confidence iteration termination condition is reached, and an optimized preset low-confidence Gaussian process model containing the target structure parameter is obtained. It can be understood that the entropy-based adaptive sampling method increases the low-confidence aerodynamic design parameters and aerodynamic output data. Specifically, in addition to the above-mentioned randomly selected training samples, other low-confidence samples are further screened from the preset low-confidence sample pool. Specifically, the current secondary screening is to calculate the first covariance matrix of the joint probability distribution of each first candidate sample point (other low-confidence samples) based on the entropy-based adaptive sampling method, and then calculate the corresponding first prediction entropy through the first covariance matrix, and then select the candidate sample point with the maximum prediction entropy as a new training sample to be added, and re-iterate the training until the maximum iteration number or the entropy reduction is less than the threshold value, to obtain the final multi-output optimized preset low-confidence Gaussian process model. The formula corresponding to the candidate sample point added to the new training sample by the entropy-based adaptive sampling method considering the correlation between outputs is as follows: ; wherein, is the first covariance matrix of the multi-output low-confidence Gaussian process model prediction in the current step of the iteration process, is the determinant of the matrix, is a constant.

[0027] Step S14: based on the optimized preset low-confidence Gaussian process model and the preset high-confidence training sample, a multi-output preset multi-confidence Gaussian process model is constructed; wherein the preset high-confidence training sample is a training sample randomly selected from the preset high-confidence sample pool.

[0028] Further, after obtaining the optimized preset low-confidence Gaussian process model that improves the global trend prediction ability of the base model at the lowest cost, a bias process is used for calibration to further obtain a high-confidence model without learning all physical laws from scratch, which greatly reduces the model complexity and the demand for the number of high-confidence samples. The specific training process is as follows: According to the preset high-trust training sample and the preset low-trust training sample, a mixed training sample set is constructed; a similarity between the aerodynamic design parameter training data of the training samples in the mixed training sample set is calculated by using the preset kernel function same as the optimized preset low-trust Gaussian process model, so as to construct a second kernel matrix; a covariance matrix same as the optimized preset low-trust Gaussian process model and representing the correlation between the aerodynamic data is used as a second covariance matrix; the second kernel matrix, the second covariance matrix and a scaling factor vector representing the correlation between different trust levels are fused, so as to construct a second total covariance structure of the preset multi-trust Gaussian process model; the vector value of the scaling factor vector, the parameters of the second kernel matrix and the parameters of the second covariance matrix are determined according to the aerodynamic output data in the preset high-trust training sample and by maximum likelihood estimation, so as to determine the initial structure parameters of the second total covariance structure, and a preset multi-trust Gaussian process model with multiple outputs is constructed. It can be understood that a multi-output multi-trust Gaussian process model is constructed according to the final optimized preset low-trust Gaussian process model with multiple outputs and the high-trust input-output training sample selected from the preset high-trust training sample pool; in the preset multi-trust Gaussian process model with multiple outputs, the second kernel function between the inputs and the second covariance matrix between the outputs are consistent with the selection in the preset low-trust Gaussian process model, and details are not described herein, wherein the scaling factor vector quantifies the global deviation relationship between the low-trust model output and the real high-trust output, and the scaling factor vector is an unknown parameter to be determined in the construction of the multi-trust Gaussian process model, which is put into a unified optimization framework together with the hyperparameters of the kernel function and the parameters of the second covariance matrix for optimization, specifically, the model maximizes the marginal likelihood function of the model by using an optimization algorithm (conjugate gradient method, BFGS, etc.), so as to simultaneously and automatically solve the optimal scaling factor vector and all other parameters, and the goal is to find a set of parameters that can best explain the observed high-trust training data.

[0029] Step S15: The entropy-based adaptive sampling method adaptively selects new training samples from the preset high-trust sample pool to join the preset high-trust training sample, and iteratively trains the preset multi-trust Gaussian process model again, so as to obtain a preset aerodynamic data generation model.

[0030] Further iteration and optimization are performed on the preset multi-trust Gaussian process model with multiple outputs, and the specific steps are as follows: Calculate a second covariance matrix of the joint probability distribution at each second candidate sample point in the preset high-confidence sample pool using the preset multi-confidence Gaussian process model of the current iteration; calculate a second prediction entropy corresponding to each second candidate sample point based on the second covariance matrix; select the candidate sample point corresponding to the maximum prediction entropy among all second prediction entropies as a new training sample to be added to the preset high-confidence training sample; use the preset high-confidence training sample to jump to execute the step of constructing a hybrid training sample set according to the preset high-confidence training sample and the preset low-confidence training sample, until the initial structure parameter of the preset multi-confidence Gaussian process model reaches a preset multi-confidence iteration termination condition, to obtain a preset aerodynamic data generation model containing a target structure parameter. It can be understood that the entropy-based adaptive sampling method is used to increase high-confidence input-output samples, and then a multi-output multi-confidence Gaussian process model is iterated until the maximum iteration number is reached or the entropy reduction is less than a threshold, to obtain a final multi-output multi-confidence Gaussian process model. It should be noted that the entropy-based adaptive sampling method used in the optimization of the preset multi-confidence Gaussian process model is consistent with the self-sampling method used in the optimization process of the preset low-confidence Gaussian process model, which will not be described here.

[0031] In this embodiment, before the aerodynamic design parameters are input into the preset aerodynamic data generation model, the preset aerodynamic data generation model needs to be constructed and trained in advance, specifically as follows: The preset aerodynamic data generation model is obtained by fusing and training based on the preset low-confidence training sample and the preset high-confidence training sample and through Gaussian process modeling and entropy-based adaptive sampling method. It can be understood that a multi-output Gaussian process base model is constructed based on a large number of easily accessible low-confidence samples, and low-confidence samples are iteratively supplemented in the area with the highest uncertainty by using the entropy-based adaptive sampling method, thereby improving the global trend prediction ability of the model at the lowest cost. Then, the optimized low-confidence model is fused with a small number of high-confidence samples to construct a multi-confidence model, and then the entropy-driven adaptive sampling is applied again to select high-confidence sample points to join the training, so that the preset aerodynamic data generation model capable of accurately predicting high-confidence and multi-output aerodynamic data is obtained under the premise of controlling the amount of high-cost samples. Step S16: Obtain the aerodynamic design parameters of the target aircraft.

[0032] In this embodiment, the design conditions or flight conditions of the target aircraft are obtained, wherein the design conditions or flight conditions are independent variables set for the target aircraft, and specifically, the aerodynamic design parameters include Mach number, angle of attack, side slip angle, control surface deflection angle, flight altitude, etc., which are not limited here.

[0033] Step S17: input the aerodynamic design parameters into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameters, and then performs fusion operation on the kernel function vector, scaling factors between different credibility levels, and a covariance matrix representing the correlation between aerodynamic data, to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameters.

[0034] In this embodiment, the aerodynamic design parameters of the target aircraft are input into the preset aerodynamic data generation model, and the model can predict aerodynamic data based on the relationships between the aerodynamic design parameters learned in the training process, the relationships between the aerodynamic design parameters and the aerodynamic data, and the correlations between the predicted aerodynamic data, to obtain an aerodynamic data prediction mean vector table representing the predicted values of different aerodynamic data labels.

[0035] Step S18: output the aerodynamic data corresponding to the aerodynamic data prediction mean vector to obtain multi-output aerodynamic data of the target aircraft.

[0036] In this embodiment, the aerodynamic data corresponding to the aerodynamic data prediction mean vector is output, specifically, the aerodynamic data (lift coefficient, drag coefficient, and pitching moment coefficient) corresponding to a vector containing a lift coefficient label, a drag coefficient label, and a pitching moment coefficient label in the table is output, and then multi-output aerodynamic data is obtained.

[0037] It can be seen that the present application discloses obtaining aerodynamic design parameter training data and corresponding aerodynamic output data under different confidence levels to construct a preset low-confidence sample pool and a preset high-confidence sample pool; a preset low-confidence Gaussian process model with multiple outputs is constructed by using preset low-confidence training samples randomly selected from the preset low-confidence sample pool; new training samples are adaptively selected from the preset low-confidence sample pool by an entropy-based adaptive sampling method to join the preset low-confidence training samples, and the preset low-confidence Gaussian process model is re-iteratively trained to obtain an optimized preset low-confidence Gaussian process model; a preset multi-confidence Gaussian process model with multiple outputs is constructed based on the optimized preset low-confidence Gaussian process model and preset high-confidence training samples; wherein the preset high-confidence training samples are training samples randomly selected from the preset high-confidence sample pool; new training samples are adaptively selected from the preset high-confidence sample pool by an entropy-based adaptive sampling method to join the preset high-confidence training samples, and the preset multi-confidence Gaussian process model is re-iteratively trained to obtain a preset aerodynamic data generation model; aerodynamic design parameters of a target aircraft are obtained; the aerodynamic design parameters are input into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameters, and then the kernel function vector, a scaling factor between different confidence levels, and a covariance matrix representing the correlation between aerodynamic data are fused to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameters; the aerodynamic data corresponding to the aerodynamic data prediction mean vector is output to obtain multi-output aerodynamic data of the target aircraft. It can be seen that, when the preset aerodynamic data generation model is used for prediction and output, other output information can be used for fusion operation through the covariance matrix and the scaling factor, so that useful information is transmitted across outputs, it is ensured that a set of physically coordinated aerodynamic data is generated, physical contradictions and information loss caused by separate prediction of each output are avoided, the problem of large demand for high-confidence sample quantity is solved through multi-confidence fusion, and meanwhile, information is fully utilized under the condition of limited samples through multi-output modeling, that is, the shortage of high-confidence sample quantity is made up by introducing a large amount of low-confidence data and using the correlation between outputs. The model can also ensure that the output multi-output aerodynamic data has high prediction accuracy and reliability even if only a small amount of high-confidence samples are used.

[0038] As Figure 2 shown, the present application discloses a construction and training process of a preset aerodynamic data generation model, and the process is as follows: Step one: obtain low-confidence input-output training samples and high-confidence input-output training samples.

[0039] In step one, the low-trust and high-trust inputs refer to the input parameters of wind tunnel tests or CFD simulations, such as Mach number, angle of attack, etc.

[0040] In step one, the low-trust and high-trust outputs refer to the outputs of wind tunnel tests or CFD simulations, such as lift coefficient, drag coefficient, pitch moment, etc.

[0041] Step two: Construct a multi-output low-trust Gaussian process model based on the low-trust input and output training samples.

[0042] In step two, the kernel function between inputs in the multi-output low-trust Gaussian process model can be Gaussian kernel function, Matérn function ( ), Matérn function ( ), etc.

[0043] In step two, the covariance matrix between outputs in the multi-output low-trust Gaussian process model uses the "free-form" parameterization method proposed by Bonilla et al., which is parameterized by Cholesky decomposition.

[0044] Step three: The entropy-based adaptive sampling method increases the low-trust input and output samples, and then iterates the multi-output low-trust Gaussian process model until it reaches the maximum number of iterations or the entropy reduction is less than the threshold, obtaining the final multi-output low-trust Gaussian process model.

[0045] In step three, the entropy-based adaptive sampling method considers the correlation between outputs for adding points: ; where, is the covariance matrix predicted by the multi-output low-trust Gaussian process model at the current step of the iteration process, is the determinant of the matrix, is a constant.

[0046] Step four: Construct a multi-output multi-trust Gaussian process model based on the final multi-output low-trust Gaussian process model and the high-trust input and output training samples.

[0047] In step four, the kernel function between inputs and the covariance matrix between outputs in the multi-output multi-trust Gaussian process model are consistent with the selection in step two.

[0048] Step five: The entropy-based adaptive sampling method increases the high-trust input and output samples, and then iterates the multi-output multi-trust Gaussian process model until it reaches the maximum number of iterations or the entropy reduction is less than the threshold, obtaining the final multi-output multi-trust Gaussian process model.

[0049] In step five, the entropy-based adaptive sampling method is consistent with the formula in step three, where the represents the covariance matrix of the multi-credibility multi-output Gaussian process model prediction of the multi-output in the current step of the iteration process.

[0050] In this way, the entropy-based adaptive sampling method takes into account the correlation between outputs, and the prediction accuracy of multiple outputs is more stable compared to the usual multi-output adaptive sampling method based on cross-validation and MSE. Compared with the random point adding multi-credibility multi-output Gaussian process model, the adaptive multi-credibility multi-output model based on entropy in the present application has better prediction effect.

[0051] The present embodiment uses MMGP (Multi-credibility-Multi-output-Gaussian-Process, multi-credibility multi-output Gaussian process model), CV-MSE-MMGP (Cross-Validation-Mean Square Error-MMGP, multi-credibility multi-output adaptive Gaussian process model based on cross-validation and mean square error), and Entropy-MMGP (multi-credibility multi-output adaptive Gaussian process model based on entropy proposed in the present application) to predict the NRMSE (Normalized Root Mean Square Error, normalized root mean square error) box plot of a numerical example with one-dimensional input and three-dimensional output to verify the adaptive multi-credibility multi-output Gaussian process model given in the present application.

[0052] The high-credibility model is: ; wherein, , , represents the three outputs of the high-credibility model, represents the nonlinear oscillation basis function, which is the nonlinear oscillation source of the high-credibility model, is a derived output that has a strong explicit correlation with is an output with similar patterns but independent of , which simulates another output in the high-credibility model that has similar frequency and nonlinear characteristics, is an output that has no explicit function relationship with .

[0053] In this way, there is a strong nonlinear transformation relationship between ; and There are similar oscillation modes among them. Therefore, the preset aerodynamic data generation model combines features such as nonlinearity, oscillation, modulation, and transformation. Its complexity simulates the correlation between multiple outputs in real aerodynamic data.

[0054] The low-confidence model is: ;

[0055] in, , , These represent the three outputs of the low-confidence model. Indicates in Based on this, approximate values ​​for the deviation and scaling of the composite function are introduced. Indicates in Based on this, frequency distortion occurred and an approximation with a trend bias was added. High-reliability independent output The approximate value is directly superimposed with additional oscillation interference.

[0056] The three surrogate model methods are MMGP, CV-MSE-MMGP, and Entropy-MMGP. The prediction accuracy is selected as NRMSE. For each dimension of the output: ; in, Indicates the first The predicted value of a certain dimension of the output for a sample. Indicates the first The true value of a sample's output in a certain dimension Indicates the number of predicted samples. It is the mean of the output of a certain dimension.

[0057] The specific steps are as follows: Step 1: Randomly sample from a uniform distribution in [0,1] to obtain 1000 low-confidence inputs. and 1000 highly reliable input samples The corresponding output samples are calculated using both the low-confidence model and the high-confidence model. and ; Randomly select 5 high-confidence input-output samples from them. and 10 low-confidence input-output samples and As the initial training set.

[0058] Step 2: Train samples based on low-confidence input and output samples and Construct a low-confidence Gaussian process model G with multiple outputs. Entropy-LThe kernel function to choose is the Matérn function ( ).

[0059] Step 3: From the remaining low-confidence input / output samples (from 1000 low-confidence input / output samples) and Excluded and From the subsequent 990 samples, an entropy-based adaptive sampling method selects one low-confidence input / output sample and adds it to the initial training samples. and In this process, the multi-output low-confidence Gaussian process model G is retrained. Entropy-L Repeat this process until the set number of low-confidence adaptive samples is reached (set to 5), to obtain the final multi-output low-confidence Gaussian process model G. Entropy-L-end .

[0060] Step 4: Based on the multi-output low-confidence Gaussian process model G obtained in Step 3 Entropy-L-end and high-confidence input / output samples and Construct a multi-output, multi-credible Gaussian process model G Entropy-LH The kernel function to choose is the Matérn function ( ).

[0061] Step 5: From the remaining high-confidence input / output samples (from 1000 high-confidence input / output samples) and Excluded and From the subsequent 995 samples, one high-confidence input / output sample is selected based on the entropy adaptive sampling method and added to the initial training samples. and In the process of retraining the multi-output, multi-credible Gaussian process model G, Entropy-LH Repeat this process until the set number of high-confidence adaptive samples is reached (set to 5), to obtain the final multi-output multi-confidence Gaussian process model G. Entropy-LH-end .

[0062] Step Six: The initial number of training samples, adaptive number of samples, and kernel function of the CV-MSE-MMGP model are kept consistent with those of the Entropy-MMGP model. The final multi-output, multi-confidence Gaussian process model is denoted as G. CV-MSE-LH-end .

[0063] Step 7: In 1000 low-confidence input-output samples ( and ) and 1000 high-confidence input-output samples ( and In the process, 10 high-confidence input-output samples and 15 low-confidence input-output samples are randomly selected to construct a multi-output, multi-confidence Gaussian process model G. MMGP .

[0064] The three models Entropy-MMGP, CV-MSE-MMGP, and MMGP were all constructed 10 times. The remaining high-confidence samples (1000 high-confidence input and output samples, excluding the initial training samples and adaptive or random point additions) were used as test samples. The NRMSE was calculated and repeated 10 times.

[0065] Combining the three outputs, the Entropy-MMGP model based on entropy, proposed in this invention, predicts NRMSE with better overall accuracy and stability than both the CV-MSE-MMGP and MMGP models.

[0066] like Figure 3 As shown, the present invention also discloses an aerodynamic data generation device for aircraft, comprising: The sample acquisition module 11 is used to acquire aerodynamic design parameter training data and corresponding aerodynamic output data under different confidence levels, so as to construct a preset low confidence sample pool and a preset high confidence sample pool respectively. The first model building module 12 is used to build a multi-output preset low-confidence Gaussian process model using preset low-confidence training samples randomly selected from the preset low-confidence sample pool. The first model optimization module 13 is used to select new training samples from the preset low confidence sample pool and add them to the preset low confidence training sample pool using an entropy-based adaptive sampling method, and to re-train the preset low confidence Gaussian process model to obtain an optimized preset low confidence Gaussian process model. The second model construction module 14 is used to construct a multi-output preset multi-confidence Gaussian process model based on the optimized preset low-confidence Gaussian process model and the preset high-confidence training samples; wherein, the preset high-confidence training samples are training samples randomly selected from the preset high-confidence sample pool. The second model optimization module 15 is used to adaptively select new training samples from the preset high-confidence sample pool and add them to the preset high-confidence training samples using an entropy-based adaptive sampling method, and to re-iterate and train the preset multi-confidence Gaussian process model to obtain a preset aerodynamic data generation model. Parameter acquisition module 16 is used to acquire the aerodynamic design parameters of the target aircraft; The multi-output prediction module 17 is configured to input the aerodynamic design parameter into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameter, and then performs fusion operation on the kernel function vector, scaling factors between different credibility levels, and a covariance matrix representing correlation between aerodynamic data, to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameter. The multi-output data generation module 18 is configured to output aerodynamic data corresponding to the aerodynamic data prediction mean vector, to obtain multi-output aerodynamic data of the target aircraft.

[0067] Therefore, the fusion operation algorithm based on the kernel function vector, the scaling factors and the covariance matrix ensures the multi-output correlation and multi-credibility fusion.

[0068] Further, the embodiment of the present application further discloses an electronic device, Figure 4 The electronic device 20 is shown in a structural diagram according to an exemplary embodiment, and the content in the diagram cannot be considered as any limitation on the use range of the present application.

[0069] Figure 4 A structural diagram of an electronic device 20 is provided in the embodiment of the present application. The electronic device 20 specifically can 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 configured to store a computer program, and the processor 21 is configured to load and execute the computer program to implement the related steps in the aircraft aerodynamic data generation method disclosed in any of the preceding embodiments. In addition, the electronic device 20 in the embodiment specifically can be an electronic computer.

[0070] In the embodiment, the power supply 23 is configured 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 external devices, and the communication protocol followed by the communication interface 24 can be any communication protocol applicable to the technical solution of the present application, which is not limited specifically herein; the input / output interface 25 is configured to obtain external input data or output data to the outside, and the specific interface type can be selected according to the specific application needs, which is not limited specifically herein.

[0071] The processor 21 can include one or more processing cores, such as a 4-core processor, an 8-core processor, etc. The processor 21 can be implemented in at least one of a hardware form of a DSP (Digital Signal Processing), an FPGA (Field-Programmable Gate Array), a PLA (Programmable Logic Array). The processor 21 can also include a main processor and a coprocessor, the main processor being a processor for processing data in a wake-up state, also known as a CPU (Central Processing Unit), and the coprocessor being a low-power processor for processing data in a standby state. In some embodiments, the processor 21 can be integrated with a GPU (Graphics Processing Unit) that is responsible for rendering and drawing the content required to be displayed by the display screen. In some embodiments, the processor 21 can also include an AI (Artificial Intelligence) processor for processing machine learning-related computing operations.

[0072] In addition, the memory 22, as a carrier for storing resources, can be a read-only memory, a random access memory, a magnetic disk or an optical disk, etc., and the resources stored thereon can include an operating system 221, a computer program 222, etc., and the storage mode can be temporary storage or permanent storage.

[0073] The operating system 221 is used to manage and control each hardware device and the computer program 222 on the electronic device 20 to realize the operation and processing of the processor 21 on the mass data 223 in the memory 22, and can be Windows Server, Netware, Unix, Linux, etc. In addition to including the computer program capable of completing the aircraft aerodynamic data generation method disclosed in any of the preceding embodiments executed by the electronic device 20, the computer program 222 can further include a computer program capable of completing other specific work. The data 223 can include data received by the electronic device from an external device, as well as data collected by the input / output interface 25, etc.

[0074] Further, the present application also discloses a computer readable storage medium for storing a computer program; wherein the computer program is executed by a processor to implement the aircraft aerodynamic data generation method disclosed above. For the specific steps of the method, please refer to the corresponding content disclosed in the preceding embodiments, which will not be repeated here.

[0075] The various embodiments described in this specification are presented purely by way of example and for purposes of illustration only. Each of the various embodiments highlights aspects that are different from those of other embodiments. The same or similar elements are denoted by the same reference numerals throughout the various drawings and the specification. As such, the various embodiments can not only disclose particular combinations of elements which are of interest, but the specification can also encompass all the possible combinations that can be achieved by a combination of the elements described or claimed. The specification can also encompass the various embodiments as they can be applied to or implemented in various apparatuses and methods.

[0076] Those skilled in the art will further appreciate that the units and algorithm steps of the various examples described in connection with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or combinations of both. To clearly illustrate this interchangeability of hardware and software, various examples disclosed herein have been described in general terms that are applicable to both. It will also be appreciated that differing embodiments of the described examples can include different combinations of the units and algorithm steps described and / or claimed, and the skilled person will appreciate that the units and algorithm steps can be implemented in electronic hardware, computer software or both. The software modules can be stored in a Random Access Memory (RAM), a memory, a Read Only Memory (ROM), an Electrically Programmable ROM (EPROM), an Electrically Erasable Programmable ROM (EEPROM), registers, a hard disk, a removable disk, a CD-ROM (Compact Disc - Read Only Memory), or any other form of storage medium known in the art.

[0077] Finally, it has to be noted that the terms like first and second, etc. are used herein for distinguishing between similar elements and not necessarily for describing a sequential or chronological order. Furthermore, the terms "comprising", "containing", or any other analogous terms are intended to encompass the presence of stated elements or features, but do not preclude the presence or addition of one or more other elements or features. In other words, it is meant that stated elements or ingredients can be present or added and still compose a part of the application. Furthermore, it is also intended that the terms "including", "containing", or any other analogous terms encompass the presence of stated elements or features, but do not preclude the presence or addition of one or more other elements or features. In other words, it is meant that stated elements or ingredients can be present or added and still compose a part of the application. In addition, it has to be noted that the terms "comprising", "containing", or any other analogous terms are used herein in an open-ended way, i.e. meaning that additional elements can be present.

[0078] The above has carried out the detailed introduction to the scheme provided by the application, the principle and implementation mode of the application are described by applying specific examples in this paper, the above example description is only applicable to help understanding the method and core idea of the application; at the same time, for the general technical personnel in the art, according to the idea of the application, the specific implementation mode and application range will have changes, and the above is described, the content of the specification should not be understood as the limitation of the application.

Claims

1. A method of generating aerodynamic data for an aircraft, characterized in that, The method comprises the following steps: acquiring aerodynamic design parameter training data and corresponding aerodynamic output data at different levels of credibility to construct a preset low-credibility sample pool and a preset high-credibility sample pool, respectively; constructing a preset low-credibility Gaussian process model with multiple outputs by using preset low-credibility training samples randomly selected from the preset low-credibility sample pool; selecting new training samples from the preset low-credibility sample pool by using an entropy-based adaptive sampling method to add the new training samples to the preset low-credibility training samples, reiteratively training the preset low-credibility Gaussian process model, and obtaining an optimized preset low-credibility Gaussian process model; constructing a preset multi-credibility Gaussian process model with multiple outputs based on the optimized preset low-credibility Gaussian process model and preset high-credibility training samples, wherein the preset high-credibility training samples are training samples randomly selected from the preset high-credibility sample pool; selecting new training samples from the preset high-credibility sample pool by using the entropy-based adaptive sampling method to add the new training samples to the preset high-credibility training samples, and reiteratively training the preset multi-credibility Gaussian process model to obtain a preset aerodynamic data generation model; acquiring aerodynamic design parameters of a target aircraft; inputting the aerodynamic design parameters into the preset aerodynamic data generation model to enable the preset aerodynamic data generation model to calculate a kernel function vector of the aerodynamic design parameters, and then performing fusion operation on the kernel function vector, scaling factors between different levels of credibility, and a covariance matrix representing the correlation between aerodynamic data to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameters; outputting aerodynamic data corresponding to the aerodynamic data prediction mean vector to obtain multi-output aerodynamic data of the target aircraft.

2. The aircraft aerodynamic data generation method of claim 1, wherein, The method of constructing a preset low-credibility Gaussian process model with multiple outputs by using preset low-credibility training samples randomly selected from the preset low-credibility sample pool comprises the following steps: calculating the similarity between the aerodynamic design parameter training data in the preset low-credibility training samples by using a preset kernel function to construct a first kernel matrix; constructing a first covariance matrix representing the correlation between aerodynamic data; combining the first kernel matrix and the first covariance matrix by Kronecker product to construct a first total covariance structure of the preset low-credibility Gaussian process model; determining the parameters of the first kernel matrix and the parameters of the first covariance matrix based on the aerodynamic output data in the preset low-credibility training samples and by maximum likelihood estimation to determine the initial structure parameters of the first total covariance structure, and constructing a preset low-credibility Gaussian process model with multiple outputs.

3. The aircraft aerodynamic data generation method of claim 2, wherein, The method of selecting new training samples from the preset low-credibility sample pool by using the entropy-based adaptive sampling method to add the new training samples to the preset low-credibility training samples, reiteratively training the preset low-credibility Gaussian process model, and obtaining an optimized preset low-credibility Gaussian process model comprises the following steps: calculating, by using the preset low-confidence Gaussian process model of the current iteration, a first covariance matrix of a joint probability distribution at each first candidate sample point in the preset low-confidence sample pool; calculating, based on the first covariance matrix, a first prediction entropy corresponding to each first candidate sample point; selecting a candidate sample point corresponding to a maximum prediction entropy among all first prediction entropies as a new training sample to be added to the preset low-confidence training sample; utilizing the preset low-confidence training sample, jumping to performing the step of calculating the similarity between the aerodynamic design parameter training data in the preset low-confidence training sample by using the preset kernel function to construct a first kernel matrix until the initial structural parameters of the preset low-confidence Gaussian process model reach a preset low-confidence iteration termination condition, and obtaining an optimized preset low-confidence Gaussian process model containing target structural parameters.

4. The aircraft aerodynamic data generation method of claim 1, wherein, The preset multi-confidence Gaussian process model is constructed based on the optimized preset low-confidence Gaussian process model and the preset high-confidence training sample, and includes: constructing a hybrid training sample set according to the preset high-confidence training sample and the preset low-confidence training sample; calculating the similarity between the training sample in the hybrid training sample set by using the same preset kernel function as the optimized preset low-confidence Gaussian process model to construct a second kernel matrix; utilizing the same covariance matrix as the optimized preset low-confidence Gaussian process model to represent the correlation between the aerodynamic data as a second covariance matrix; fusing the second kernel matrix, the second covariance matrix and a scaling factor vector representing the association relationship between different confidence levels to construct a second total covariance structure of the preset multi-confidence Gaussian process model; determining the vector value of the scaling factor vector, the parameters of the second kernel matrix and the parameters of the second covariance matrix according to the aerodynamic output data in the preset high-confidence training sample and by maximum likelihood estimation, to determine the initial structural parameters of the second total covariance structure, and construct a multi-output preset multi-confidence Gaussian process model.

5. The aircraft aerodynamic data generation method according to claim 4, wherein, The entropy-based adaptive sampling method adaptively selects a new training sample from the preset high-confidence sample pool to be added to the preset high-confidence training sample, and iteratively trains the preset multi-confidence Gaussian process model to obtain a preset aerodynamic data generation model, and includes: calculating, by using the preset multi-confidence Gaussian process model of the current iteration, a second covariance matrix of a joint probability distribution at each second candidate sample point in the preset multi-high-confidence sample pool; calculating, based on the second covariance matrix, a second prediction entropy corresponding to each second candidate sample point; selecting a candidate sample point corresponding to a maximum prediction entropy among all second prediction entropies as a new training sample to be added to the preset high-confidence training sample; The preset high-confidence training sample is used to jump to perform the step of constructing a mixed training sample set according to the preset high-confidence training sample and the preset low-confidence training sample until an initial structure parameter of the preset multi-confidence Gaussian process model reaches a preset multi-confidence iteration termination condition, and a preset aerodynamic data generation model containing a target structure parameter is obtained.

6. An aircraft aerodynamic data generating apparatus, characterized by, The method comprises the steps of: a sample acquisition module, configured to acquire aerodynamic design parameter training data and corresponding aerodynamic output data under different confidence levels, to construct a preset low-confidence sample pool and a preset high-confidence sample pool, respectively; a first model construction module, configured to construct a preset low-confidence Gaussian process model with multiple outputs by using preset low-confidence training samples randomly selected from the preset low-confidence sample pool; a first model optimization module, configured to select new training samples from the preset low-confidence sample pool by using an entropy-based adaptive sampling method, to add the new training samples to the preset low-confidence training samples, and to re-iterate training of the preset low-confidence Gaussian process model to obtain an optimized preset low-confidence Gaussian process model; a second model construction module, configured to construct a preset multi-confidence Gaussian process model with multiple outputs based on the optimized preset low-confidence Gaussian process model and preset high-confidence training samples, wherein the preset high-confidence training samples are training samples randomly selected from the preset high-confidence sample pool; a second model optimization module, configured to select new training samples from the preset high-confidence sample pool by using the entropy-based adaptive sampling method, to add the new training samples to the preset high-confidence training samples, and to re-iterate training of the preset multi-confidence Gaussian process model to obtain a preset aerodynamic data generation model; a parameter acquisition module, configured to acquire aerodynamic design parameters of a target aircraft; a multi-output prediction module, configured to input the aerodynamic design parameters into the preset aerodynamic data generation model, so that the preset aerodynamic data generation model calculates a kernel function vector of the aerodynamic design parameters, and then performs fusion operation on the kernel function vector, scaling factors between different confidence levels, and a covariance matrix representing correlation between aerodynamic data, to generate a multi-output aerodynamic data prediction mean vector corresponding to the aerodynamic design parameters; a multi-output data generation module, configured to output aerodynamic data corresponding to the aerodynamic data prediction mean vector, to obtain multi-output aerodynamic data of the target aircraft.

7. An electronic device, comprising: The method comprises the steps of: a memory, configured to save a computer program; a processor, configured to execute the computer program to implement the steps of the aircraft aerodynamic data generation method according to any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, a memory, configured to save a computer program; a processor, configured to execute the computer program to implement the steps of the aircraft aerodynamic data generation method according to any one of claims 1 to 5.

Images