Method and device for constructing aerodynamic coefficient prediction model

Abstract

The invention provides an aerodynamic coefficient prediction model construction method and device, and the method comprises the steps: obtaining an aerodynamic training set of an aircraft and a corresponding flow field training set, and carrying out the order reduction of the flow field training set, and obtaining an order reduction flow field feature; a flow field prediction model is obtained based on the aerodynamic force training set and reduced-order flow field feature training; forming a data set by the reduced-order flow field prediction value output by the flow field prediction model and an aerodynamic force training set, and training based on the data set to obtain a linear prediction model and a nonlinear prediction model; and fusing the output of the linear prediction model and the output of the nonlinear prediction model in a prediction layer to obtain a constructed aerodynamic coefficient prediction model. According to the method, reduced-order flow field features are introduced in the model construction process, and related physical information output by a flow field prediction model is fused into prediction of an aerodynamic coefficient; a linear prediction model and a non-linear prediction model are divided when the incidence relation between the data set and the aerodynamic coefficient is analyzed, so that prediction can be flexibly applied to various situations.

Description

Technical Field

[0001] The present invention relates to the field of aircraft technology, and in particular to a method and device for constructing an aerodynamic coefficient prediction model. Background Art

[0002] In the field of aircraft design, it is necessary to evaluate the aerodynamic characteristics of the aircraft, among which the aerodynamic coefficient is a dimensionless parameter that describes the force exerted on the aircraft during its motion.

[0003] In the related technologies, the evaluation methods of aircraft aerodynamic characteristics mainly rely on engineering algorithms such as empirical formulas, Newton's method, and quasi-Newton's method; computational fluid dynamics methods such as Reynolds average equations and large eddy simulation; and other wind tunnel test methods; the prediction accuracy of the above technical means varies, but with the iterative design of aircraft shapes, it is necessary to quickly and accurately predict the aerodynamic coefficients of a large number of aircraft with different shapes and different flight states. Therefore, using machine learning methods to generate a proxy model between aircraft states and aerodynamic coefficients has become another feasible solution;

[0004] Machine learning methods include models such as Gaussian regression, support vector machine and neural network, but the above-mentioned technical means often directly fit the function mapping between the aircraft state and the aerodynamic coefficient, and cannot utilize the physical information of the flow field around the aircraft; in addition, in order to improve the prediction accuracy, the depth and width of the network are usually increased, which is prone to overfitting on a limited training set, affecting the generalization ability of the network.

[0005] Based on the above analysis of the development status of this technical field, the existing technology lacks a solution for integrating flow field information into the model during the modeling process and realizing the prediction of aerodynamic coefficients using a specific network structure. Summary of the invention

[0006] The purpose of the present invention is to provide a method and device for constructing an aerodynamic coefficient prediction model, aiming to solve the above-mentioned problems in the prior art.

[0007] According to a first aspect of an embodiment of the present invention, a method for constructing an aerodynamic coefficient prediction model is provided, comprising:

[0008] Obtaining an aerodynamic training set of an aircraft and a corresponding flow field training set, performing order reduction processing on the flow field training set, and obtaining reduced-order flow field characteristics;

[0009] The flow field prediction model is obtained based on the aerodynamic training set and the reduced-order flow field feature training;

[0010] The reduced-order flow field prediction value output by the flow field prediction model and the aerodynamic training set are combined into a data group, and a linear prediction model and a nonlinear prediction model are obtained based on the training of the data group;

[0011] In the prediction layer, the outputs of the linear prediction model and the nonlinear prediction model are fused to obtain the constructed aerodynamic coefficient prediction model.

[0012] According to a second aspect of an embodiment of the present invention, there is provided a device for constructing an aerodynamic coefficient prediction model, comprising:

[0013] The training set processing module is used to obtain the aerodynamic training set of the aircraft and the corresponding flow field training set, and perform order reduction processing on the flow field training set to obtain the reduced-order flow field characteristics;

[0014] A flow field prediction training module is used to obtain a flow field prediction model based on an aerodynamic training set and reduced-order flow field feature training;

[0015] A dual-branch prediction module is used to form a data group with the reduced-order flow field prediction value output by the flow field prediction model and the aerodynamic training set, and obtain a linear prediction model and a nonlinear prediction model based on the data group training;

[0016] The result fusion module is used to fuse the outputs of the linear prediction model and the nonlinear prediction model at the prediction layer to obtain a constructed aerodynamic coefficient prediction model.

[0017] According to a third aspect of an embodiment of the present invention, there is provided an electronic device, comprising: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the method for constructing an aerodynamic coefficient prediction model as provided in the first aspect of the present disclosure are implemented.

[0018] According to a fourth aspect of an embodiment of the present invention, a computer-readable storage medium is provided, on which a program for implementing information transmission is stored, and when the program is executed by a processor, the steps of the method for constructing an aerodynamic coefficient prediction model provided in the first aspect of the present disclosure are implemented.

[0019] The technical solution provided by the embodiment of the present invention includes the following beneficial effects: introducing reduced-order flow field features in the model construction process, integrating relevant physical information output by the flow field prediction model into the prediction of the aerodynamic coefficients, and reducing the tendency of overfitting in the overall model through the model structure; when analyzing the correlation between the data group and the aerodynamic coefficients, it is divided into a linear prediction model and a nonlinear prediction model, so that the prediction can be flexibly applied to situations dominated by linear or nonlinear factors, thereby improving the accuracy of the prediction of the aerodynamic coefficients.

[0020] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate one or more embodiments of this specification or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in this specification. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.

[0022] Figure 1 is a flow chart of a method for constructing an aerodynamic coefficient prediction model according to an embodiment of the present invention;

[0023] Figure 2 is a schematic diagram of data stream transmission according to an embodiment of the present invention;

[0024] Figure 3 is a schematic diagram of the overall architecture of the prediction model of an embodiment of the present invention;

[0025] Figure 4 is a schematic diagram of a device for constructing an aerodynamic coefficient prediction model according to an embodiment of the present invention;

[0026] Figure 5 is a schematic diagram of an electronic device according to an embodiment of the present invention. DETAILED DESCRIPTION

[0027] In order to enable those skilled in the art to better understand the technical solutions in one or more embodiments of this specification, the following will be combined with the drawings in one or more embodiments of this specification to clearly and completely describe the technical solutions in one or more embodiments of this specification. Obviously, the described embodiments are only part of the embodiments of this specification, not all of the embodiments. Based on one or more embodiments of this specification, all other embodiments obtained by ordinary technicians in this field without creative work should fall within the scope of protection of this document.

[0028] For the convenience of the following description, the technical terms involved in the present invention are explained as follows:

[0029] (1) Aerodynamic coefficients: Aerodynamic force is the force exerted by air on an aircraft when the aircraft moves relative to the air. The corresponding aerodynamic coefficients are obtained by non-dimensionalizing the aerodynamic force using the aircraft reference length, reference area and incoming flow pressure as reference quantities. The aerodynamic coefficients include the lift coefficient, drag coefficient, lateral force coefficient, rolling moment coefficient, yaw moment coefficient and pitch moment coefficient.

[0030] (2) Lift: The component of the aerodynamic force vector that is perpendicular to the direction of flight speed within the longitudinal symmetry plane of the aircraft.

[0031] (3) Drag: The component of the aerodynamic force vector parallel to the flight speed.

[0032] (4) Lateral force: The component remaining after deducting the lift and drag components from the aerodynamic force vector.

[0033] (5) Rolling moment: The moment component of the aerodynamic force generated in the axial direction of the aircraft body axis.

[0034] (6) Yaw moment: The moment component of the aerodynamic force in the normal direction of the aircraft body axis system.

[0035] (7) Pitch moment: The moment component of the aerodynamic force generated laterally on the vehicle's body axis.

[0036] Method Embodiment

[0037] According to an embodiment of the present invention, a method for constructing an aerodynamic coefficient prediction model is provided. Figure 1 is a flow chart of a method for constructing an aerodynamic coefficient prediction model according to an embodiment of the present invention, such as Figure 1 As shown, the method for constructing the aerodynamic coefficient prediction model according to an embodiment of the present invention specifically includes:

[0038] In step S110, an aerodynamic training set of the aircraft and a corresponding flow field training set are obtained, and the flow field training set is reduced to obtain reduced-order flow field features, specifically including:

[0039] The data in the training set are directly obtained through CFD software;

[0040] The Latin hypercube method is used to obtain an aerodynamic training set and a corresponding flow field training set. The Latin hypercube method is a stratified sampling method. Preferably, the optimized Latin hypercube method can be selected to replace the original standard Latin hypercube method to avoid excessive concentration or dispersion of sample points.

[0041] The aerodynamic training set includes state point data and aerodynamic coefficients; the state point data is a set of parameters describing the aircraft under specific flight conditions, and each state point defines a set of flight conditions, such as angle of attack, side attack angle, rudder deflection angle, etc. In the embodiment of the present invention, it is composed of the parametric shape parameters and flight condition parameters of the aircraft, and five shape parameters and three flight condition parameters are selected. Since the relationship between data and aerodynamic coefficients needs to be explored during the model training process, the aerodynamic coefficients exist as known quantities;

[0042] The flow field data of an aircraft is a data set that describes the detailed information of the airflow around the aircraft;

[0043] The variational autoencoder or the intrinsic orthogonal decomposition algorithm is used to reduce the order of the flow field training set to obtain the reduced-order flow field characteristics; the main purpose of reducing the order of the flow field training set is to reduce the computational cost, improve the computational efficiency, simplify the model and realize easy real-time prediction. Figure 2Schematic diagram of data stream transmission according to an embodiment of the present invention. Figure 2 As shown, the reduced-order flow field features are introduced during the training process.

[0044] In step S120, a flow field prediction model is obtained based on the aerodynamic training set and the reduced-order flow field feature training, specifically including:

[0045] The state point data in the aerodynamic training set and the reduced-order flow field characteristics form data pairs. During the training phase, it is necessary to explore the relationship between the aerodynamic state points and the reduced-order flow field.

[0046] A four-layer fully connected neural network is trained using data point pairs, and a reduced-order flow field prediction value is output through the four-layer fully connected neural network. In the embodiment of the present invention, the widths of the intermediate layers are 8, 16, and 16 respectively, and the loss between the reduced-order flow field prediction value and the reduced-order flow field feature is evaluated using mean square error calculation;

[0047] The flow field prediction model is obtained by iteratively using the gradient algorithm according to the loss, wherein the activation function of the flow field prediction model is a hyperbolic tangent function, and the output layer is a linear layer without an activation function; thus, the four-layer fully connected neural network becomes a regression prediction model;

[0048] When the training performance index reaches the predetermined requirement, the trained flow field prediction model is obtained. It is necessary to save all parameters in the flow field prediction model before the next stage of training.

[0049] In step S130, the reduced-order flow field prediction value output by the flow field prediction model and the aerodynamic training set are combined into a data group, and a linear prediction model and a nonlinear prediction model are obtained based on the data group training, specifically including:

[0050] Using the data set to train a single-layer neural network, outputting a first predicted value of the aerodynamic coefficient through the single-layer neural network, and calculating the loss of the first predicted value and the aerodynamic coefficient in the data set using a mean square error;

[0051] A linear prediction model is obtained by iteratively using a gradient algorithm according to the loss, wherein the linear prediction model does not use an activation function;

[0052] Using the data set to train a multi-hidden layer fully connected neural network, outputting a second predicted value of the aerodynamic coefficient through the multi-hidden layer fully connected neural network, and using a mean square error to calculate the loss of the second predicted value and the aerodynamic coefficient in the data set;

[0053] According to the loss, the gradient algorithm is used to iterate to obtain a nonlinear prediction model;

[0054] The linear prediction model can capture the linear relationship between the flow field characteristics and state points and the aerodynamic coefficients; the nonlinear prediction model can capture the nonlinear relationship between the flow field characteristics and state points and the aerodynamic coefficients; the linear relationship indicates that the changes between variables are proportional, and the nonlinear relationship can capture more complex relationships; the linear prediction model and the nonlinear prediction model are set at the same time to adapt to different situations, for example, in some cases, the nonlinear characteristics or linear characteristics are more obvious;

[0055] After the training is completed, the weight parameters of the linear prediction model and the parameters in the nonlinear prediction model are saved, depending on the form of the nonlinear prediction model. In the embodiment of the present invention, the nonlinear prediction model includes but is not limited to a neural network, a decision tree, a support vector machine, etc.

[0056] In step S140, the outputs of the linear prediction model and the nonlinear prediction model are fused at the prediction layer to obtain a constructed aerodynamic coefficient prediction model, which specifically includes:

[0057] The outputs of the linear prediction model and the nonlinear prediction model are fused using the weighted summation method. The prediction layer can determine the weight coefficients and calculate the final output using Formula 1:

[0058] Final output = w*first predicted value + (1-w)*second predicted value Formula 1;

[0059] Wherein, w represents a weight parameter, and 0≤w≤1, and the outputs of the two modules are integrated into a final output value. Since the independent variables in Formula 1 of the embodiment of the present invention only include two values, a statistical method can be used to determine the weight coefficient in the embodiment of the present invention.

[0060] It should be noted that in all models in the embodiments of the present invention, different choices of structural parameters, or other models that can play a consistent role with the corresponding models in the embodiments of the present invention, all produce the same technical effects, such as support vector machines, decision trees, etc.; in addition, replacing the activation function form and loss function in each model does not cause essential changes to the technical solution disclosed in the present invention.

[0061] The method further comprises:

[0062] In step S150, the aerodynamic force verification set including the state point data is input into the aerodynamic force coefficient prediction model to obtain the prediction result of the aerodynamic force coefficient, which specifically includes:

[0063] Since the current stage is the prediction verification stage of the existing model, the aerodynamic coefficient is not included in the aerodynamic verification set. Except for the aerodynamic coefficient, other indicators are consistent with the aerodynamic training set;

[0064] Inputting the aircraft state point into the flow field prediction model to predict the reduced-order flow field prediction value, combining the reduced-order flow field prediction value with the aircraft state point, respectively inputting the linear prediction model and the nonlinear prediction model, outputting the first prediction value and the second prediction value, and obtaining the prediction result of the aerodynamic coefficient after weighted summation;

[0065] During the prediction process, the aircraft state point can be understood as an independent variable, and the output aerodynamic coefficient can be understood as a dependent variable.

[0066] The above technical solutions of the embodiments of the present invention are illustrated with reference to the following drawings.

[0067] Figure 3 Schematic diagram of the overall architecture of the prediction model of the embodiment of the present invention. Figure 3 As shown, the established dual-branch architecture of the prediction model is demonstrated. The aircraft state point data is input into the flow field prediction model, and after combining the reduced-order flow field prediction value and the aircraft state point, they are input into the linear prediction model and the nonlinear prediction model respectively, and then a weighted summation process is performed.

[0068] To sum up, in response to the existing problems, the method for constructing an aerodynamic coefficient prediction model of this invention introduces reduced-order flow field features in the model construction process, and integrates the relevant physical information output by the flow field prediction model into the prediction of the aerodynamic coefficient. The model structure can effectively reduce the tendency of overfitting in the overall model; the reduced-order flow field features used to train the flow field prediction model are obtained based on the original data reduction preprocessing to reduce the computational cost and improve the computational efficiency; when analyzing the correlation between the data group and the aerodynamic coefficient, it is divided into a linear prediction model and a nonlinear prediction model, so that the prediction can be flexibly applied to situations that are mainly linear or nonlinear, thereby improving the accuracy of the aerodynamic coefficient prediction; when the constructed aerodynamic coefficient prediction model is actually applied, it only needs to obtain the state point data to output the predicted aerodynamic coefficient.

[0069] Device Embodiment

[0070] According to an embodiment of the present invention, a device for constructing an aerodynamic coefficient prediction model is provided. Figure 4 Schematic diagram of a device for constructing an aerodynamic coefficient prediction model according to an embodiment of the present invention. Figure 4 As shown, the device for constructing the aerodynamic coefficient prediction model according to an embodiment of the present invention specifically includes:

[0071] The training set processing module 40 is used to obtain the aerodynamic training set of the aircraft and the corresponding flow field training set, and perform order reduction processing on the flow field training set to obtain reduced-order flow field features, which is specifically used for:

[0072] The Latin hypercube method is used to obtain an aerodynamic training set and a corresponding flow field training set, wherein the aerodynamic training set includes state point data and aerodynamic coefficients.

[0073] The flow field training set is preprocessed by reducing the order using a variational autoencoder or an intrinsic orthogonal decomposition algorithm to obtain reduced-order flow field features.

[0074] The flow field prediction training module 42 is used to obtain a flow field prediction model based on the aerodynamic training set and the reduced-order flow field feature training, and is specifically used for:

[0075] The state point data in the aerodynamic training set and the reduced-order flow field features are combined into data pairs;

[0076] A four-layer fully connected neural network is trained using data point pairs, and the reduced-order flow field prediction value is output through the four-layer fully connected neural network. The loss between the reduced-order flow field prediction value and the reduced-order flow field feature is evaluated using mean square error calculation;

[0077] The flow field prediction model is obtained by iteratively using the gradient algorithm according to the loss, wherein the activation function of the flow field prediction model is a hyperbolic tangent function, and the output layer is a linear layer without an activation function.

[0078] The dual-branch prediction module 44 is used to form a data group with the reduced-order flow field prediction value output by the flow field prediction model and the aerodynamic training set, and obtain a linear prediction model and a nonlinear prediction model based on the data group training, specifically for:

[0079] Using the data set to train a single-layer neural network, outputting a first predicted value of the aerodynamic coefficient through the single-layer neural network, and calculating the loss of the first predicted value and the aerodynamic coefficient in the data set using a mean square error;

[0080] A linear prediction model is obtained by iteratively using a gradient algorithm according to the loss, wherein the linear prediction model does not use an activation function;

[0081] Using the data set to train a multi-hidden layer fully connected neural network, outputting a second predicted value of the aerodynamic coefficient through the multi-hidden layer fully connected neural network, and using a mean square error to calculate the loss of the second predicted value and the aerodynamic coefficient in the data set;

[0082] The nonlinear prediction model is obtained by iteratively using the gradient algorithm according to the loss.

[0083] The result fusion module 46 is used to fuse the outputs of the linear prediction model and the nonlinear prediction model at the prediction layer to obtain a constructed aerodynamic coefficient prediction model, which is specifically used for:

[0084] The outputs of the linear prediction model and the nonlinear prediction model are fused using a weighted summation method.

[0085] To sum up, in response to the existing problems, the device for constructing an aerodynamic coefficient prediction model of this invention introduces reduced-order flow field features in the model construction process, and integrates the relevant physical information output by the flow field prediction model into the prediction of the aerodynamic coefficient. The model structure can effectively reduce the tendency of overfitting in the overall model; the reduced-order flow field features used to train the flow field prediction model are obtained based on the original data reduction preprocessing to reduce the calculation cost and improve the calculation efficiency; when analyzing the correlation between the data group and the aerodynamic coefficient, it is divided into a linear prediction model and a nonlinear prediction model, so that the prediction can be flexibly applied to situations dominated by linear or nonlinear, thereby improving the accuracy of the aerodynamic coefficient prediction; when the constructed aerodynamic coefficient prediction model is actually applied, it only needs to obtain the state point data to output the predicted aerodynamic coefficient.

[0086] Electronic device embodiment

[0087] Figure 5 Schematic diagram of an electronic device according to an embodiment of the present invention. The electronic device 500 may include at least one processor 510 and a memory 520. The processor 510 may execute instructions stored in the memory 520. The processor 510 is connected to the memory 520 through a data bus. In addition to the memory 520, the processor 510 may also be connected to an input device 530, an output device 540, and a communication device 550 through a data bus.

[0088] The processor 510 may be any conventional processor, such as a commercially available CPU. The processor may also include a graphics processor (Graphic Process Unit, GPU), a field programmable gate array (Field Programmable Gate Array, FPGA), a system on chip (System on Chip, SOC), an application specific integrated circuit (Application Specific Integrated Circuit, ASIC) or a combination thereof.

[0089] The memory 520 may be implemented by any type of volatile or nonvolatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic memory, flash memory, magnetic disk or optical disk.

[0090] In the embodiment of the present disclosure, executable instructions are stored in the memory 520, and the processor 510 can read the executable instructions from the memory 520 and execute the instructions to implement all or part of the steps of the method for constructing the aerodynamic coefficient prediction model of any of the above exemplary embodiments.

[0091] Computer Readable Storage Medium Embodiments

[0092] In addition to the above-mentioned methods and devices, an exemplary embodiment of the present disclosure may also be a computer program product or a computer-readable storage medium storing the computer program product, wherein the computer product includes computer program instructions that can be executed by a processor to implement all or part of the steps described in the method for constructing an aerodynamic coefficient prediction model in any of the above-mentioned exemplary embodiments.

[0093] The computer program product may be written in any combination of one or more programming languages ​​to write program codes for performing the operations of the embodiments of the present application, including object-oriented programming languages ​​such as Java, C++, etc., and also conventional procedural programming languages ​​such as "C" language or similar programming languages ​​and scripting languages ​​(e.g., Python). The program code may be executed entirely on the user computing device, partially on the user computing device, as an independent software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server.

[0094] Computer readable storage media can adopt any combination of one or more readable media. The readable medium can be a readable signal medium or a readable storage medium. The readable storage medium can include, for example, but is not limited to, a system, device or device of electricity, magnetism, light, electromagnetic, infrared, or semiconductor, or any combination of the above. More specific examples of readable storage media include: a static random access memory (SRAM) with one or more wires electrically connected, an electrically erasable programmable read-only memory (EEPROM), an erasable programmable read-only memory (EPROM), a programmable read-only memory (PROM), a read-only memory (ROM), a magnetic memory, a flash memory, a magnetic disk or an optical disk, or any suitable combination of the above.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for constructing an aerodynamic coefficient prediction model, characterized in that: include: Acquire an aerodynamic training set and a corresponding flow field training set of an aircraft, and perform order reduction processing on the flow field training set to obtain reduced-order flow field features; Obtaining a flow field prediction model based on the aerodynamic training set and the reduced-order flow field feature training; The reduced-order flow field prediction value output by the flow field prediction model and the aerodynamic force training set are combined into a data group, and a linear prediction model and a nonlinear prediction model are obtained through training based on the data group; The outputs of the linear prediction model and the nonlinear prediction model are fused at the prediction layer to obtain a constructed aerodynamic coefficient prediction model.

2. The method according to claim 1, characterized in that The method further comprises: The aerodynamic force verification set including the state point data is input into the aerodynamic force coefficient prediction model to obtain the prediction result of the aerodynamic force coefficient.

3. The method according to claim 1, characterized in that The step of obtaining the aerodynamic training set of the aircraft and the corresponding flow field training set specifically includes: The aerodynamic training set and the corresponding flow field training set are obtained using the Latin hypercube method, wherein the aerodynamic training set includes state point data and aerodynamic coefficients.

4. The method according to claim 1, characterized in that: The reducing order processing of the flow field training set to obtain the reduced order flow field features specifically includes: The flow field training set is preprocessed by reducing the order using a variational autoencoder or an intrinsic orthogonal decomposition algorithm to obtain the reduced-order flow field features.

5. The method according to claim 1, characterized in that The flow field prediction model obtained by training based on the aerodynamic training set and the reduced-order flow field characteristics specifically includes: The state point data in the aerodynamic training set and the reduced-order flow field features form a data pair; Using the data point pairs to train a four-layer fully connected neural network, outputting a reduced-order flow field prediction value through the four-layer fully connected neural network, and using mean square error calculation to evaluate the loss between the reduced-order flow field prediction value and the reduced-order flow field feature; A flow field prediction model is obtained by iteratively adopting a gradient algorithm according to the loss, wherein the activation function of the flow field prediction model is a hyperbolic tangent function, and the output layer is a linear layer without an activation function.

6. The method according to claim 1, characterized in that The linear prediction model and the nonlinear prediction model obtained by training based on the data set specifically include: Using the data set to train a single-layer neural network, outputting a first predicted value of an aerodynamic coefficient through the single-layer neural network, and calculating a loss between the first predicted value and the aerodynamic coefficient in the data set using a mean square error; Iteratively obtaining a linear prediction model using a gradient algorithm according to the loss, wherein the linear prediction model does not use an activation function; Using the data set to train a multi-hidden layer fully connected neural network, outputting a second predicted value of the aerodynamic coefficient through the multi-hidden layer fully connected neural network, and calculating the loss between the second predicted value and the aerodynamic coefficient in the data set using a mean square error; The nonlinear prediction model is obtained by iteratively using the gradient algorithm according to the loss.

7. The method according to claim 1, characterized in that The fusing the outputs of the linear prediction model and the nonlinear prediction model at the prediction layer specifically includes: The outputs of the linear prediction model and the nonlinear prediction model are fused using a weighted summation method.

8. A device for constructing an aerodynamic coefficient prediction model, characterized in that: include: A training set processing module is used to obtain an aerodynamic training set of an aircraft and a corresponding flow field training set, and to reduce the order of the flow field training set to obtain reduced-order flow field features; A flow field prediction training module, used for obtaining a flow field prediction model based on the aerodynamic training set and the reduced-order flow field feature training; A dual-branch prediction module, used for combining the reduced-order flow field prediction value output by the flow field prediction model and the aerodynamic training set into a data group, and obtaining a linear prediction model and a nonlinear prediction model based on the training of the data group; The result fusion module is used to fuse the outputs of the linear prediction model and the nonlinear prediction model at the prediction layer to obtain a constructed aerodynamic coefficient prediction model.

9. An electronic device, characterized in that: include: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the computer program is executed by the processor, the steps of the method for constructing an aerodynamic coefficient prediction model as described in any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores an implementation program for information transmission, and when the program is executed by a processor, the steps of the method for constructing an aerodynamic coefficient prediction model as described in any one of claims 1 to 7 are implemented.

Images