A method, system, electronic device and storage medium for predicting aerodynamic characteristics of airfoil
Through the multi-task multi-fidelity model, the problem of low efficiency and insufficient accuracy of traditional aerodynamic appearance optimization design is solved, and efficient aerodynamic characteristic prediction is achieved to meet the needs of aircraft refined design.
Patent Information
- Application Number
- CN202310606863.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2023-05-26
- Publication Date
- 2024-12-03
- Estimated Expiration
- 2043-05-26
AI Technical Summary
Traditional aerodynamic appearance optimization design relies on rules of thumb, with a long design cycle and it is difficult to obtain the optimal appearance under multiple constraints and multi-design conditions. The high-fidelity simulation calculation efficiency based on CFD is low. Although XFOIL software accelerates the optimization process, the accuracy is insufficient. Meet engineering needs.
Using multi-task multi-fidelity model, including the first and second multi-task expert hybrid model, is constructed through multi-task learning methods and gating mechanisms, and aerodynamic characteristics prediction is achieved through potential variables and flight conditions of the generated adversarial network to achieve low-income protection. The conversion of real data to high-fidelity data.
It significantly improves the accuracy and speed of aerodynamic characteristic data, shortens the design cycle, and can obtain high-precision aerodynamic characteristic data within a few seconds, meeting the needs of aircraft refined design.
Smart Images

Figure CN116628854B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of aerodynamic shape optimization design, and in particular to a method, system, electronic equipment and storage medium for predicting aerodynamic characteristics of an airfoil. Background Art
[0002] Aerodynamic shape optimization design is a discipline that studies how to combine modern CFD technology with numerical optimization algorithms to automatically find the optimal shape that meets performance requirements through computers. The traditional trial and error method relies on theoretical derivation and experimental verification, and optimizes aerodynamic shape by constantly adjusting the shape to determine whether it meets the optimization goal. This method heavily relies on the designer's experience, the design cycle is very long, and usually the optimal shape under multiple constraints and multiple design states cannot be obtained.
[0003] With the continuous development of aircraft technology and computer technology, CFD technology has made great progress in theories such as computational accuracy and grid generation. High-fidelity simulation based on computational fluid dynamics (CFD) has gradually become the main method for evaluating aerodynamic performance. However, aerodynamic shape optimization based on CFD technology still takes a long time. One analysis may take hours or even days. Thousands of CFD calls will lead to "computational disasters". In order to improve optimization efficiency and reduce design cycles, many practitioners use XFOIL software instead of CFD simulation for aerodynamic optimization, which can complete a simulation in a few seconds. It does speed up the optimization process, but with the increasing requirements for refined aerodynamic shape optimization design of aircraft, the accuracy calculated by XFOIL often does not meet the requirements of engineering design. Summary of the invention
[0004] The purpose of the present invention is to provide a method, system, electronic device and storage medium for predicting the aerodynamic characteristics of an airfoil, which can convert low-precision data into high-precision data and improve the accuracy and speed of data acquisition.
[0005] To achieve the above object, the present invention provides the following solutions:
[0006] A method for predicting aerodynamic characteristics of an airfoil, comprising:
[0007] Obtaining parametric variables and flight conditions of the target aircraft airfoil shape;
[0008] A multi-task multi-fidelity model is used to predict the aerodynamic characteristics of the parameterized variables and the flight conditions, and determine the high-fidelity aerodynamic characteristics data of the target aircraft airfoil shape; the multi-task multi-fidelity model includes a first multi-task expert hybrid model and a second multi-task expert hybrid model; the first multi-task expert hybrid model and the second multi-task expert hybrid model are both constructed according to a multi-task learning method and a gating mechanism; the first multi-task expert hybrid model is used to determine the low-fidelity aerodynamic characteristics data of the target aircraft airfoil shape; the second multi-task expert hybrid model is used to convert the low-fidelity aerodynamic characteristics data into high-fidelity aerodynamic characteristics data.
[0009] Optionally, the flight conditions are operating conditions of the aircraft during flight; the operating conditions include angle of attack, Mach number and Reynolds number.
[0010] Optionally, the method for determining the parameterized variables of the target aircraft airfoil shape is: determining the latent variables of the generative adversarial network of the target aircraft airfoil shape as parameterized variables.
[0011] Optionally, the using of a multi-task multi-fidelity model to predict aerodynamic characteristics of the parameterized variables and the flight conditions to determine high-fidelity aerodynamic characteristic data of the airfoil shape of the target aircraft specifically includes:
[0012] Inputting the parameterized variables and the flight conditions into the first multi-task expert hybrid model to determine low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape;
[0013] The low-fidelity aerodynamic characteristic data is input into the second multi-task expert hybrid model, and the low-fidelity aerodynamic characteristic data is transformed to determine the high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape.
[0014] Optionally, the method for determining the multi-task multi-fidelity model is:
[0015] Acquiring training data; the training data includes training parameter variables, flight training conditions and corresponding high-fidelity aerodynamic characteristics prediction results;
[0016] The training data is input into the first multi-task expert mixture model, and the low-fidelity aerodynamic characteristic data output by the first multi-task expert mixture model is used as the input of the second multi-task expert mixture model. The back-propagation method is used to train the model with the goal of minimizing the loss between the high-fidelity aerodynamic characteristic data output by the second multi-task expert mixture model and the high-fidelity aerodynamic characteristic prediction result, so as to obtain the multi-task multi-fidelity model.
[0017] The present invention also provides an airfoil aerodynamic characteristics prediction system, comprising:
[0018] A data acquisition module, used to obtain parameterized variables and flight conditions of the target aircraft airfoil shape;
[0019] A characteristic prediction module is used to use a multi-task multi-fidelity model to predict the aerodynamic characteristics of the parameterized variables and the flight conditions, and determine the high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the multi-task multi-fidelity model includes a first multi-task expert hybrid model and a second multi-task expert hybrid model; the first multi-task expert hybrid model and the second multi-task expert hybrid model are both constructed according to a multi-task learning method and a gating mechanism; the first multi-task expert hybrid model is used to determine the low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the second multi-task expert hybrid model is used to convert the low-fidelity aerodynamic characteristic data into high-fidelity aerodynamic characteristic data.
[0020] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned airfoil aerodynamic characteristics prediction method.
[0021] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method for predicting aerodynamic characteristics of an airfoil as described above.
[0022] According to the specific embodiments provided by the present invention, the present invention discloses the following technical effects:
[0023] The present invention discloses a method, system, electronic device and storage medium for predicting aerodynamic characteristics of an airfoil. The method comprises using a multi-task multi-fidelity model to predict aerodynamic characteristics of parameterized variables and flight conditions of an acquired target aircraft airfoil shape, and determining high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape, wherein the multi-task multi-fidelity model comprises a first multi-task expert hybrid model and a second multi-task expert hybrid model, wherein the first multi-task expert hybrid model is used to determine low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape, and the second multi-task expert hybrid model is used to convert the low-fidelity aerodynamic characteristic data into high-fidelity aerodynamic characteristic data, and the two hybrid models can convert low-precision data into high-precision data, thereby improving data acquisition accuracy and speed. BRIEF DESCRIPTION OF THE DRAWINGS
[0024] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative labor.
[0025] Figure 1 It is a schematic flow chart of the method for predicting aerodynamic characteristics of an airfoil of the present invention;
[0026] Figure 2 This is a schematic diagram of the results of the XFOIL solver in this embodiment;
[0027] Figure 3 This is a diagram showing the meshing effect required to obtain a high-fidelity simulation in this embodiment;
[0028] Figure 4 A comparison diagram of the pressure distribution of XFOIL calculation, high-fidelity simulation and wind tunnel test under the same working conditions in this embodiment;
[0029] Figure 5 Schematic diagram of the MMOE network structure in this embodiment;
[0030] Figure 6 Schematic diagram of the training process of MMOE-MF in this embodiment;
[0031] Figure 7 The predicted effect diagram of lift, drag and pitching moment predicted by MMOE-MF under subsonic conditions in this embodiment; wherein (a) is C l Prediction effect diagram; (b) is C d Prediction effect diagram; (c) is C m Prediction effect diagram;
[0032] Figure 8 : is the prediction effect diagram of lift, drag and pitching moment predicted by MMOE-MF under transonic working condition in this embodiment; wherein (a) is C l Prediction effect diagram; (b) is C d Prediction effect diagram; (c) is C m Prediction effect diagram. DETAILED DESCRIPTION
[0033] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0034] The purpose of the present invention is to provide a method, system, electronic device and storage medium for predicting the aerodynamic characteristics of an airfoil, which can convert low-precision data into high-precision data and improve the accuracy and speed of data acquisition.
[0035] In order to make the above-mentioned objects, features and advantages of the present invention more obvious and easy to understand, the present invention is further described in detail below with reference to the accompanying drawings and specific embodiments.
[0036] like Figure 1 As shown, the present invention provides a method for predicting aerodynamic characteristics of an airfoil, comprising:
[0037] Step 100: Obtain parameterized variables and flight conditions of the target aircraft airfoil shape. The parameterized variables of the target aircraft airfoil shape are determined by: determining the latent variables of the generative adversarial network of the target aircraft airfoil shape as parameterized variables. The flight conditions are the operating conditions of the aircraft during flight; the operating conditions include angle of attack, Mach number and Reynolds number.
[0038] Step 200: Utilize a multi-task multi-fidelity model to predict aerodynamic characteristics of the parameterized variables and the flight conditions, and determine high-fidelity aerodynamic characteristics data of the target aircraft airfoil shape; the multi-task multi-fidelity model includes a first multi-task expert hybrid model and a second multi-task expert hybrid model; the first multi-task expert hybrid model and the second multi-task expert hybrid model are both constructed according to a multi-task learning method and a gating mechanism; the first multi-task expert hybrid model is used to determine low-fidelity aerodynamic characteristics data of the target aircraft airfoil shape; the second multi-task expert hybrid model is used to convert the low-fidelity aerodynamic characteristics data into high-fidelity aerodynamic characteristics data.
[0039] As a specific implementation of step 200, it includes:
[0040] The parameterized variables and the flight conditions are input into the first multi-task expert hybrid model to determine low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the low-fidelity aerodynamic characteristic data is input into the second multi-task expert hybrid model to transform the low-fidelity aerodynamic characteristic data to determine high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape.
[0041] The method for determining the multi-task multi-fidelity model is as follows:
[0042] Acquire training data; the training data includes training parameter variables, flight training conditions and corresponding high-fidelity aerodynamic characteristic prediction results; input the training data into the first multi-task expert hybrid model, use the low-fidelity aerodynamic characteristic data output by the first multi-task expert hybrid model as the input of the second multi-task expert hybrid model, and use the back propagation method to train with the goal of minimizing the loss between the high-fidelity aerodynamic characteristic data output by the second multi-task expert hybrid model and the high-fidelity aerodynamic characteristic prediction results, so as to obtain the multi-task multi-fidelity model.
[0043] On the basis of the above technical solution, the following embodiments are also provided.
[0044] In this embodiment, the method for converting low-precision aerodynamic characteristics of an airfoil into high-precision mainly includes airfoil shape parameterized variables, flight conditions, a low-fidelity aerodynamic database, a high-fidelity aerodynamic database, construction of an MMOE-MF multi-fidelity model, and a result output layer.
[0045] Among them, airfoil shape parameterization variables: in aerodynamic shape optimization design, airfoil shape parameterization is the most important content of aerodynamic shape optimization design, which determines the quality and efficiency of optimization. The present invention uses advanced generative adversarial network intelligent parameterization technology, and uses the potential variables of the generative adversarial network as parameterization variables.
[0046] Flight conditions: the operating conditions of the aircraft during flight, such as angle of attack, Mach number, Reynolds number, etc.
[0047] Low-fidelity aerodynamic database: After the airfoil shape is obtained, it is input into the XFOIL solver, and the corresponding flight conditions are input to obtain low-fidelity aerodynamic characteristics including lift, drag, pitch moment, etc. The low-fidelity aerodynamic database is composed of the airfoil shape parameterized variables, the flight conditions, and the low-fidelity aerodynamic characteristics.
[0048] High-fidelity aerodynamic database: After obtaining the airfoil shape, high-fidelity aerodynamic characteristic data is obtained through high-fidelity simulation. First, the shape is input into Pointwise software for meshing, and then the generated mesh is input into Fluent software for high-fidelity simulation to obtain high-fidelity aerodynamic characteristic data. Similarly, the airfoil shape parameterized variables are integrated with flight conditions, and high-fidelity aerodynamic characteristics are added to form a high-fidelity aerodynamic database.
[0049] Construct a multi-task multi-fidelity model (MMOE-MF): Couple two multi-task expert hybrid models to establish a multi-fidelity proxy model to predict the multi-fidelity aerodynamic characteristics of the airfoil. The first multi-task expert hybrid model is used to predict low-precision aerodynamic data, and then integrate the features of high-fidelity aerodynamic data into the second multi-task expert hybrid model. The second multi-task expert hybrid model is used to explore the relationship between low precision and high precision. Once the model is trained, we can obtain high-fidelity aerodynamic data within a few seconds by integrating flight conditions only through the airfoil shape parameterized variables.
[0050] Output of results: For each shape, predict its high-fidelity aerodynamic properties.
[0051] Based on each component, the specific application steps are as follows:
[0052] Step 1: By inputting the aerodynamic shape into the XFOIL solver, a large amount of cheap low-fidelity aerodynamic data is obtained by setting flight conditions;
[0053] Step 2: Get a small amount of expensive high-fidelity aerodynamic characteristics by meshing the shape and high-fidelity simulation;
[0054] Step 3: Use the low-fidelity dataset to train the first MMOE model to predict low-fidelity aerodynamic characteristics;
[0055] Step 4: Use the features of the high-fidelity dataset to fuse the low-fidelity aerodynamic characteristics to train a second MMOE model to predict high-fidelity aerodynamic characteristics data;
[0056] Step 5: Once the multi-fidelity proxy model is trained, we can input the high-fidelity aerodynamic characteristics obtained by the MMOE-MF model into the high-fidelity database.
[0057] Among them, the MMOE-MF model is constructed according to the following formula:
[0058] y k =h k (f k (x)
[0059]
[0060] g k (x) = softmax(W x)
[0061] Among them, k is the number of tasks, E i (x) is the i-th expert network, For the kth gating network of the i-th expert, softmax is used to activate the input features, f k (x) is the weighted sum of all experts, y k is the output result for each task. x is the input data, which integrates the airfoil shape parameterization variables and flight conditions. For different tasks, the model selection is different. The output of a specific Gate K represents the probability of different experts being selected. The weighted sum of multiple experts is obtained to obtain f k (x) and outputs it to a specific Tower model, which finally outputs the result.
[0062] The MMOE-MF model is used to convert low-precision airfoil aerodynamic data into high-precision data. After the MMOE-MF model is constructed, the method further comprises: optimizing and updating the parameters of the MMOE-MF model by a back-propagation method.
[0063] The parameters of the MMOE-MF model are optimized and updated by the following formula:
[0064]
[0065] Among them, n is the number of training samples, y pred is the model prediction value, y true is the true value of the training sample.
[0066] Therefore, outputting high-precision data requires the following steps:
[0067] Step 1: Input the aerodynamic shape parameterization variables and flight conditions into the MMOE-MF model and use the first MMOE model to predict the low-fidelity aerodynamic characteristics;
[0068] Step 2: Fusion of high-fidelity features and flight conditions with low-fidelity aerodynamic characteristics to obtain a merged layer;
[0069] Step 3: Input the merged layer into the MMOE-MF model and use the second MMOE model to obtain high-precision aerodynamic characteristics.
[0070] As another embodiment, in the aerodynamic shape optimization design, obtaining aerodynamic characteristic data is a very important step. Figure 2 shows the aerodynamic characteristics calculated using a low-fidelity solver, Figure 3 A hybrid grid is shown to accelerate convergence. In order to verify the independence of the grid, as shown in Table 1, this embodiment selects grid amounts of 14577 (M0), 43741 (M1), and 83871 (M2) for verification, and the results are shown in Table 1, where represents the Mach number, represents the Reynolds number, and represents the angle of attack.
[0071] Table 1: Under subsonic conditions, M = 0.45, Re = 6.0 × 10 6 , AOA=1.5901
[0072] Grid volume <![CDATA[C l (10 -3 )]]> <![CDATA[C d (10 -4 )]]> <![CDATA[C m (10 -4 )]]> M0 14539 820.02 193.14 -979.11 M1 34359 812.84 186.83 -956.48 M2 83907 826.09 191.30 -984.43
[0073] In order to verify the accuracy of the calculation in this embodiment, the calculation results are compared with the results of the wind tunnel test (such as Figure 4 The specific comparison results are shown in Table 2, where the aerodynamic coefficient results of the wind tunnel test are: C l =0.802, C d =0.0175, C m =-0.1.
[0074] Table 2 Comparison of aerodynamic coefficients and calculation time of high and low fidelity RAE2822 airfoil
[0075] Solver <![CDATA[C l ]]> <![CDATA[ΔC l ]]> <![CDATA[C d ]]> <![CDATA[ΔC d ]]> <![CDATA[C m ]]> <![CDATA[ΔC m ]]> CPU time XFOIL 0.7381 0.064 0.0083 0.0092 -0.0740 0.026 0.26s CFD 0.8200 0.018 0.0193 0.0018 -0.0979 0.002 112s
[0076] Considering the need to predict multiple aerodynamic characteristics such as lift, drag, and pitching moment, this embodiment is modeled based on a multi-task learning method, dividing the shared layer into multiple experts, and introducing a gating mechanism to set an independent gating weight for each task, and then weighted sum the contribution of each expert. Traditional multi-task learning requires strong correlation between tasks to obtain better prediction accuracy.
[0077] This embodiment better solves the limitation of strong task correlation and improves the prediction accuracy of tasks by fully considering the correlation between tasks. This embodiment uses two MMOE models (such as Figure 5 A multi-task multi-fidelity agent model (MMOE-MF) is constructed (as shown in Figure 1), and a prediction method for low-precision aerodynamic characteristics of aircraft is proposed to convert them into high-precision. In the MMOE-MF model, the input is a merging layer, including airfoil shape parameterized variables and flight conditions. The output is three prediction tasks, including lift, drag, and pitch moment.
[0078] like Figure 4 As shown, the MMOE-MF model is constructed according to the following formula.
[0079] y k =h k (f k (x)
[0080]
[0081] g k (x) = softmax(W x)
[0082] Among them, k is the number of tasks, E i (x) is the i-th expert network, For the kth gating network of the i-th expert, soffmax is used to activate the input features, f k (x) is the weighted sum of all experts, y k is the output result for each task. x is the input data, which integrates the airfoil shape parameterization variables and flight conditions. For different tasks, the model selection is different. The output of a specific Gate K represents the probability of different experts being selected. The weighted sum of multiple experts is obtained to obtain f k (x) and outputs it to a specific Tower model, which finally outputs the result.
[0083] Finally, the MMOE-MF model is used to convert the low-precision aerodynamic data of the airfoil into high-precision data.
[0084] like Figure 6As shown, in some embodiments, the parameters of the model are optimized and updated by back propagation, and the model is trained, wherein the loss function is shown in the following formula:
[0085]
[0086] Among them, n is the number of training samples, y pred is the model prediction value, y true is the true value of the training sample. For example, the specific parameter settings of the MMOE model are shown in Table 3
[0087] Table 3 Specific parameter settings of MMOE-MF model
[0088] Network Layer Neurons quantity Activation Function Input 25 1 - Experts 25×16 10 Relu Tasks 25×10 3 Softmax Towers 512×16 3 Relu Output 1×3 3 Linear
[0089] like Figure 7-Figure 8 As shown, in some implementations, separate models are constructed for subsonic and transonic speeds, respectively, and the MMOE-MF model obtains prediction results through the following steps:
[0090] Step 1: Input the aerodynamic shape parameterization variables and flight conditions into the MMOE-MF model and use the first MMOE model to predict the low-fidelity aerodynamic characteristics;
[0091] Step 2: Fusion of high-fidelity features and flight conditions with low-fidelity aerodynamic characteristics to obtain a merged layer;
[0092] Step 3: Input the merged layer into the MMOE-MF model and use the second MMOE model to obtain high-precision aerodynamic characteristics.
[0093] According to the following embodiments, the method of this embodiment is compared with multi-task learning and multi-fidelity in the prior art to verify the effectiveness and advancement of the method provided by the embodiment of the present invention.
[0094] Table 4 Model comparison
[0095]
[0096] This embodiment has the following beneficial effects:
[0097] According to the aircraft aerodynamic shape optimization method, device and medium based on multi-fidelity simulation of each scheme of the present invention, a multi-fidelity proxy model is established by constructing two multi-task expert hybrid models to predict the aerodynamic characteristics of the airfoil. The first multi-task expert hybrid model is used to predict low-precision aerodynamic data, which fully explores the relationship between low-precision aerodynamic data. The second multi-task expert hybrid model is used to explore the relationship between low precision and high precision. The two prediction models are combined to realize the conversion of low-fidelity aerodynamic characteristics into high-fidelity aerodynamic characteristics within a Mach number range of [0.3-0.74] within a few seconds. It greatly saves time costs and computing resources, and provides a more efficient method for aerodynamic shape optimization design.
[0098] In addition, the present invention also provides an airfoil aerodynamic characteristics prediction system, comprising:
[0099] A data acquisition module, used to obtain parameterized variables and flight conditions of the target aircraft airfoil shape;
[0100] A characteristic prediction module is used to use a multi-task multi-fidelity model to predict the aerodynamic characteristics of the parameterized variables and the flight conditions, and determine the high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the multi-task multi-fidelity model includes a first multi-task expert hybrid model and a second multi-task expert hybrid model; the first multi-task expert hybrid model and the second multi-task expert hybrid model are both constructed according to a multi-task learning method and a gating mechanism; the first multi-task expert hybrid model is used to determine the low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the second multi-task expert hybrid model is used to convert the low-fidelity aerodynamic characteristic data into high-fidelity aerodynamic characteristic data.
[0101] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the above-mentioned airfoil aerodynamic characteristics prediction method.
[0102] The present invention also provides a computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the method for predicting aerodynamic characteristics of an airfoil as described above.
[0103] The various embodiments in this specification are described in a progressive manner, and each embodiment focuses on the differences from other embodiments. The same or similar parts between the various embodiments can be referenced to each other.
[0104] This article uses specific examples to illustrate the principles and implementation methods of the present invention. The above examples are only used to help understand the core idea of the present invention. At the same time, for those skilled in the art, according to the idea of the present invention, there will be changes in the specific implementation methods and application scope. In summary, the content of this specification should not be understood as limiting the present invention.
Claims
1. A method for predicting aerodynamic characteristics of an airfoil, characterized in that: include: Obtaining parametric variables and flight conditions of the target aircraft airfoil shape; A multi-task multi-fidelity model is used to predict the aerodynamic characteristics of the parameterized variables and the flight conditions, and determine the high-fidelity aerodynamic characteristics data of the target aircraft airfoil shape; the multi-task multi-fidelity model includes a first multi-task expert hybrid model and a second multi-task expert hybrid model; the first multi-task expert hybrid model and the second multi-task expert hybrid model are both constructed according to a multi-task learning method and a gating mechanism; the first multi-task expert hybrid model is used to determine the low-fidelity aerodynamic characteristics data of the target aircraft airfoil shape; the second multi-task expert hybrid model is used to convert the low-fidelity aerodynamic characteristics data into high-fidelity aerodynamic characteristics data; The flight conditions are the operating conditions of the aircraft during flight; the operating conditions include angle of attack, Mach number and Reynolds number; The method for determining the parameterized variables of the target aircraft airfoil shape is as follows: determining the latent variables of the generative adversarial network of the target aircraft airfoil shape as parameterized variables; The method of using a multi-task multi-fidelity model to predict aerodynamic characteristics of the parameterized variables and the flight conditions and determining high-fidelity aerodynamic characteristic data of the airfoil shape of the target aircraft specifically includes: Inputting the parameterized variables and the flight conditions into the first multi-task expert hybrid model to determine low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; Inputting the low-fidelity aerodynamic characteristic data into the second multi-task expert hybrid model, transforming the low-fidelity aerodynamic characteristic data, and determining the high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; specifically: fusing the high-fidelity features and the flight conditions to fuse the low-fidelity aerodynamic characteristics to obtain a merged layer, and inputting the merged layer into the second multi-task expert hybrid model to obtain high-precision aerodynamic characteristics; The method for determining the multi-task multi-fidelity model is: Acquiring training data; the training data includes training parameter variables, flight training conditions and corresponding high-fidelity aerodynamic characteristics prediction results; The training data is input into the first multi-task expert mixture model, and the low-fidelity aerodynamic characteristic data output by the first multi-task expert mixture model is used as the input of the second multi-task expert mixture model. The back-propagation method is used to train the model with the goal of minimizing the loss between the high-fidelity aerodynamic characteristic data output by the second multi-task expert mixture model and the high-fidelity aerodynamic characteristic prediction result, so as to obtain the multi-task multi-fidelity model.
2. A system for predicting airfoil aerodynamic characteristics, characterized in that: include: A data acquisition module, used to obtain parameterized variables and flight conditions of the target aircraft airfoil shape; A characteristic prediction module, used for predicting aerodynamic characteristics of the parameterized variables and the flight conditions by using a multi-task multi-fidelity model, and determining high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the multi-task multi-fidelity model includes a first multi-task expert hybrid model and a second multi-task expert hybrid model; the first multi-task expert hybrid model and the second multi-task expert hybrid model are both constructed according to a multi-task learning method and a gating mechanism; the first multi-task expert hybrid model is used for determining low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; the second multi-task expert hybrid model is used for converting the low-fidelity aerodynamic characteristic data into high-fidelity aerodynamic characteristic data; The flight conditions are the operating conditions of the aircraft during flight; the operating conditions include angle of attack, Mach number and Reynolds number; The method for determining the parameterized variables of the target aircraft airfoil shape is as follows: determining the latent variables of the generative adversarial network of the target aircraft airfoil shape as parameterized variables; The method of using a multi-task multi-fidelity model to predict aerodynamic characteristics of the parameterized variables and the flight conditions and determining high-fidelity aerodynamic characteristic data of the airfoil shape of the target aircraft specifically includes: Inputting the parameterized variables and the flight conditions into the first multi-task expert hybrid model to determine low-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; The low-fidelity aerodynamic characteristic data is input into the second multi-task expert hybrid model, and the low-fidelity aerodynamic characteristic data is transformed to determine the high-fidelity aerodynamic characteristic data of the target aircraft airfoil shape; specifically: the high-fidelity features and flight conditions are integrated to integrate the low-fidelity aerodynamic characteristics to obtain a merged layer, and the merged layer is input into the second multi-task expert hybrid model to obtain high-precision aerodynamic characteristics.
3. An electronic device, characterized in that: The electronic device comprises a memory and a processor, wherein the memory is used to store a computer program, and the processor runs the computer program to enable the electronic device to execute the method for predicting aerodynamic characteristics of an airfoil according to claim 1.
4. A computer-readable storage medium, characterized in that: The computer program is stored therein, and when the computer program is executed by a processor, the method for predicting the aerodynamic characteristics of an airfoil as claimed in claim 1 is implemented.
