An aircraft aerodynamic optimization method based on a mixed-precision surrogate model

Abstract

This invention belongs to the field of aircraft aerodynamic optimization and discloses an aircraft aerodynamic optimization method based on a hybrid precision surrogate model, comprising: constructing a geometric model of the aircraft and a sample point set; constructing a high-precision sample point set and a low-precision sample point set based on the sample point set and the aircraft's high-precision CFD model and low-precision CFD model, respectively, for training the low-precision surrogate model and the hybrid precision surrogate model; constructing individuals based on design parameters to form an initial population and iterating the population; in each generation of the population, replacing the cruise lift-to-drag ratio determined by the hybrid precision surrogate model with the cruise lift-to-drag ratio determined by the high-precision CFD model and the low-precision CFD model for some individuals, and updating the hybrid precision surrogate model; the design parameters contained in the optimal individuals output after stopping the iteration are the optimal aerodynamic optimization design scheme of the aircraft.

Description

Technical Field

[0001] This invention belongs to the field of aircraft aerodynamic optimization, specifically relating to an aircraft aerodynamic optimization method based on a hybrid precision surrogate model. Background Technology

[0002] Beginning in the 1990s, computational fluid dynamics (CFD) methods of varying precision emerged, and by the early 2000s, they had become the core computational method for aerodynamic design in the early preliminary design phase of aircraft. With the gradual improvement of these methods, the precision of CFD calculations has steadily increased, along with the number of computational grids and computation time, evolving from tens of thousands of grids to nearly ten million grids, and from early structured grids to hybrid grids of unstructured and structured grids. This aerodynamic computation method has significantly increased the computational cost of early preliminary design; when this method is integrated into existing optimization methods, the computational cost of aerodynamic optimization surges. The Vortex Lattice Method (VLM), developed from the lift line theory, has a simple calculation principle and low computational cost, making it the most cost-effective calculation method in the conceptual design stage. It can generally achieve a calculation speed of seconds in a computer, and its fundamental calculation error lies in the assumption of ignoring viscosity and only calculating induced drag, which makes the resulting systematic error predictable. In addition, the VLM method is still very reliable in calculating the lift coefficient trend corresponding to the geometric layout, which is sufficient to help the optimization model screen poor design areas and predict the angle of attack corresponding to the required lift coefficient in the early stage of optimization.

[0003] According to publicly available domestic literature, existing aerodynamic optimization design methods are mainly based on computational fluid dynamics. These optimization design methods have very high computational costs and involve a large number of unnecessary high-precision CFD calculations. There is a lack of aircraft optimization methods that use VLM methods and a combination of CFD calculations of different precision levels. Summary of the Invention

[0004] The purpose of this invention is to provide an aerodynamic optimization method for aircraft based on a hybrid precision surrogate model, so as to solve the problem of high computational overhead in existing optimization design methods.

[0005] To achieve the above objectives, the present invention employs the following technical solution:

[0006] An aerodynamic optimization method for aircraft based on a hybrid precision surrogate model includes:

[0007] A geometric model of the aircraft is constructed to determine the design parameters. Based on the design parameters, a sample point set is constructed, and a corresponding lift coefficient matrix is ​​constructed for each sample point. The design lift coefficient of the aircraft is obtained, and the cruise angle of attack corresponding to the design lift coefficient is determined for each sample point based on the lift coefficient matrix. The sample points in the sample point set are divided into high-precision sample points and low-precision sample points. The first cruise lift-to-drag ratio and the second cruise lift-to-drag ratio are calculated using the high-precision CFD model and the low-precision CFD model of the aircraft, respectively, in combination with the cruise angle of attack, thereby constructing the high-precision sample point set and the low-precision sample point set.

[0008] A low-precision proxy model is constructed and trained using a set of low-precision sample points. Based on the low-precision proxy model, a mixed-precision proxy model is constructed by combining a Gaussian process, and the mixed-precision proxy model is trained and optimized using a set of high-precision sample points. The output response of the mixed-precision proxy model is the third cruise lift-to-drag ratio.

[0009] Individuals are constructed based on design parameters to form an initial population, and the population is iterated. In each generation of the population, the third cruise lift-to-drag ratio corresponding to the individual is determined using a mixed-precision surrogate model and sorted. A preset number of individuals are selected from the sorted results, and the third cruise lift-to-drag ratio of some individuals is replaced with the corresponding first cruise lift-to-drag ratio, while the third cruise lift-to-drag ratio of the remaining individuals is replaced with the corresponding second cruise lift-to-drag ratio. The first cruise lift-to-drag ratio and the corresponding individual are used as new low-precision sample points to update the mixed-precision surrogate model. When the preset conditions are met, the iteration stops and the optimal individual is output. The design parameters contained in the optimal individual are the optimal aerodynamic optimization design scheme of the aircraft.

[0010] Furthermore, a corresponding value range is set for each design parameter, and the value range of all design parameters is sampled using the Latin hypercube sampling method. Each set of design parameters obtained from each sampling is taken as a sample point, thus forming a sample point set.

[0011] Based on the aircraft's geometric model, the lift coefficient corresponding to each angle of attack within the preset angle of attack range is calculated using the vortex lattice method for each sample point, and a lift coefficient matrix is ​​constructed.

[0012] Furthermore, based on the lift coefficient matrix, for each sample point in the sample point set, the correlation coefficient between its multiple angles of attack within the preset angle of attack range and the corresponding lift coefficient is calculated; if the correlation coefficient is greater than the preset value, the linearization theoretical criterion of the sample point is set to 1; otherwise, the sample point is discarded.

[0013] By using the slope and intercept of the fitted lines for all lift coefficients corresponding to the sample points, the cruise angle of attack corresponding to the design lift coefficient can be solved.

[0014] Furthermore, the sample points in the sample point set are divided into two parts by random partitioning, with one part serving as high-precision sample points and the other part serving as low-precision sample points.

[0015] High-precision CFD models and low-precision CFD models were constructed for the aircraft; the high-precision CFD model used a higher total number of computational grids than the low-precision CFD model.

[0016] For each high-precision sample point, determine whether the linearization theoretical criterion of the high-precision sample point is 1; if so, use the high-precision CFD model to calculate the first cruise lift-to-drag ratio of the high-precision sample point based on the cruise angle of attack.

[0017] For each low-precision sample point, determine whether the linearization criterion of the low-precision sample point is 1; if so, use the low-precision CFD model to calculate the second cruise lift-to-drag ratio of the low-precision sample point based on the cruise angle of attack.

[0018] Construct a high-precision sample point set, which includes high-precision sample points as input data and a first cruise lift-to-drag ratio as output response; construct a low-precision sample point set, which includes low-precision sample points as input data and a second cruise lift-to-drag ratio as output response.

[0019] Furthermore, the low-precision surrogate model adopts the Kriging model; using the low-precision surrogate model as the trend term, a mixed-precision surrogate model is constructed by combining it with a Gaussian process; the mixed-precision surrogate model is expressed as follows:

[0020] ;

[0021] in, It is the scaling factor; It is a Gaussian process with zero mean; The output response of the hybrid precision surrogate model is the predicted third cruise lift-to-drag ratio. For the Kriging model, input data The prediction process;

[0022] The maximum likelihood estimation method is employed, combined with a high-precision sample point set to optimize the mixed-precision surrogate model; the scaling factor is adjusted. This ensures that the confidence level of the trained mixed-precision surrogate model meets the design requirements.

[0023] Furthermore, during the population iteration process, for each generation of the population, the following operations are performed:

[0024] The preset number of calculations is M; each individual is used as input data, and the corresponding third cruise lift-to-drag ratio is obtained using the current mixed precision surrogate model; the individuals are sorted in descending order of the third cruise lift-to-drag ratio, and the top M individuals are selected;

[0025] The first individual ranked among the first M individuals is used to calculate its first cruise lift-to-drag ratio using the high-precision CFD model, and the first cruise lift-to-drag ratio is used to replace the third cruise lift-to-drag ratio of the individual. The remaining M-1 individuals are used to calculate their second cruise lift-to-drag ratio using the low-precision CFD model, and the second cruise lift-to-drag ratio is used to replace the third cruise lift-to-drag ratio of the individual. Then, crossover and mutation operations are performed on the population.

[0026] Furthermore, among the top M individuals selected from each generation of the population, the top-ranked individual and its first cruise lift-to-drag ratio calculated using a high-precision CFD model are added to the low-precision sample point set as a new low-precision sample point. The low-precision surrogate model is then retrained using the updated low-precision sample point set, and the mixed-precision surrogate model is updated. The next generation of the population then uses the updated mixed-precision surrogate model to calculate the third cruise lift-to-drag ratio.

[0027] Furthermore, iteration stops when one of the following preset conditions is met:

[0028] Condition 1: The maximum number of iterations reaches the preset number of iterations;

[0029] Condition 2: In k consecutive iterations, the change in the first cruise lift-to-drag ratio calculated by the high-precision CFD model in each generation of the population compared to the previous generation is less than 1%, and the error accuracy of the high-precision CFD model and the hybrid precision surrogate model meets the preset requirements, as shown below:

[0030] ;

[0031] Where k ranges from 2 to 10; The lift coefficient;

[0032] After the iteration stops, the individual corresponding to the maximum value of the third cruise lift-to-drag ratio predicted by the mixed precision surrogate model in each generation of the population during the historical iteration process is output as the optimal individual.

[0033] A terminal device includes a processor, a memory, and a computer program stored in the memory; when the processor executes the computer program, it implements the aircraft aerodynamic optimization method based on a hybrid precision proxy model.

[0034] A computer-readable storage medium storing a computer program; when executed by a processor, the computer program implements the aircraft aerodynamic optimization method based on a hybrid precision proxy model.

[0035] Compared with the prior art, the present invention has the following technical features:

[0036] This invention ensures high accuracy in the final calculation results while reducing the proportion of high-precision CFD calculations and the computational overhead of the overall optimization cycle, thereby effectively improving the overall computational efficiency of the aircraft design optimization process. Attached Figure Description

[0037] Figure 1 This is a schematic flowchart of the method of the present invention;

[0038] Figure 2 This is a schematic diagram of the geometric model of the aircraft in an embodiment of the present invention;

[0039] Figure 3 This is a schematic diagram of the computational grid division for a high-precision CFD model in an embodiment of the present invention;

[0040] Figure 4 This is a schematic diagram of the computational grid division for a low-precision CFD model in an embodiment of the present invention. Detailed Implementation

[0041] This invention provides an aerodynamic optimization method for aircraft based on a mixed-precision surrogate model. Based on the aircraft's geometric model, a computational model with mixed high and low precision is established. A genetic algorithm assisted by the mixed-precision surrogate model is used for optimization. During the iterative solution process, high-precision and low-precision CFD calculations are used to update the optimization process, achieving rapid optimization of the aircraft's aerodynamic performance. See also... Figure 1 The specific steps of the method of the present invention are as follows:

[0042] Step 1: Construct the geometric model of the aircraft and determine the design parameters; construct a sample point set based on the design parameters, and construct a corresponding lift coefficient matrix for each sample point; obtain the design lift coefficient of the aircraft, and determine the cruise angle of attack corresponding to the design lift coefficient for each sample point based on the lift coefficient matrix; divide the sample points in the sample point set into high-precision sample points and low-precision sample points, and calculate the first cruise lift-to-drag ratio and the second cruise lift-to-drag ratio using the high-precision CFD model and the low-precision CFD model of the aircraft, respectively, in combination with the cruise angle of attack, thereby constructing the high-precision sample point set and the low-precision sample point set.

[0043] Step 1.1: Construct the geometric model of the aircraft in CAD software, ensuring a complete and smooth shape while minimizing the presence of any sharp triangular points that would make mesh generation difficult. Based on the geometric model, determine the design parameters; the dimensions of these design parameters are... Design parameters can be, for example, various geometric parameters of the aircraft; such as... Figure 2In this embodiment, the design parameters used are leading edge sweep angle, aspect ratio, and wingtip twist angle; the aircraft in this embodiment is a combination of wing and fuselage.

[0044] Step 1.2: Set a corresponding value range for each design parameter. Sample the value ranges of all design parameters using the Latin hypercube sampling method. Each set of design parameters obtained from each sampling is considered a sample point, resulting in a total of N sample points, forming a sample point set. In this embodiment, N depends on the dimension m of the design parameter, and its calculation formula is: .

[0045] Step 1.3: Based on the geometric model of the aircraft, the lift coefficient corresponding to each angle of attack within the preset angle of attack range is calculated using the vortex lattice method for each sample point, and the lift coefficient matrix is ​​constructed.

[0046] In this embodiment, the preset angle of attack range is -2° to +4°, with an interval step of 1°, so a total of 7 angles of attack are taken; the lift coefficient corresponding to each angle of attack can be obtained by the vortex lattice method and stored in the lift coefficient matrix.

[0047] Step 1.4: Obtain the design lift coefficient of the aircraft. The lift coefficient of this design The values ​​are given during the preliminary design phase of the aircraft and range from 0 to 1; for each sample point, the design lift coefficient is determined based on the aforementioned lift coefficient matrix. The corresponding cruise angle of attack is as follows:

[0048] (1) Based on the lift coefficient matrix, for the sample points in the sample point set The lift coefficients corresponding to the seven angles of attack within the preset angle of attack range of -2° to +4° are used to calculate the correlation coefficient with the seven angles of attack. If the correlation coefficient (for example, the Pearson correlation coefficient can be used) is greater than the preset value of 0.99, then the sample points are set. Linearization Theory Criterion Otherwise, the sample points will be... The discard and linearization criterion is set as follows .

[0049] (2) Solve for the design lift coefficient Corresponding cruise angle of attack :

[0050] ;

[0051] in, , Sample points The slope and intercept of the fitted lines for all corresponding lift coefficients.

[0052] The cruise angle of attack for each sample point can be obtained using the above method. and linearization theory criteria .

[0053] Step 1.5: Set a calculation ratio η, with a range of 0 < η < 1; in this example, η = 50%; use the calculation ratio η to divide the N sample points in the sample point set into two parts, one part as high-precision sample points and the other part as low-precision sample points.

[0054] Step 1.6: Construct both a high-precision CFD model and a low-precision CFD model for the aircraft. The difference is that the high-precision CFD model uses a higher total number of computational grids. In this embodiment, the high-precision CFD model uses 5 million computational grids, and the computational physics model uses the SST k-ω model, calculating the convergence residual as follows: The calculation count is 1000 steps. See [link / reference] Figure 3 The low-precision CFD model uses a total of 800,000 computational grids, and the computational convergence residuals are relaxed to [a certain value]. The calculation count is 500 steps. See below. Figure 4 .

[0055] Step 1.7, for each high-precision sample point Determine the high-precision sample point Linearization Theory Criterion Is it 1? If so, then use a high-precision CFD model to calculate high-precision sample points based on the cruise angle of attack. First cruise lift-to-drag ratio That is, the ratio of the lift coefficient to the drag coefficient.

[0056] For each low-precision sample point Determine the low-precision sample point Linearization Theory Criterion Is it 1? If yes, then use a low-precision CFD model to calculate low-precision sample points based on the cruise angle of attack. Second cruise lift-to-drag ratio .

[0057] Step 1.8: Construct a high-precision sample point set, which includes the high-precision sample points used as input data. And the first cruise lift-to-drag ratio as the output response Construct a low-precision sample point set, which includes the low-precision sample points used as input data. And the second cruise lift-to-drag ratio as the output response .

[0058] Step 2: Construct a low-precision surrogate model and train it using a set of low-precision sample points; based on the low-precision surrogate model, construct a mixed-precision surrogate model (MPSM) by combining a Gaussian process, and train and optimize the mixed-precision surrogate model using a set of high-precision sample points; the output response of the mixed-precision surrogate model is the third cruise lift-to-drag ratio.

[0059] Step 2.1: Train a low-precision surrogate model using a low-precision sample point set; the low-precision surrogate model adopts the Kriging model, as shown below:

[0060] ;

[0061] in, This indicates that the Kriging model responds to the input data. The prediction process, This is the output response of the Kriging model.

[0062] Step 2.2: Using the low-precision surrogate model as the trend term, construct a mixed-precision surrogate model by combining it with a Gaussian process; train the mixed-precision surrogate model using a high-precision sample point set; the mixed-precision surrogate model is represented as follows:

[0063] ;

[0064] in, It is a scaling factor; It is a Gaussian process with zero mean, used to capture the deviation between high-precision sample points and low-precision surrogate models; The output response of the hybrid precision surrogate model is the predicted third cruise lift-to-drag ratio.

[0065] Step 2.3: The maximum likelihood estimation (MLE) method is used, combined with a high-precision sample point set to optimize the mixed-precision surrogate model; the scaling factor is adjusted. Ultimately, the confidence level of the trained mixed-precision surrogate model reached 0.99, meeting the design requirements.

[0066] Step 3: Construct individuals based on design parameters to form an initial population, and iterate the population. In each generation of the population, use a mixed-precision surrogate model to determine the third cruise lift-to-drag ratio corresponding to the individual and sort them. Select a preset number of individuals from the sorted results, replace the third cruise lift-to-drag ratio of some individuals with the corresponding first cruise lift-to-drag ratio, and replace the third cruise lift-to-drag ratio of the remaining individuals with the corresponding second cruise lift-to-drag ratio. Use the first cruise lift-to-drag ratio and the corresponding individual as new low-precision sample points to update the mixed-precision surrogate model. Stop the iteration when the preset conditions are met and output the optimal individual. The design parameters contained in the optimal individual are the optimal aerodynamic optimization design scheme of the aircraft.

[0067] Step 3.1: Set the corresponding value range for each design parameter, and re-perform Latin hypercube sampling. Each set of design parameters obtained is used as an individual in the initial population of the genetic algorithm; the total number of individuals is... Individual requirements must be met. That is, the individual The linearization criterion needs to be determined to be 1 using the method in step 1.5.

[0068] Step 3.2, during the population iteration process, for each generation of the population, perform the following operations:

[0069] The preset number of calculations is M; in this example, M = m + 1; each individual is used as input data, and the corresponding output response, i.e., the predicted third cruise lift-to-drag ratio, is obtained using the current hybrid precision surrogate model. According to the third cruise lift-to-drag ratio The individuals are sorted in descending order, and the top M individuals are selected. The first individual among the top M individuals is then used to calculate its first cruise lift-to-drag ratio using the high-precision CFD model. and utilize Replace the individual corresponding The remaining M-1 individuals had their second cruise lift-to-drag ratio calculated using the aforementioned low-precision CFD model. and utilize Replace the individual corresponding Then, crossover and mutation operations are performed on the population.

[0070] Step 3.3: Sort the population of each generation and select the top M individuals. Then, select the first individual in the sorted population and calculate the first cruise lift-to-drag ratio of that individual using a high-precision CFD model. As a new low-precision sample point, it is added to the low-precision sample point set. Using the updated low-precision sample point set, the low-precision surrogate model is retrained in step 2, and the mixed-precision surrogate model is updated. The next generation population then uses the updated mixed-precision surrogate model. The calculation.

[0071] Step 3.4: Stop the iteration when one of the following preset conditions is met:

[0072] Condition 1: The maximum number of iterations reaches the preset number of iterations; in this example, the preset number of iterations is 50, which can be adjusted according to actual needs and computing power limitations.

[0073] Condition 2: In each of the k consecutive iterations, the first cruise lift-to-drag ratio calculated by the high-precision CFD model in each generation of the population. The change compared to the previous generation is less than 1%, and the error accuracy of both the high-precision CFD model and the mixed-precision surrogate model meets the preset requirements, as shown below:

[0074] ;

[0075] After the iteration stops, the third cruise lift-to-drag ratio predicted by the mixed-precision surrogate model in each generation of the population during the historical iteration process will be used. The individual corresponding to the maximum value is output as the optimal individual; the design parameters contained in the optimal individual are the optimal aerodynamic optimization design scheme of the aircraft; in the example, the value of k is 5, and its value range can generally be set from 2 to 10.

[0076] The optimal individual in the embodiment includes the following design parameters: leading edge sweep angle of 54°, aspect ratio of 3.1, and wingtip twist angle of 4.9°; the corresponding third cruise lift-to-drag ratio is 12.1.

[0077] Compared with optimization methods that rely entirely on high-precision CFD aerodynamic simulation, this invention reduces the number of CFD aerodynamic simulation calls (i.e., the number of calls to the high-precision CFD model and the low-precision CFD model in this invention) by approximately 37%, and 85% of the high-precision CFD aerodynamic simulations are replaced with low-precision CFD aerodynamic simulations, significantly improving optimization efficiency.

[0078] The above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of this application, and should all be included within the protection scope of this application.

Claims

1. A hybrid-precision proxy model based aerodynamic optimization method for an aircraft, characterized in that, include: Construct the geometric model of the aircraft and determine the design parameters; A set of sample points is constructed based on the design parameters, and a corresponding lift coefficient matrix is ​​constructed for each sample point; Obtain the design lift coefficient of the aircraft, and determine the cruise angle of attack corresponding to the design lift coefficient for each sample point based on the lift coefficient matrix; The sample points in the sample point set are divided into high-precision sample points and low-precision sample points. The first cruise lift-to-drag ratio and the second cruise lift-to-drag ratio are calculated using the high-precision CFD model and the low-precision CFD model of the aircraft, respectively, in combination with the cruise angle of attack, thereby constructing the high-precision sample point set and the low-precision sample point set. Based on the geometric model of the aircraft, the lift coefficient corresponding to each angle of attack in the preset angle of attack range is calculated for each sample point using the vortex lattice method, and the lift coefficient matrix is ​​constructed. A low-precision surrogate model is constructed and trained using a set of low-precision sample points. Based on this low-precision surrogate model, a mixed-precision surrogate model is constructed using a Gaussian process, and then trained and optimized using a set of high-precision sample points. The output response of the mixed-precision surrogate model is the third cruise lift-to-drag ratio. The low-precision surrogate model employs the Kriging model. Using the low-precision surrogate model as a trend term, a mixed-precision surrogate model is constructed using a Gaussian process. The mixed-precision surrogate model is represented as follows: ； in, It is the scaling factor; It is a Gaussian process with zero mean; The output response of the hybrid precision surrogate model is the predicted third cruise lift-to-drag ratio. For the Kriging model, input data The prediction process; The maximum likelihood estimation method is employed, combined with a high-precision sample point set to optimize the mixed-precision surrogate model; the scaling factor is adjusted. This ensures that the confidence level of the trained mixed-precision surrogate model meets the design requirements. Individuals are constructed based on design parameters to form an initial population, and the population is iterated. In each generation of the population, the third cruise lift-to-drag ratio corresponding to the individual is determined and sorted using a mixed-precision surrogate model. A preset number of individuals are selected from the sorted results, and the third cruise lift-to-drag ratio of some individuals is replaced with the corresponding first cruise lift-to-drag ratio, while the third cruise lift-to-drag ratio of the remaining individuals is replaced with the corresponding second cruise lift-to-drag ratio. The first cruise lift-to-drag ratio and the corresponding individual are used as new low-precision sample points to update the mixed-precision surrogate model. When the preset conditions are met, the iteration stops and the optimal individual is output. The design parameters contained in the optimal individual are the optimal aerodynamic optimization design scheme of the aircraft.

2. The aerodynamic optimization method for aircraft based on a hybrid precision surrogate model according to claim 1, characterized in that, Each design parameter is assigned a corresponding value range. The value range of all design parameters is sampled using the Latin hypercube sampling method. Each set of design parameters obtained from each sampling is taken as a sample point, thus forming a sample point set.

3. The aerodynamic optimization method for aircraft based on a hybrid precision surrogate model according to claim 1, characterized in that, Based on the lift coefficient matrix, for each sample point in the sample point set, the correlation coefficient between its multiple angles of attack within the preset angle of attack range and the corresponding lift coefficient is calculated; if the correlation coefficient is greater than the preset value, the linearization theoretical criterion of the sample point is set to 1; otherwise, the sample point is discarded. By using the slope and intercept of the fitted lines for all lift coefficients corresponding to the sample points, the cruise angle of attack corresponding to the design lift coefficient can be solved.

4. The aerodynamic optimization method for aircraft based on a hybrid precision surrogate model according to claim 3, characterized in that, The sample points in the sample point set are divided into two parts by random partitioning, one part is used as high-precision sample points and the other part is used as low-precision sample points. High-precision CFD models and low-precision CFD models were constructed for the aircraft respectively. The high-precision CFD model uses a higher total number of computational grids than the low-precision CFD model. For each high-precision sample point, determine whether the linearization theoretical criterion for the high-precision sample point is 1; if so, calculate the first cruise lift-to-drag ratio of the high-precision sample point using the high-precision CFD model. For each low-precision sample point, determine whether the linearization criterion for the low-precision sample point is 1; if so, calculate the second cruise lift-to-drag ratio of the low-precision sample point using the low-precision CFD model. Construct a high-precision sample point set, which includes high-precision sample points as input data and a first cruise lift-to-drag ratio as output response; construct a low-precision sample point set, which includes low-precision sample points as input data and a second cruise lift-to-drag ratio as output response.

5. The aerodynamic optimization method for aircraft based on a hybrid precision surrogate model according to claim 1, characterized in that, During population iteration, for each generation of the population, the following operations are performed: The preset number of calculations is M; each individual is used as input data, and the corresponding third cruise lift-to-drag ratio is obtained using the current mixed precision surrogate model; the individuals are sorted in descending order of the third cruise lift-to-drag ratio, and the top M individuals are selected; The first individual ranked among the first M individuals is used to calculate its first cruise lift-to-drag ratio using the high-precision CFD model, and the first cruise lift-to-drag ratio is used to replace the third cruise lift-to-drag ratio of the individual. The remaining M-1 individuals are used to calculate their second cruise lift-to-drag ratio using the low-precision CFD model, and the second cruise lift-to-drag ratio is used to replace the third cruise lift-to-drag ratio of the individual. Then, crossover and mutation operations are performed on the population.

6. The aerodynamic optimization method for aircraft based on a hybrid precision surrogate model according to claim 1, characterized in that, In each generation of the population, the top M individuals are sorted, and the first individual in the sorted population, along with the first cruise lift-to-drag ratio calculated using a high-precision CFD model, are added to the low-precision sample point set as a new low-precision sample point. The low-precision surrogate model is then retrained using the updated low-precision sample point set, and the mixed-precision surrogate model is updated. The next generation of the population then uses the updated mixed-precision surrogate model to calculate the third cruise lift-to-drag ratio.

7. The aerodynamic optimization method for aircraft based on a hybrid precision surrogate model according to claim 1, characterized in that, The iteration stops when one of the following preset conditions is met: Condition 1: The maximum number of iterations reaches the preset number of iterations; Condition 2: In k consecutive iterations, the change in the first cruise lift-to-drag ratio calculated by the high-precision CFD model in each generation of the population compared to the previous generation is less than 1%, and the error accuracy of the high-precision CFD model and the hybrid precision surrogate model meets the preset requirements, as shown below: ； Where k ranges from 2 to 10; After the iteration stops, the individual corresponding to the maximum value of the third cruise lift-to-drag ratio predicted by the mixed precision surrogate model in each generation of the population during the historical iteration process is output as the optimal individual.

8. A terminal device, comprising a processor, a memory, and a computer program stored in the memory; characterized in that, When the processor executes the computer program, it implements the aircraft aerodynamic optimization method based on a hybrid precision proxy model as described in any one of claims 1-7.

9. A computer-readable storage medium storing a computer program; characterized in that, When the computer program is executed by the processor, it implements the aircraft aerodynamic optimization method based on a hybrid precision proxy model as described in any one of claims 1-7.

Images