Aerodynamic force / torque coefficient small sample modeling method based on sign regression
Through the symbol regression method, the correlation parameter expression of aerodynamic/moment coefficient is constructed, the constant term is optimized, and the polynomial fit of aerodynamic/moment coefficient is realized, which solves the problem of high cost of aerodynamic data acquisition and incomplete high dimensions, and reduces the aerodynamic The number of modeled samples improves modeling efficiency and extrapolation capabilities.
Patent Information
- Application Number
- CN202411185588.2
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-08-27
- Publication Date
- 2024-12-03
AI Technical Summary
The existing aerodynamic data acquisition methods have large samples demands and high cost. The high-dimensional incompleteness makes it difficult to achieve small sample modeling for aerodynamic prediction. The existing black box and white box modeling methods have their own limitations, making it difficult to effectively reduce pneumatic construction. number of samples.
Using a small sample modeling method of aerodynamic/moment coefficient based on symbol regression, the aerodynamic data of the first aircraft is collected, and the associated parameter expression is constructed, and the optimized expression is used for polynomial fitting to achieve the aerodynamic/moment coefficient. Polynomial expression, reducing the number of samples required for new profile aerodynamic modeling.
Through nonlinear dimensionality reduction and optimization of associated parameters, effective modeling of high-dimensional state space is achieved, reducing the number of samples required for aerodynamic modeling, reducing costs, and improving extrapolation capabilities and interpretability.
Smart Images

Figure CN119066870A_ABST
Abstract
Description
Technical Field
[0001] The invention belongs to the technical field of aerodynamic modeling in aerodynamics, and in particular relates to a small sample modeling method of aerodynamic force / torque coefficients based on symbolic regression. Background Art
[0002] With the rapid development of aerospace technology, obtaining the aerodynamic characteristics of aircraft has become the key to advanced aircraft design.
[0003] The methods for obtaining aerodynamic characteristics mainly include numerical simulation, wind tunnel test and flight test. When a large number of aerodynamic samples are needed, all three methods will face the problem of high cost. Although the numerical simulation method can realize multi-state and multi-condition parallel calculation, it has high requirements for computer hardware and consumes a lot of computing resources; the method of measuring aerodynamic force through wind tunnel test has high precision and high confidence, but the wind tunnel test has high requirements for wind tunnel facilities, and conducting wind tunnel tests with high Mach numbers or high Reynolds numbers is costly and difficult; flight tests have the highest credibility, but flight tests require the construction of ready-made prototypes, and the risk is relatively high. Its cost is the highest among the three methods. The acquisition of aerodynamic data faces the problem of large sample requirements and high costs. The high cost of money and time is in conflict with the need for efficient design.
[0004] In addition, the high-dimensional incompleteness of aerodynamic data exacerbates this contradiction. The flight parameter space is a high-dimensional parameter space, which contains multi-dimensional state variables such as Mach number, Reynolds number, angle of attack, sideslip angle, and rudder angle. The high-dimensional characteristics lead to a sharp increase in the aerodynamic samples required to cover the flight envelope; and the flight test data is sparsely sampled, and the state in the flight test cannot completely cover the complete flight envelope. This incompleteness often requires data extrapolation when using existing aerodynamic data for aerodynamic prediction.
[0005] Data-driven modeling methods are divided into black-box modeling methods and white-box modeling methods. At present, the existing aerodynamic data-driven methods still find it difficult to achieve small sample modeling with a sample size of less than 10; among them, black-box data-driven modeling mainly establishes the mapping between state quantities and aerodynamic / torque coefficients based on neural networks, and then combines algorithms such as transfer learning to improve the model prediction accuracy, which can achieve the purpose of efficient and fast reasoning to obtain aerodynamic coefficients. However, the generalization and extrapolation capabilities of black-box models are poor, and the extrapolation prediction accuracy of small sample data is very low. The sample size required for accurate modeling is often less than 10. 2The existing white-box modeling methods such as symbolic regression and sparse recognition can construct algebraic expressions of aerodynamic force / torque coefficients, have strong generalization and interpretability, and strong extrapolation ability. The sample size required for modeling is usually smaller than that of the black-box model, at the order of tens. In complex modeling problems, the model parameters that need to be calibrated are often large, and it is also difficult to achieve aerodynamic modeling of small samples. As shown in Table 1, the existing aerodynamic modeling methods and their advantages and disadvantages are given.
[0006] Table 1 Data-driven modeling methods
[0007] Summary of the invention
[0008] In view of the above-mentioned deficiencies in the prior art, the small sample modeling method of aerodynamic force / torque coefficient based on symbolic regression provided by the present invention solves the problems of large demand for aerodynamic data samples, high acquisition cost, and high-dimensional incompleteness in existing aerodynamic acquisition methods.
[0009] In order to achieve the above-mentioned invention object, the technical solution adopted by the present invention is: a small sample modeling method of aerodynamic force / torque coefficient based on symbolic regression, comprising the following steps:
[0010] S1. Collecting aerodynamic data of the first aircraft under various flow conditions; the aerodynamic data includes flow state variables and aerodynamic force / torque coefficients;
[0011] S2. Based on the flow state variables, the associated parameter expressions of the aerodynamic force / torque coefficients of the aircraft are constructed by the symbolic regression method;
[0012] S3, collecting aerodynamic data of a small sample for the second aircraft, and optimizing the constant term in the expression of the associated parameter based on the aerodynamic data;
[0013] S4. Use the optimized associated parameter expression to perform polynomial fitting on the small sample aerodynamic data to obtain the polynomial expression of the corresponding aerodynamic force / torque coefficient, and complete the small sample modeling of the aerodynamic force / torque coefficient.
[0014] Furthermore, the step S1 is specifically as follows:
[0015] Numerical simulation or experiment is performed on the first aircraft under various flow conditions to collect corresponding aerodynamic data; wherein the flow state variables include the incoming flow Mach number Ma, the Reynolds number Re, the angle of attack α, the sideslip angle β, the rudder deflection angle δ x ,δ y ,δ z ; The aerodynamic force / torque coefficient C is obtained based on numerical simulation or experiment.
[0016] Furthermore, the definition of the associated parameters in step S2 is:
[0017] For a high-dimensional function C=f(x1,x2,...), if C=g(s(x1,x2,...)), then s(x1,x2,...) is called the associated parameter of C, that is, the aerodynamic force / torque coefficient is a composite function of its corresponding associated parameter, and a smooth curve with a unified trend can be drawn with the associated parameter s as the horizontal coordinate and the associated force / torque coefficient C as the vertical coordinate.
[0018] Furthermore, the step S2 is specifically as follows:
[0019] S21, the flow state variables (Ma, Re, α, β, δ x ,δ y ,δ z ,...) as the input variable of symbolic regression, and the aerodynamic force / moment coefficient C as the correlation target, and then use symbolic regression to generate the candidate expression of the correlation parameter s=s(Ma,Re,α,β,δ x ,δ y ,δ z ,...,c1,c2,...); where (c1,c2,...) represents the constant term in the associated parameter expression;
[0020] S22, using the polynomial fitting of the selected expression s to fit the aerodynamic force / torque coefficient C, to obtain the polynomial fitting value C of the aerodynamic force / torque coefficient of the associated quantity polyfit (s,s 2 ,...);
[0021] S23, taking the mean square error between the fitted value of the aerodynamic force / torque coefficient C and the true value of the aerodynamic force / torque coefficient C as the loss function as the fitness function in the genetic algorithm, evaluating the fitness of each candidate expression, and taking each candidate expression as an individual to perform inheritance, mutation, crossover and recombination operations, and updating the individual position;
[0022] S24, repeating steps S21 to S23 until the end condition set by the genetic algorithm is met, and selecting the associated parameter expression of the aerodynamic force / torque coefficient based on the fitness value of each candidate expression.
[0023] Furthermore, in step S23, the loss function loss is expressed as:
[0024]
[0025] In the formula, C i It represents the true value of the aerodynamic force / torque coefficient of the i-th aerodynamic sample. The subscript i represents the aerodynamic sample index, and n represents the total number of aerodynamic samples.
[0026] Furthermore, in step S3, the method for collecting small sample pneumatic data is specifically as follows:
[0027] Collect three or more sample points randomly selected in the flight envelope of the second aircraft, perform numerical calculations or wind tunnel tests, and collect corresponding aerodynamic data.
[0028] Furthermore, in step S3, a gradient optimization algorithm or a non-gradient optimization algorithm is used to optimize the constant term in the associated parameter expression so as to minimize the mean square error of the associated parameter fitting to the aerodynamic force / torque coefficient.
[0029] The beneficial effects of the present invention are:
[0030] (1) The method of the present invention is used to extract correlation parameters, which can combine high-dimensional flight state variables into low-dimensional correlation parameters through a regression method, and can express the associated aerodynamic force / torque coefficients into a polynomial form of correlation parameters, thereby achieving nonlinear dimensionality reduction of the high-dimensional state space.
[0031] (2) As the associated parameters have good generalization properties for variable shapes and states and good extrapolation capabilities, migrating the associated parameter formula to the new aerodynamic shape and associating the force coefficients of the new shape can effectively reduce the number of samples required for aerodynamic modeling of the new aerodynamic shape. Aerodynamic modeling of the new shape can be achieved with at least three samples, thereby reducing costs and increasing efficiency. Compared with traditional methods, the number of samples required for modeling can be reduced by one or more orders of magnitude. BRIEF DESCRIPTION OF THE DRAWINGS
[0032] Figure 1 A flow chart of a small sample modeling method of aerodynamic force / torque coefficients based on symbolic regression provided by the present invention.
[0033] Figure 2 The axial force coefficient C of HB-2 before association provided by the present invention is A With the distribution of Ma and α.
[0034] Figure 3 The axial force correlation parameter s of HB-2 provided by the present invention is A Axial force coefficient C A The associated effect.
[0035] Figure 4 The axial force coefficient C established by the HBS profile small sample data provided by the present invention is A Polynomial model.
[0036] Figure 5 The HBS axial force coefficient C provided by the present invention is A The prediction effect of the polynomial model on other aerodynamic conditions. DETAILED DESCRIPTION
[0037] The specific implementation modes of the present invention are described below so that those skilled in the art can understand the present invention. However, it should be clear that the present invention is not limited to the scope of the specific implementation modes. For those of ordinary skill in the art, as long as various changes are within the spirit and scope of the present invention as defined and determined by the attached claims, these changes are obvious, and all inventions and creations utilizing the concept of the present invention are protected.
[0038] Embodiment 1:
[0039] The embodiment of the present invention provides a small sample modeling method of aerodynamic force / torque coefficient based on symbolic regression, such as Figure 1 As shown, the following steps are included:
[0040] S1. Collecting aerodynamic data of the first aircraft under various flow conditions; the aerodynamic data includes flow state variables and aerodynamic force / torque coefficients;
[0041] S2. Based on the flow state variables, the associated parameter expressions of the aerodynamic force / torque coefficients of the aircraft are constructed by the symbolic regression method;
[0042] S3, collecting aerodynamic data of a small sample for the second aircraft, and optimizing the constant term in the expression of the associated parameter based on the aerodynamic data;
[0043] S4. Use the optimized associated parameter expression to perform polynomial fitting on the small sample aerodynamic data to obtain the polynomial expression of the corresponding aerodynamic force / torque coefficient, and complete the small sample modeling of the aerodynamic force / torque coefficient.
[0044] Step S1 of the embodiment of the present invention is specifically as follows:
[0045] Numerical simulation or experiment is performed on the first aircraft under various flow conditions to collect corresponding aerodynamic data; wherein the flow state variables include the incoming flow Mach number Ma, the Reynolds number Re, the angle of attack α, the sideslip angle β, the rudder deflection angle δ x ,δ y ,δ z ; The aerodynamic force / torque coefficient C is obtained based on numerical simulation or experiment.
[0046] In step S2 of the embodiment of the present invention, the association parameter is defined as:
[0047] For a high-dimensional function C=f(x1,x2,...), if C=g(s(x1,x2,...)), then s(x1,x2,...) is called the associated parameter of C, that is, the aerodynamic force / torque coefficient is a composite function of its corresponding associated parameter, and a smooth curve with a unified trend can be drawn with the associated parameter s as the horizontal coordinate and the associated force / torque coefficient C as the vertical coordinate.
[0048] Step S2 of the embodiment of the present invention is specifically as follows:
[0049] S21, the flow state variables (Ma, Re, α, β, δ x ,δ y ,δ z ,...) as the input variable of symbolic regression, and the aerodynamic force / moment coefficient C as the correlation target, and then use symbolic regression to generate the candidate expression of the correlation parameter s=s(Ma,Re,α,β,δ x ,δ y ,δ z ,...,c1,c2,...); where (c1,c2,...) represents the constant term in the associated parameter expression;
[0050] S22, using the polynomial fitting of the selected expression s to fit the aerodynamic force / torque coefficient C, to obtain the polynomial fitting value C of the aerodynamic force / torque coefficient of the associated quantity polyfit (s,s 2 ,...);
[0051] S23, taking the mean square error between the fitted value of the aerodynamic force / torque coefficient C and the true value of the aerodynamic force / torque coefficient C as the loss function as the fitness function in the genetic algorithm, evaluating the fitness of each candidate expression, and taking each candidate expression as an individual to perform inheritance, mutation, crossover and recombination operations, and updating the individual position;
[0052] S24, repeating steps S21 to S23 until the end condition set by the genetic algorithm is met, and selecting the associated parameter expression of the aerodynamic force / torque coefficient based on the fitness value of each candidate expression.
[0053] In step S23 of this embodiment, the expression of the loss function loss is:
[0054]
[0055] In the formula, C i It represents the true value of the aerodynamic force / torque coefficient of the i-th aerodynamic sample. The subscript i represents the aerodynamic sample index, and n represents the total number of aerodynamic samples.
[0056] In step S3 of this embodiment, the method for collecting small sample pneumatic data is specifically as follows:
[0057] Collect 3 or more sample points in the flight envelope of the second aircraft, perform numerical calculations or wind tunnel tests, and collect corresponding aerodynamic data; the flow state of the sample points includes the incoming flow Mach number Ma, Reynolds number Re, angle of attack α, sideslip angle β, rudder deflection angle δ x ,δ y ,δz , and the aerodynamic force / moment coefficient C measured by numerical simulation or wind tunnel test.
[0058] In step S3 of the embodiment of the present invention, a gradient optimization algorithm or a gradient-free optimization algorithm is used to optimize the constant terms in the correlation parameter expression, so that the mean square error of the fitting of the correlation parameters to the aerodynamic force / moment coefficient is minimized.
[0059] Specifically, according to the small sample data, the constant terms c1, c2,... in the correlation parameter s = s(Ma, Re, α, β, δ x , δ y , δ z ,..., c1, c2,...) are optimized. The optimization objective is to minimize the mean square error of the fitting of the correlation parameters to the aerodynamic force / moment coefficient, and the optimization objective is
[0060] Optionally, the gradient optimization algorithm in this embodiment is the BFGS algorithm, and the gradient-free optimization algorithm is the Nelder-Mead algorithm.
[0061] In step S4 of the embodiment of the present invention, the correlation parameter expression s' = s(Ma, Re, α, β, δ x , δ y , δ z ,..., c1', c2',...) after optimizing the constant terms is used to perform polynomial fitting on the aerodynamic force / moment coefficient of the small sample data, so as to obtain the polynomial expression C = C(s', s' 2 ,...) of the force / moment coefficient corresponding to the second aircraft with the correlation parameters.
[0062] Embodiment 2:
[0063] This embodiment provides a specific example based on the method described in Embodiment 1.
[0064] In this embodiment, the correlation parameters of the axial force coefficient of the HB-2 aircraft shape are extracted, and the correlation parameters are migrated to the HBS shape and applied to the small sample modeling of the axial force coefficient of the HBS shape. The specific method is as follows:
[0065] S1: The SST turbulence model is used to numerically calculate the aerodynamic force of the HB-2 shape. The calculation ranges of each state variable are: the Mach number range is 1.5 < Ma < 11, the angle of attack range is -25° < α < 25°, the Reynolds number range is 10 4 < Re < 3×10 7 , and the sideslip angle and rudder deflection angle are both 0. The axial force coefficient C of HB-2 A The distribution law with Ma and α is as Figure 2 shown.
[0066] S2: Extraction of axial force coefficient C based on symbolic regression method A The associated parameter S A , using the extracted correlation parameter S A =f(Ma,Re,α,β,δ x ,δ y ,δ z ,...,c1,c2,...) for the axial force coefficient C A The association effect is as follows Figure 3 shown.
[0067] S3: Select three different state points (Ma, Re, α, β, δ x ,δ y ,δ z ), the aerodynamic force of the HBS shape is numerically calculated using the SST turbulence model, and the axial force coefficient C corresponding to the three sample points is obtained. A .
[0068] S4: Using these three samples, the correlation parameter S A =f(Ma,Re,α,β,δ x ,δ y ,δ z ,...,c1,c2,...) are optimized. The optimization goal is to minimize the mean square error of the associated parameters for the force coefficient fitting. The optimization goal is The optimization algorithm uses the Nelder-Mead algorithm. The correlation parameter expression S after optimizing the constant term is A '=f(Ma,Re,α,β,δ x ,δ y ,δ z ,...,c1',c2',...), perform polynomial fitting on the force coefficient of the small sample data to obtain the corresponding expression of the axial force coefficient. The polynomial model of the axial force coefficient established by the small sample data of HBS shape is as follows: Figure 4 As shown in Figure 2, the model predicts the axial force coefficient of other aerodynamic samples of HBS. Figure 5 shown.
[0069] The HBS axial force coefficient polynomial model established using three samples has a relative error of 3.35% in predicting the axial force coefficients of the remaining 85 different state points, achieving a high accuracy.
[0070] Based on the correlation parameters extracted in the above embodiments, and using the correlation parameters for small sample modeling, not only can nonlinear dimensionality reduction of high-dimensional state space be achieved, but also a new aerodynamic model of the aircraft shape can be accurately established through a small sample, greatly reducing the number of samples required for aerodynamic modeling. As a white box model, the correlation parameters also have good interpretability and extrapolation capabilities.
[0071] In the description of the present invention, it is necessary to understand that the orientation or positional relationship indicated by the terms "center", "thickness", "upper", "lower", "horizontal", "top", "bottom", "inner", "outer", "radial", etc. is based on the orientation or positional relationship shown in the drawings, and is only for the convenience of describing the present invention and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present invention. In addition, the terms "first", "second", and "third" are used only for descriptive purposes, and cannot be understood as indicating or implying the relative importance or the number of implicitly specified technical features. Therefore, the features defined by "first", "second", and "third" may explicitly or implicitly include one or more of the features.
[0072] The present invention uses specific embodiments to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the method of the present invention and its core idea. 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 a limitation on the present invention.
[0073] Those skilled in the art will appreciate that the embodiments described herein are intended to help readers understand the principles of the present invention, and should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those skilled in the art can make various other specific variations and combinations that do not deviate from the essence of the present invention based on the technical revelations disclosed by the present invention, and these variations and combinations are still within the protection scope of the present invention.
Claims
1. A small sample modeling method for aerodynamic force / torque coefficients based on symbolic regression, It is characterized in that The following steps are involved: S1. Collecting aerodynamic data of the first aircraft under various flow conditions; the aerodynamic data includes flow state variables and aerodynamic force / torque coefficients; S2. Based on the flow state variables, the associated parameter expressions of the aerodynamic force / torque coefficients of the aircraft are constructed by the symbolic regression method; S3, collecting aerodynamic data of a small sample for the second aircraft, and optimizing the constant term in the expression of the associated parameter based on the aerodynamic data; S4. Use the optimized associated parameter expression to perform polynomial fitting on the small sample aerodynamic data to obtain the polynomial expression of the corresponding aerodynamic force / torque coefficient, and complete the small sample modeling of the aerodynamic force / torque coefficient.
2. The aerodynamic force / torque coefficient small sample modeling method based on symbolic regression according to claim 1, It is characterized in that The step S1 is specifically as follows: Numerical simulation or experiment is performed on the first aircraft under various flow conditions to collect corresponding aerodynamic data; wherein the flow state variables include the incoming flow Mach number Ma, the Reynolds number Re, the angle of attack α, the sideslip angle β, the rudder deflection angle δ x ,δ y ,δ z ; The aerodynamic force / torque coefficient C is obtained based on numerical simulation or experiment.
3. The aerodynamic force / torque coefficient small sample modeling method based on symbolic regression according to claim 1, It is characterized in that The definition of the associated parameters in step S2 is: For a high-dimensional function C = f(x 1 ,x 2 ,...), if C=g(s(x 1 ,x 2 ,...)), then s(x 1 ,x 2 ,...) is called the associated parameter of C, that is, the aerodynamic force / torque coefficient is a composite function of its corresponding associated parameter. With the associated parameter s as the horizontal coordinate and the associated force / torque coefficient C as the vertical coordinate, it can be drawn into a smooth curve with a unified trend.
4. The aerodynamic force / torque coefficient small sample modeling method based on symbolic regression according to claim 2, It is characterized in that The step S2 is specifically as follows: S21, the flow state variables (Ma, Re, α, β, δ x ,δ y ,δ z ,...) as the input variable of symbolic regression, and the aerodynamic force / moment coefficient C as the correlation target, and then use symbolic regression to generate the candidate expression of the correlation parameter s=s(Ma,Re,α,β,δ x ,δ y ,δ z ,...,c 1 ,c 2 ,...); where (c 1 ,c 2 ,...) represents the constant term in the associated parameter expression; S22, using the polynomial fitting of the selected expression s to fit the aerodynamic force / torque coefficient C, to obtain the polynomial fitting value C of the aerodynamic force / torque coefficient of the associated quantity polyfit (s,s 2 ,...); S23, taking the mean square error between the fitted value of the aerodynamic force / torque coefficient C and the true value of the aerodynamic force / torque coefficient C as the loss function as the fitness function in the genetic algorithm, evaluating the fitness of each candidate expression, and taking each candidate expression as an individual to perform inheritance, mutation, crossover and recombination operations, and updating the individual position; S24, repeating steps S21 to S23 until the end condition set by the genetic algorithm is met, and selecting the associated parameter expression of the aerodynamic force / torque coefficient based on the fitness value of each candidate expression.
5. The small sample modeling method of aerodynamic force / torque coefficient based on symbolic regression according to claim 4, It is characterized in that In step S23, the expression of the loss function loss is: In the formula, C i It represents the true value of the aerodynamic force / torque coefficient of the i-th aerodynamic sample. The subscript i represents the aerodynamic sample index, and n represents the total number of aerodynamic samples.
6. The aerodynamic force / torque coefficient small sample modeling method based on symbolic regression according to claim 2, It is characterized in that In step S3, the method for collecting small sample pneumatic data is specifically as follows: Collect three or more sample points randomly selected in the flight envelope of the second aircraft, perform numerical calculations or wind tunnel tests, and collect corresponding aerodynamic data.
7. The small sample modeling method of aerodynamic force / torque coefficient based on symbolic regression according to claim 4, It is characterized in that In step S3, a gradient optimization algorithm or a non-gradient optimization algorithm is used to optimize the constant term in the expression of the associated parameters so as to minimize the mean square error of the associated parameters fitting the aerodynamic force / moment coefficient.
Citation Information
Cited By
Small sample migration modeling optimization method and system based on comparative learning
CN121525535A
Small sample transfer modeling optimization method and system based on contrastive learning
CN121525535B
Symbol regression-based high-speed aerodynamic derivative analysis modeling method and system
CN122021722A
A high-speed aerodynamic derivative analytical modeling method and system based on symbolic regression
CN122021722B
Aerodynamic modeling transfer learning method based on sign regression
CN122087963A
