A machine learning optimization design method for wing lift-to-drag ratio based on information fusion

Abstract

The present invention discloses a wing lift-to-drag ratio machine learning optimization design method based on information fusion, comprising: (1) constructing a wing lift-to-drag ratio and stiffness simulation model in Xfoil and Abaqus respectively; (2) iteratively optimizing the correlation between individual vectors of wing interlayer parameters to establish a population; (3) using DE / current / 0 to expand the population, and fusing three differential evolutions to generate 1500 wing interlayer parameter vectors; (4) fusing angle and distance information to construct a radial basis function machine learning prediction model; (5) evaluating the wing lift-to-drag ratio and stiffness; (6) based on a fitness function, screening the best wing interlayer parameter vector, updating the population and going to step (3) until the optimized structure meets the requirements. The present invention improves the distribution of offspring by population expansion, provides accurate prediction by a radial basis function machine learning prediction model fusing angle and distance information, and balances the optimization of wing lift-to-drag ratio and stiffness by a fitness function, thereby improving the wing lift-to-drag ratio optimization performance.

Description

Technical Field

[0001] The present invention relates to the fields of general artificial intelligence technology and swarm intelligence, and more specifically, to a machine learning optimization design method for a wing lift-to-drag ratio based on information fusion. Background Art

[0002] The problem of improving the lift-to-drag ratio of the wing is one of the key requirements for the performance design of the aircraft wing. The design purpose is to maximize the lift-to-drag ratio of the wing while meeting the wing stiffness. The lift-to-drag ratio directly reflects the aerodynamic efficiency of the aircraft during flight. A higher lift-to-drag ratio usually means less fuel consumption, lower flight resistance, and longer flight distance. The optimization design of wing structures involves time-consuming simulation calculations. It is unrealistic to use traditional heuristic algorithms to solve the problem of wing structure optimization design.

[0003] Machine learning models can fit the complex patterns and historical trends of real functions through evaluated historical calculation data, and the time spent on prediction using machine learning models is often negligible compared to real simulation calculations. Therefore, machine learning models can be established for the wing lift-to-drag ratio and stiffness to guide the optimization design process, and can be combined with classic optimization algorithms (such as differential evolution algorithms) to form an efficient design optimization framework.

[0004] Furthermore, the performance of the differential evolution algorithm depends on the distribution of the offspring population and the selection strategy. Existing differential evolution algorithms only improve the distribution of the offspring population by adjusting parameters and replacing evolutionary operations; and existing selection techniques do not balance objectives and constraints well; in addition, ordinary machine learning models find it difficult to provide accurate prediction results. Summary of the invention

[0005] In view of the above limitations of the prior art or the need for improved technology, the present invention proposes a machine learning optimization design method for the lift-to-drag ratio of a wing based on information fusion. Based on the time-consuming simulation calculation characteristics involved in the optimization design of the wing structure and the need for lift-to-drag ratio optimization design, the present invention designs a novel way to improve the distribution of offspring, a selection strategy, and a machine learning model modeling method. It considers combining machine learning technology with a differential evolution algorithm, and improving the distribution of offspring through population expansion, providing accurate predictions through a machine learning model that fuses angle information and distance information, and balancing the optimization of the wing lift-to-drag ratio and stiffness through an improved fitness function. The machine learning optimization design method designed by the present invention has stronger robustness, improves the optimization efficiency for the wing lift-to-drag ratio optimization problem, and can not only be used for the wing lift-to-drag ratio optimization design, but also provides a usable algorithm structure for the performance design of other complex structures.

[0006] To achieve the above object, according to one aspect of the present invention, a machine learning optimization design method for wing lift-to-drag ratio based on information fusion is provided, and the method comprises the following steps:

[0007] (1) Taking the center and thickness of each wing sandwich member as the optimization design parameters, and the wing stiffness and wing lift-to-drag ratio as the constraint conditions and optimization objectives, respectively, the wing lift-to-drag ratio and stiffness simulation models were constructed in Xfoil and Abaqus according to the loads and boundary constraints;

[0008] (2) According to the geometric constraints of the wing structure, a design space consisting of the optimized design parameters of the center and thickness of each member is constructed. The correlation between the individual vectors of the wing sandwich parameters is iteratively optimized in the design space to obtain the initialization population. The wing lift-to-drag ratio and wing stiffness of all the individual vectors of the wing sandwich parameters in the initialization population are evaluated in Xfoil and Abaqus, respectively. All the individual vectors of the wing sandwich parameters and their wing lift-to-drag ratio and wing stiffness are saved in the database.

[0009] (3) Based on the Euclidean distance between the individual vectors of wing interlayer parameters and the individual vectors of wing interlayer parameters, the local population was obtained, the differential evolution operation DE / current / 0 was used to expand the local population, and three differential evolution operations were applied to generate 1500 offspring wing interlayer parameter vectors;

[0010] (4) Calculate the angle information and distance information, and construct a radial basis function machine learning prediction model that integrates the angle information and distance information, and predict the wing lift-to-drag ratio and stiffness value of the offspring wing sandwich parameter vector;

[0011] (5) Based on the Spearman rank correlation coefficient, the local correlation of the wing lift-to-drag ratio and stiffness optimization of the local population is calculated, and the fitness function based on the correlation design is derived. The optimal offspring wing sandwich parameter vector is selected according to the fitness function, and the wing lift-to-drag ratio and stiffness of the optimal offspring wing sandwich parameter vector are evaluated in Xfoil and Abaqus respectively, and stored in the database;

[0012] (6) Based on the fitness function that integrates the wing lift-to-drag ratio and the stiffness value, the fitness values ​​of the optimal offspring wing sandwich parameter vector and the parent wing sandwich parameter vector are calculated and compared. The wing sandwich parameter vector with a larger fitness value is updated and enters the next generation population to determine whether the optimized structure meets the design performance. If so, the optimized structure is output; otherwise, go to step (3) until the optimized structure meets the requirements.

[0013] Furthermore, the wing lift-to-drag ratio and stiffness simulation model in step (1) has the following mathematical expression:

[0014]

[0015] In the above formula, P is the structural optimization parameter vector in the wing optimization problem, T is the transpose of the matrix, L and D These are the lift and drag simulated in Xfoil. Stiff is the wing stiffness simulated in Abaqus, is the number of structural optimization parameters, f and t are the Neumann boundary Γ t The body force density on the surface and the preset surface traction, is the Heaviside function, is the topological description function of the wing structure, u is the displacement field, is the random variable of the wing structure volume, is the wing structure displacement random variable, are the upper and lower bounds of the structural optimization parameters, is the lower limit of the wing structure stiffness, Find is the optimal solution to the wing optimization problem, Max is to maximize the wing lift-to-drag ratio, and St is the constraint that needs to be satisfied.

[0016] Furthermore, the specific steps of step (2) are as follows:

[0017] The first step is to determine the number of variables that need to be sampled N , and construct a design space consisting of the center and thickness optimization design parameters of each member according to the geometric constraints of the wing structure;

[0018] In the second step, the value range of the center and thickness optimization design parameters of each member is divided into N non-overlapping subintervals, and then optimize the design parameters of the center and thickness of each member. N The sub-intervals are randomly arranged, and the center and thickness of each member are optimized. N A sampling point is randomly selected in each subinterval, and the random sampling points of the center and thickness optimization design parameters of each member are combined to generate N indivual n Sample points in dimensional space;

[0019] The third step is to evaluate the Pearson correlation coefficient between all sample points in each dimension to obtain the correlation matrix. The Pearson correlation coefficient formula is as follows:

[0020] ,

[0021] In the above formula, Indicates i Dimensions and j The correlation between the dimensions It is k The sample point is iThe values ​​in the dimensions, All sample points in i The average value of the dimensions. It is k The sample point is j The values ​​in the dimensions, All sample points in j The average value of the dimensions. N is the total number of sample points;

[0022] The fourth step is to change the order of sampling points between dimensions to minimize the correlation matrix;

[0023] Step 5: Repeat step 3 and step 4 until the set iteration limit is reached to obtain the initialized population.

[0024] The sixth step is to simulate and evaluate the wing lift-to-drag ratio and wing stiffness of all wing sandwich parameter individual vectors in the initialized population in Xfoil and Abaqus respectively, and save all wing sandwich parameter individual vectors and their wing lift-to-drag ratio and stiffness data into the database.

[0025] Furthermore, the specific steps of step (3) are as follows:

[0026] The first step is to select the neighboring individual vectors of the wing layer parameters in the current population according to the Euclidean distance in the database. K Individual vectors of wing interlayer parameters are obtained to obtain the local population;

[0027] The second step is to use the differential evolution operation DE / current / 0 to expand the local population into a wing interlayer parameter individual vector that is three times the size of the population, and combine the local population with the expanded wing interlayer parameter individual vector as the parent population of the differential evolution operation;

[0028] In the third step, three differential evolution operations, DE / rand / 1, DE / rand / 2, and DE / current-to-rand / 1, are used to generate 1500 offspring wing layer parameter vectors for the parent population. The number of offspring wing layer parameter vectors generated by the three differential evolution operations is 500 respectively. The differential evolution operation expressions used are as follows:

[0029] DE / rand / 1: ,

[0030] DE / rand / 2: ,

[0031] DE / current / 0: ,

[0032] DE / current-to-rand / 1: ,

[0033] In the above formula, DE stands for differential evolution, rand indicates that the starting point of evolution is a random individual, current indicates that the starting point of evolution is a specific parent individual, to indicates the evolutionary direction of the differential evolution operation, 0, 1 and 2 indicate the number of random differential terms, and x r1 、x r2 、x r3 、x r4 and x r5 are five wing layer parameter vectors randomly selected from the parent population, x cur is the selected wing sandwich parameter vector, G 1 and G 2 is the scaling factor, is the wing layer parameter vector obtained after differential mutation.

[0034] Furthermore, the specific steps of step (4) are as follows:

[0035] The first step is to extract the wing lift-to-drag ratio and wing stiffness of the local population in the database and calculate the angle information. The expression is as follows:

[0036] ,

[0037] In the above formula, Ang is the angle information between the predicted wing sandwich parameter vector and the center point of the basis function, c o Represents the center point of the basis function o The value of x in the dimension o Represents the predicted wing sandwich parameter vector in o The values ​​in the dimensions, n is the total number of dimensions of the predicted wing sandwich parameter vector;

[0038] The second step is to extract the wing lift-to-drag ratio and wing stiffness of the local population in the database and calculate the distance information. The expression is as follows:

[0039] ,

[0040] In the above formula, Dist is the distance information between the predicted wing sandwich parameter vector and the center point of the basis function, c o Represents the center point of the basis function o The value of x in the dimension o Represents the predicted wing sandwich parameter vector in o The values ​​in the dimensions, nis the total number of dimensions of the predicted wing sandwich parameter vector;

[0041] The third step is to build a radial basis function machine learning prediction model that integrates angle information and distance information. The expression is as follows:

[0042] ,

[0043] In the above formula, A radial basis function machine learning prediction model that integrates angle information and distance information is built for the wing lift-to-drag ratio and wing stiffness. For the k Distance Information Basis Function The corresponding weight parameter, For the k Angle Information Basis Function The corresponding weight parameter, K is the total number of basis functions, P (x) represents a first-order polynomial. The cubic kernel function is used as the basis function for distance information. The expression is as follows: , the basis function for the angle information uses a linear kernel function, and the expression is as follows: ;

[0044] The fourth step is to use the constructed radial basis function machine learning prediction model that integrates angle information and distance information to predict the wing lift-to-drag ratio and stiffness value of the offspring wing sandwich parameter vector.

[0045] Furthermore, the specific steps of step (5) are as follows:

[0046] In the first step, the local correlation of the wing lift-to-drag ratio and the optimization of the stiffness value of the local population is calculated based on the Spearman rank correlation coefficient, and the fitness function designed based on the correlation is derived. The expression of the Spearman rank correlation coefficient is as follows:

[0047] ,

[0048] In the above formula, r h It is h The rank difference of the wing layer parameter vectors of the offspring generations, H is the total number of wing sandwich parameter vectors used to calculate the local correlation between the lift-to-drag ratio and the optimization of the stiffness value, is the Spearman rank correlation coefficient;

[0049] The fitness function formula based on correlation design is as follows:

[0050] ,

[0051] ,

[0052] In the above formula, C is the Spearman rank correlation coefficient The value of is adjusted to the coefficient value of 0-1. fit is the fitness value, It is i The offspring wing layer parameter vectors, It is i The value of the wing lift-to-drag ratio of the child wing sandwich parameter vector, It is i The values ​​of the wing stiffness of the child wing sandwich parameter vector;

[0053] In the second step, the predicted wing lift-to-drag ratio and stiffness values ​​are substituted into the fitness function, and the optimal offspring wing sandwich parameter vector with the largest fitness value is selected. The wing lift-to-drag ratio and stiffness of the optimal offspring wing sandwich parameter vector are evaluated in Xfoil and Abaqus respectively, and stored in the database.

[0054] Furthermore, in step (6), based on the fitness function integrating the wing lift-to-drag ratio and the stiffness value, the fitness values ​​of the best offspring wing sandwich parameter vector and the parent wing sandwich parameter vector are calculated and compared, and the wing sandwich parameter vector with a larger fitness value is updated to enter the next generation population. The specific steps are as follows:

[0055] The fitness value of the optimal offspring wing sandwich parameter vector is calculated based on the fitness function that combines the wing lift-to-drag ratio and the stiffness value;

[0056] The fitness value of the parent wing sandwich parameter vector is calculated based on the fitness function that combines the wing lift-to-drag ratio and the stiffness value;

[0057] The fitness values ​​of the two are compared, and the wing layer parameter vector with a larger fitness value is updated and enters the next generation population.

[0058] On the other hand, the present invention provides a computer device, including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a machine learning optimization design method for a wing lift-to-drag ratio based on information fusion as described in the aforementioned embodiment are implemented.

[0059] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a machine learning optimization design method for a wing lift-to-drag ratio based on information fusion as described in the aforementioned embodiment.

[0060] In summary, compared with the prior art, the above technical solution conceived by the present invention provides a machine learning optimization design method for wing lift-to-drag ratio based on information fusion, which mainly has the following beneficial effects:

[0061] 1. Considering the correlation between individual vectors of wing layer parameters, the population individuals are initialized, and the correlation between individual vectors of wing layer parameters is continuously improved through iterative optimization, ensuring that the generated population individuals are both uniform in space and minimize the correlation between dimensions, laying a solid foundation for subsequent optimization;

[0062] 2. After expanding the local population, the differential evolution operation DE / current / 0 fully retains the optimization information of the existing population and makes the offspring evolution path more diverse. It also uses three differential evolution operations to obtain offspring with richer characteristics.

[0063] 3. The radial basis function machine learning model constructed by integrating angle information and distance information has higher prediction accuracy, improves the efficiency of wing lift-to-drag ratio optimization design, and shortens the design cycle;

[0064] 4. The corresponding fitness function is designed by using the local correlation between the lift-to-drag ratio and stiffness optimization of the wing, and the search direction is dynamically adjusted to achieve a balance between the lift-to-drag ratio optimization and the stiffness optimization. In addition, the trend relationship between the lift-to-drag ratio optimization and the stiffness optimization in the local space can be captured by using local correlation instead of global correlation;

[0065] 5. The present invention can effectively solve complex optimization problems involving time-consuming simulation models and significantly improve the optimization efficiency. The method is not only applicable to the design optimization of expensive problems, but also has a wide range of promotion and application prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0066] Figure 1 A simplified flow chart of a wing lift-to-drag ratio machine learning optimization design method based on information fusion provided by the present invention. DETAILED DESCRIPTION

[0067] In order to make the purpose, technical scheme and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention. In addition, the technical features involved in the various embodiments of the present invention described below can be combined with each other as long as they do not conflict with each other.

[0068] See also Figure 1 The present invention provides a wing lift-to-drag ratio machine learning optimization design method based on information fusion, which is applicable to the wing lift-to-drag ratio optimization problem. Specifically, the method includes steps (1) to (6).

[0069] Step (1): Taking the center and thickness of each wing sandwich member as the optimization design parameters, and the wing stiffness and wing lift-to-drag ratio as the constraint conditions and optimization targets respectively, the wing lift-to-drag ratio and stiffness simulation models are constructed in Xfoil and Abaqus according to the loads and boundary constraints.

[0070] The mathematical expression of the wing lift-to-drag ratio and stiffness simulation model in step (1) is as follows:

[0071]

[0072] In the above formula, P is the structural optimization parameter vector in the wing optimization problem, T is the transpose of the matrix, L and D These are the lift and drag simulated in Xfoil. Stiff is the wing stiffness simulated in Abaqus, is the number of structural optimization parameters, f and t are the Neumann boundary Γ t The body force density on the surface and the preset surface traction, is the Heaviside function, is the topological description function of the wing structure, u is the displacement field, is the random variable of the wing structure volume, is the wing structure displacement random variable, are the upper and lower bounds of the structural optimization parameters, is the lower limit of the wing structure stiffness, Find is the optimal solution to the wing optimization problem, Max is to maximize the wing lift-to-drag ratio, and St is the constraint that needs to be satisfied.

[0073] Step (2): According to the geometric constraints of the wing structure, a design space consisting of the center and thickness optimization design parameters of each member is constructed. The correlation between the individual vectors of the wing sandwich parameters is iteratively optimized in the design space to obtain an initialization population. The wing lift-to-drag ratio and wing stiffness of all the individual vectors of the wing sandwich parameters in the initialization population are evaluated in Xfoil and Abaqus respectively, and all the individual vectors of the wing sandwich parameters and their wing lift-to-drag ratio and wing stiffness are saved in the database.

[0074] The specific steps of step (2) are as follows:

[0075] The first step is to determine the number of variables that need to be sampled N , and construct a design space consisting of the center and thickness optimization design parameters of each member according to the geometric constraints of the wing structure;

[0076] In the second step, the value range of the center and thickness optimization design parameters of each member is divided into N non-overlapping subintervals, and then optimize the design parameters of the center and thickness of each member. NThe sub-intervals are randomly arranged, and the center and thickness of each member are optimized. N A sampling point is randomly selected in each subinterval, and the random sampling points of the center and thickness optimization design parameters of each member are combined to generate N indivual n Sample points in dimensional space;

[0077] The third step is to evaluate the Pearson correlation coefficient between all sample points in each dimension to obtain the correlation matrix. The Pearson correlation coefficient formula is as follows:

[0078] ,

[0079] In the above formula, Indicates i Dimensions and j The correlation between the dimensions, It is k The sample point is i The values ​​in the dimensions, All sample points in i The average value of the dimensions. It is k The sample point is j The values ​​in the dimensions, All sample points in j The average value of the dimensions. N is the total number of sample points;

[0080] The fourth step is to change the order of sampling points between dimensions to minimize the correlation matrix;

[0081] Step 5: Repeat step 3 and step 4 until the set iteration limit is reached to obtain the initialized population.

[0082] The sixth step is to simulate and evaluate the wing lift-to-drag ratio and wing stiffness of all wing sandwich parameter individual vectors in the initialized population in Xfoil and Abaqus respectively, and save all wing sandwich parameter individual vectors and their wing lift-to-drag ratio and stiffness data into the database.

[0083] It should be noted that the total number of dimensions in step (2) is the same as the number of structural optimization parameters in step (1).

[0084] Step (3): Based on the Euclidean distance between the individual vectors of wing interlayer parameters and the individual vectors of wing interlayer parameters, the local population is obtained, the local population is expanded using the differential evolution operation DE / current / 0, and three differential evolution operations are applied to generate 1500 offspring wing interlayer parameter vectors.

[0085] The specific steps of step (3) are as follows:

[0086] The first step is to select the neighboring individual vectors of the wing layer parameters in the current population according to the Euclidean distance in the database. K Individual vectors of wing interlayer parameters are obtained to obtain the local population;

[0087] The second step is to use the differential evolution operation DE / current / 0 to expand the local population into a wing interlayer parameter individual vector that is three times the size of the population, and combine the local population with the expanded wing interlayer parameter individual vector as the parent population of the differential evolution operation;

[0088] In the third step, three differential evolution operations, DE / rand / 1, DE / rand / 2, and DE / current-to-rand / 1, are used to generate 1500 offspring wing layer parameter vectors for the parent population. The number of offspring wing layer parameter vectors generated by the three differential evolution operations is 500 respectively. The differential evolution operation expressions used are as follows:

[0089] DE / rand / 1: ,

[0090] DE / rand / 2: ,

[0091] DE / current / 0: ,

[0092] DE / current-to-rand / 1: ,

[0093] In the above formula, DE stands for differential evolution, rand indicates that the starting point of evolution is a random individual, current indicates that the starting point of evolution is a specific parent individual, to indicates the evolutionary direction of the differential evolution operation, 0, 1 and 2 indicate the number of random differential terms, and x r1 、x r2 、x r3 、x r4 and x r5 are five wing layer parameter vectors randomly selected from the parent population, x cur is the selected wing sandwich parameter vector, G 1 and G 2 is the scaling factor, is the wing layer parameter vector obtained after differential mutation.

[0094] Step (4): Calculate the angle information and distance information, and construct a radial basis function machine learning prediction model that integrates the angle information and distance information, and predict the wing lift-to-drag ratio and stiffness value of the offspring wing sandwich parameter vector.

[0095] The specific steps of step (4) are as follows:

[0096] The first step is to extract the wing lift-to-drag ratio and wing stiffness of the local population in the database and calculate the angle information. The expression is as follows:

[0097] ,

[0098] In the above formula, Ang is the angle information between the predicted wing sandwich parameter vector and the center point of the basis function, c o Represents the center point of the basis function o The value of x in the dimension o Represents the predicted wing sandwich parameter vector in o The values ​​in the dimensions, n is the total number of dimensions of the predicted wing sandwich parameter vector;

[0099] The second step is to extract the wing lift-to-drag ratio and wing stiffness of the local population in the database and calculate the distance information. The expression is as follows:

[0100] ,

[0101] In the above formula, Dist is the distance information between the predicted wing sandwich parameter vector and the center point of the basis function, c o Represents the center point of the basis function o The value of x in the dimension o Represents the predicted wing sandwich parameter vector in o The values ​​in the dimensions, n is the total number of dimensions of the predicted wing sandwich parameter vector;

[0102] The third step is to build a radial basis function machine learning prediction model that integrates angle information and distance information. The expression is as follows:

[0103] ,

[0104] In the above formula, A radial basis function machine learning prediction model that integrates angle information and distance information is built for the wing lift-to-drag ratio and wing stiffness. For the k Distance Information Basis Function The corresponding weight parameter, For the k Angle Information Basis Function The corresponding weight parameter, K is the total number of basis functions, P(x) represents a first-order polynomial. The cubic kernel function is used as the basis function for distance information. The expression is as follows: , the basis function for the angle information uses a linear kernel function, and the expression is as follows: ;

[0105] The fourth step is to use the constructed radial basis function machine learning prediction model that integrates angle information and distance information to predict the wing lift-to-drag ratio and stiffness value of the offspring wing sandwich parameter vector.

[0106] Step (5): Based on the Spearman rank correlation coefficient, the local correlation of the wing lift-to-drag ratio and the stiffness value optimization of the local population is calculated, and the fitness function designed based on the correlation is derived. The optimal offspring wing sandwich parameter vector is selected according to the fitness function, and the wing lift-to-drag ratio and stiffness of the optimal offspring wing sandwich parameter vector are evaluated in Xfoil and Abaqus respectively, and stored in the database.

[0107] The specific steps of step (5) are as follows:

[0108] In the first step, the local correlation of the wing lift-to-drag ratio and the optimization of the stiffness value of the local population is calculated based on the Spearman rank correlation coefficient, and the fitness function designed based on the correlation is derived. The expression of the Spearman rank correlation coefficient is as follows:

[0109] ,

[0110] In the above formula, r h It is h The rank difference of the wing layer parameter vectors of the offspring generations, H is the total number of wing sandwich parameter vectors used to calculate the local correlation between the lift-to-drag ratio and the optimization of the stiffness value, is the Spearman rank correlation coefficient; it should be noted that ,in, and are the ranks of the wing lift-to-drag ratio and stiffness value, respectively;

[0111] The fitness function formula based on correlation design is as follows:

[0112] ,

[0113] ,

[0114] In the above formula, C is the Spearman rank correlation coefficient The value of is adjusted to the coefficient value of 0-1. fit is the fitness value, It is i The offspring wing layer parameter vectors, It is i The value of the wing lift-to-drag ratio of the child wing sandwich parameter vector, It is i The value of the wing stiffness of the child wing sandwich parameter vector; it should be noted that, C The closer it is to 0, the higher the correlation between the lift-to-drag ratio of the wing structure and the stiffness optimization; when the wing stiffness constraint condition is met, its value is taken as the lower limit of the wing stiffness value, that is, no optimization is performed;

[0115] In the second step, the predicted wing lift-to-drag ratio and stiffness values ​​are substituted into the fitness function, and the optimal offspring wing sandwich parameter vector with the largest fitness value is selected. The wing lift-to-drag ratio and stiffness of the optimal offspring wing sandwich parameter vector are evaluated in Xfoil and Abaqus respectively, and stored in the database.

[0116] Step (6): Based on the fitness function that integrates the wing lift-to-drag ratio and the stiffness value, the fitness values ​​of the optimal offspring wing sandwich parameter vector and the parent wing sandwich parameter vector are calculated and compared. The wing sandwich parameter vector with a larger fitness value is updated and entered into the next generation population to determine whether the optimized structure meets the design performance. If so, the optimized structure is output. Otherwise, go to step (3) until the optimized structure meets the requirements.

[0117] In step (6), based on the fitness function that integrates the wing lift-to-drag ratio and the stiffness value, the fitness values ​​of the best offspring wing sandwich parameter vector and the parent wing sandwich parameter vector are calculated and compared. The wing sandwich parameter vector with a larger fitness value is updated and entered into the next generation population. The specific steps are as follows:

[0118] The fitness value of the optimal offspring wing sandwich parameter vector is calculated based on the fitness function that combines the wing lift-to-drag ratio and the stiffness value;

[0119] The fitness value of the parent wing sandwich parameter vector is calculated based on the fitness function that combines the wing lift-to-drag ratio and the stiffness value;

[0120] The fitness values ​​of the two are compared, and the wing layer parameter vector with a larger fitness value is updated and enters the next generation population.

[0121] Example

[0122] See also Figure 1 In this embodiment, a benchmark test function is used to illustrate the optimization performance of a wing lift-to-drag ratio machine learning optimization design method based on information fusion provided in this embodiment. The expression of the benchmark test function is as follows:

[0123] ,

[0124]

[0125] In the above formula, represents the objective function to be minimized, Indicates i Optimization parameters, Indicates i +1 optimization parameter, n represents the number of optimization parameters, M represents the rotation matrix, y represents the optimized parameter vector after transformation by the rotation matrix, z represents the optimization parameter vector, represents the first optimized parameter vector after transformation by the rotation matrix i A value, represents the first constraint function, represents the second constraint function, Represents the third constraint function, Min means minimizing the objective function, st represents the constraints that need to be satisfied, and where represents the value range.

[0126] The above-mentioned benchmark test function is processed respectively through steps (1) to (6) of a wing lift-to-drag ratio machine learning optimization design method based on information fusion provided by the present invention to obtain experimental results.

[0127] In order to further illustrate the present embodiment, a machine learning optimization design method for wing lift-to-drag ratio based on information fusion of the present embodiment is compared with another algorithm MPMLS, and the maximum number of simulation evaluations of the present embodiment is set to 3000. The experimental results are shown in Table 1. Under the same number of simulation samples, the method of the present embodiment obtains the effect of approximating the global optimal solution and is significantly better than MPMLS, which reflects the optimization efficiency of the machine learning optimization design method for wing lift-to-drag ratio based on information fusion of the present embodiment. It can be considered that the method of the present embodiment has good performance in the optimization problem of lift-to-drag ratio of wing structure.

[0128] Table 1 Comparison of optimization results of different methods

[0129]

[0130] The present invention provides a machine learning optimization design method for wing lift-to-drag ratio based on information fusion, which combines machine learning technology with a differential evolution algorithm, improves the distribution of offspring through population expansion, provides accurate predictions through a machine learning model that fuses angle information and distance information, and balances the optimization of the wing lift-to-drag ratio and stiffness through a fitness function that integrates the wing lift-to-drag ratio and stiffness value. The method provides a systematic solution to the wing lift-to-drag ratio optimization problem.

[0131] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, and when the processor executes the computer program, the steps of a machine learning optimization design method for a wing lift-to-drag ratio based on information fusion of the aforementioned embodiment are implemented.

[0132] In a third aspect, the present invention provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of a machine learning optimization design method for a wing lift-to-drag ratio based on information fusion in the aforementioned embodiment.

[0133] It will be easily understood by those skilled in the art that the above are only preferred embodiments of the present invention and are not intended to limit the present invention. Any modifications, equivalent substitutions and improvements made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A machine learning optimization design method for wing lift-to-drag ratio based on information fusion, characterized in that: The method comprises the following steps: (1) Taking the center and thickness of each wing sandwich member as the optimization design parameters, and the wing stiffness and wing lift-to-drag ratio as the constraint conditions and optimization objectives, respectively, the wing lift-to-drag ratio and stiffness simulation models were constructed in Xfoil and Abaqus according to the loads and boundary constraints; (2) According to the geometric constraints of the wing structure, a design space consisting of the optimized design parameters of the center and thickness of each member is constructed. The correlation between the individual vectors of the wing sandwich parameters is iteratively optimized in the design space to obtain the initialization population. The wing lift-to-drag ratio and wing stiffness of all the individual vectors of the wing sandwich parameters in the initialization population are evaluated in Xfoil and Abaqus, respectively. All the individual vectors of the wing sandwich parameters and their wing lift-to-drag ratio and wing stiffness are saved in the database. (3) Based on the Euclidean distance between the individual vectors of wing interlayer parameters and the individual vectors of wing interlayer parameters, the local population was obtained, the differential evolution operation DE / current / 0 was used to expand the local population, and three differential evolution operations were applied to generate 1500 offspring wing interlayer parameter vectors; (4) Calculate the angle information and distance information, and construct a radial basis function machine learning prediction model that integrates the angle information and distance information, and predict the wing lift-to-drag ratio and stiffness value of the offspring wing sandwich parameter vector; (5) Based on the Spearman rank correlation coefficient, the local correlation of the wing lift-to-drag ratio and stiffness optimization of the local population is calculated, and the fitness function based on the correlation design is derived. The optimal offspring wing sandwich parameter vector is selected according to the fitness function, and the wing lift-to-drag ratio and stiffness of the optimal offspring wing sandwich parameter vector are evaluated in Xfoil and Abaqus respectively, and stored in the database; (6) Based on the fitness function that integrates the wing lift-to-drag ratio and the stiffness value, the fitness values ​​of the optimal offspring wing sandwich parameter vector and the parent wing sandwich parameter vector are calculated and compared. The wing sandwich parameter vector with a larger fitness value is updated and enters the next generation population to determine whether the optimized structure meets the design performance. If so, the optimized structure is output; otherwise, go to step (3) until the optimized structure meets the requirements.

2. The method according to claim 1, characterized in that The mathematical expression of the wing lift-to-drag ratio and stiffness simulation model in step (1) is as follows: In the above formula, P is the structural optimization parameter vector in the wing optimization problem, T is the transpose of the matrix, L and D These are the lift and drag simulated in Xfoil. Stiff is the wing stiffness simulated in Abaqus, is the number of structural optimization parameters, f and t are the Neumann boundary Γ t The body force density on the surface and the preset surface traction, is the Heaviside function, is the topological description function of the wing structure, u is the displacement field, is the random variable of the wing structure volume, is the wing structure displacement random variable, are the upper and lower bounds of the structural optimization parameters, is the lower limit of the wing structure stiffness, Find is the optimal solution to the wing optimization problem, Max is to maximize the wing lift-to-drag ratio, and St is the constraint that needs to be satisfied.

3. The method according to claim 1, characterized in that The specific steps of step (2) are as follows: The first step is to determine the number of variables that need to be sampled N , and construct a design space consisting of the center and thickness optimization design parameters of each member according to the geometric constraints of the wing structure; In the second step, the value range of the center and thickness optimization design parameters of each member is divided into N non-overlapping subintervals, and then optimize the design parameters of the center and thickness of each member. N The sub-intervals are randomly arranged, and the center and thickness of each member are optimized. N A sampling point is randomly selected in each subinterval, and the random sampling points of the center and thickness optimization design parameters of each member are combined to generate N indivual n Sample points in dimensional space; The third step is to evaluate the Pearson correlation coefficient between all sample points in each dimension to obtain the correlation matrix. The Pearson correlation coefficient formula is as follows: ， In the above formula, Indicates i Dimensions and j The correlation between the dimensions It is k The sample point is i The values ​​in the dimensions, All sample points in i The average value of the dimensions. It is k The sample point is j The values ​​in the dimensions, All sample points in j The average value of the dimensions. N is the total number of sample points; The fourth step is to change the order of sampling points between dimensions to minimize the correlation matrix; Step 5: Repeat step 3 and step 4 until the set iteration limit is reached to obtain the initialized population. The sixth step is to simulate and evaluate the wing lift-to-drag ratio and wing stiffness of all wing sandwich parameter individual vectors in the initialized population in Xfoil and Abaqus respectively, and save all wing sandwich parameter individual vectors and their wing lift-to-drag ratio and stiffness data into the database.

4. The method according to claim 1, characterized in that: The specific steps of step (3) are as follows: The first step is to select the neighboring individual vectors of the wing layer parameters in the current population according to the Euclidean distance in the database. K Individual vectors of wing interlayer parameters are obtained to obtain the local population; The second step is to use the differential evolution operation DE / current / 0 to expand the local population into a wing interlayer parameter individual vector that is three times the size of the population, and combine the local population with the expanded wing interlayer parameter individual vector as the parent population of the differential evolution operation; In the third step, three differential evolution operations, DE / rand / 1, DE / rand / 2, and DE / current-to-rand / 1, are used to generate 1500 offspring wing layer parameter vectors for the parent population. The number of offspring wing layer parameter vectors generated by the three differential evolution operations is 500 respectively. The differential evolution operation expressions used are as follows: FROM / row / 1: , DE / row / 2: , EN / current / 0: , DE / current-to-rand / 1: , In the above formula, DE stands for differential evolution, rand indicates that the starting point of evolution is a random individual, current indicates that the starting point of evolution is a specific parent individual, to indicates the evolutionary direction of the differential evolution operation, 0, 1 and 2 indicate the number of random differential terms, and x r1 、x r2 、x r3 、x r4 and x r5 are five wing layer parameter vectors randomly selected from the parent population, x cur is the selected wing sandwich parameter vector, G 1 and G 2 is the scaling factor, is the wing layer parameter vector obtained after differential mutation.

5. The method according to claim 1, characterized in that The specific steps of step (4) are as follows: The first step is to extract the wing lift-to-drag ratio and wing stiffness of the local population in the database and calculate the angle information. The expression is as follows: ， In the above formula, Ang is the angle information between the predicted wing sandwich parameter vector and the center point of the basis function, c o Represents the center point of the basis function o The value of x in the dimension o Represents the predicted wing sandwich parameter vector in o The values ​​in the dimensions, n is the total number of dimensions of the predicted wing sandwich parameter vector; The second step is to extract the wing lift-to-drag ratio and wing stiffness of the local population in the database and calculate the distance information. The expression is as follows: ， In the above formula, Dist is the distance information between the predicted wing sandwich parameter vector and the center point of the basis function, c o Represents the center point of the basis function o The value of x in the dimension o Represents the predicted wing sandwich parameter vector in o The values ​​in the dimensions, n is the total number of dimensions of the predicted wing sandwich parameter vector; The third step is to build a radial basis function machine learning prediction model that integrates angle information and distance information. The expression is as follows: ， In the above formula, A radial basis function machine learning prediction model that integrates angle information and distance information is built for the wing lift-to-drag ratio and wing stiffness. For the k Distance Information Basis Function The corresponding weight parameter, For the k Angle Information Basis Function The corresponding weight parameter, K is the total number of basis functions, P (x) represents a first-order polynomial. The cubic kernel function is used as the basis function for distance information. The expression is as follows: , the basis function for the angle information uses a linear kernel function, and the expression is as follows: ; The fourth step is to use the constructed radial basis function machine learning prediction model that integrates angle information and distance information to predict the wing lift-to-drag ratio and stiffness value of the offspring wing sandwich parameter vector.

6. The method according to claim 1, characterized in that The specific steps of step (5) are as follows: In the first step, the local correlation of the wing lift-to-drag ratio and the optimization of the stiffness value of the local population is calculated based on the Spearman rank correlation coefficient, and the fitness function designed based on the correlation is derived. The expression of the Spearman rank correlation coefficient is as follows: ， In the above formula, r h It is h The rank difference of the wing layer parameter vectors of the offspring generations, H is the total number of wing sandwich parameter vectors used to calculate the local correlation between the lift-to-drag ratio and the optimization of the stiffness value, is the Spearman rank correlation coefficient; The fitness function formula based on correlation design is as follows: ， ， In the above formula, C is the Spearman rank correlation coefficient The value of is adjusted to the coefficient value of 0-1. fit is the fitness value, It is i The offspring wing layer parameter vector, It is i The value of the wing lift-to-drag ratio of the child wing sandwich parameter vector, It is i The values ​​of the wing stiffness of the child wing sandwich parameter vector; In the second step, the predicted wing lift-to-drag ratio and stiffness values ​​are substituted into the fitness function, and the optimal offspring wing sandwich parameter vector with the largest fitness value is selected. The wing lift-to-drag ratio and stiffness of the optimal offspring wing sandwich parameter vector are evaluated in Xfoil and Abaqus respectively, and stored in the database.

7. The method according to claim 1, characterized in that In step (6), based on the fitness function integrating the wing lift-to-drag ratio and the stiffness value, the fitness values ​​of the best offspring wing sandwich parameter vector and the parent wing sandwich parameter vector are calculated and compared, and the wing sandwich parameter vector with a larger fitness value is updated to enter the next generation population. The specific steps are as follows: The fitness value of the optimal offspring wing sandwich parameter vector is calculated based on the fitness function that combines the wing lift-to-drag ratio and the stiffness value; The fitness value of the parent wing sandwich parameter vector is calculated based on the fitness function that combines the wing lift-to-drag ratio and the stiffness value; The fitness values ​​of the two are compared, and the wing layer parameter vector with a larger fitness value is updated and enters the next generation population.

8. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images