WO2021142916A1 - Proxy-assisted evolutionary algorithm-based airfoil optimization method and apparatus - Google Patents

Abstract

A proxy-assisted evolutionary algorithm-based airfoil optimization method and apparatus, and a computer device, and a storage medium. By performing multiple global searches and local searches on an initial data point set in a sample library, the number of data points in the initial data point set in the sample library is increased to be greater than or equal to the maximum number of real evaluations.

Description

Airfoil optimization method and device based on agent-assisted evolutionary algorithm

This application claims the priority of a Chinese patent application filed with the Chinese Patent Office on January 15, 2020, the application number is 202010041514.7, and the application title is "Agent-assisted evolutionary algorithm-based airfoil optimization method and device", the entire content of which is incorporated by reference Incorporated in this application.

Technical field

This application relates to the field of data processing technology, and in particular to an airfoil optimization method, device, computer equipment, and storage medium based on an agent-assisted evolutionary algorithm.

Background technique

Evolutionary algorithm, also known as meta-heuristic algorithm, has been widely used in various engineering optimizations. However, if these evolutionary algorithms are applied to expensive optimization problems involving computationally expensive simulations, the computational cost will be huge. Expensive optimization problems are very different from ordinary optimization problems. The response function of expensive optimization problems is usually a simulation model, such as finite element analysis (abbreviated as FEA) and computational fluid dynamics (abbreviated as CFD).

At present, agent-assisted evolutionary algorithms that combine agent-assisted and evolutionary algorithms are getting more and more attention in dealing with expensive optimization problems, that is, adding points to sample data points that are lacking in expensive problems to improve sample data points. The total number is important consideration.

Commonly used points-adding criteria include PoI point-adding criterion (that is, point-adding criterion based on improved probability), ExI-pointing criterion (that is, point-adding criterion based on expected improvement), and LCB-pointing criterion (that is, point-adding criterion based on lower confidence bound), these three points-adding criteria Both effectively enhance the accuracy of the agent model in the application of agent assistance and evolutionary algorithms, thereby accelerating the final convergence of the algorithm. However, the above three points-adding criteria simply use a single criterion or combine multiple criteria into a scalar criterion, resulting in a poorer effect of adding sample points in a limited data sample, and the increased sample points have an impact on the proxy model. The increase in accuracy is less.

Summary of the invention

The embodiments of this application provide an airfoil optimization method, device, computer equipment, and storage medium based on the agent-assisted evolutionary algorithm, aiming to solve the problem of using point addition criteria based on improved probability, point addition criteria based on expected improvement, or point addition criteria based on improved probability in the prior art. When dealing with the addition of points in expensive optimization problems, the lower confidence bound’s adding point criterion simply uses a single criterion or combines multiple criteria into a scalar criterion, which results in a poor effect of adding sample points in a limited data sample. Moreover, the increased sample points have fewer problems to improve the accuracy of the proxy model.

In the first aspect, an embodiment of the present application provides an airfoil optimization method based on an agent-assisted evolutionary algorithm, which includes:

Determine whether the data point adding request sent by the client is received;

If a data point addition request sent by the client is received, the initial data point set in the sample library is obtained, and the maximum number of real evaluations is obtained according to the total number of data points in the initial data point set and the total number of optimized points set in advance; Among them, each data point includes a decision variable corresponding to the wing geometric shape control point and an evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector;

Judging whether the total number of data points in the initial data point set is less than the maximum number of real evaluation times;

If the total number of data points in the initial data point set is less than the maximum number of real evaluations, each data point in the initial data point set is used as the training sample of the first agent model to be trained to obtain the corresponding first current The agent model, according to the first current agent model and preset first individual screening conditions, searches the first type final population generated according to the first type of initial population genetic evolution to obtain the first target individual and the second target individual;

The data points corresponding to the first target individual and the data points corresponding to the second target individual are both added to the initial data point set to obtain the current data point set; wherein the data point corresponding to the first target individual is determined by The first target individual and the first target individual are input to the pre-stored target function for calculation and are composed of the corresponding first true function value; the data point corresponding to the second target individual is composed of the second target individual and the second target individual Input to the objective function corresponding to the second real function value composition;

Obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set, and each data point in the target data point set is used as the second target data point set. Train the training samples of the agent model to obtain the corresponding second current agent model, and perform the process on the second type final population generated based on the genetic evolution of the second type initial population according to the second current agent model and preset second individual screening conditions Search to get the third target individual;

The data point corresponding to the third target individual is added to the current data point set to obtain the final data point set, the final data point set is used as the initial data point set, and the execution is returned to determine the total number of data points in the initial data point set. The step of whether the number is less than the maximum number of real evaluation times; wherein the data point corresponding to the third target individual is input by the third target individual and the third target individual to the third true function value corresponding to the target function Composition; and

If the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, the initial data point set is sent to the client.

In the second aspect, an embodiment of the present application provides an airfoil optimization device based on the agent-assisted evolutionary algorithm, which includes a unit for executing the airfoil optimization method based on the agent-assisted evolution algorithm described in the first aspect. In a third aspect, an embodiment of the present application provides a computer device, which includes a memory, a processor, and a computer program stored on the memory and running on the processor, and the processor executes the computer The program implements the airfoil optimization method based on the agent-assisted evolution algorithm described in the first aspect above.

In a fourth aspect, an embodiment of the present application also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the processor executes the above-mentioned first On the one hand, the airfoil optimization method based on the agent-assisted evolutionary algorithm.

Description of the drawings

In order to explain the technical solutions of the embodiments of the present application more clearly, the following will briefly introduce the drawings used in the description of the embodiments. Obviously, the drawings in the following description are some embodiments of the present application. Ordinary technicians can obtain other drawings based on these drawings without creative work.

FIG. 1 is a schematic diagram of an application scenario of an airfoil optimization method based on an agent-assisted evolution algorithm provided by an embodiment of the application;

2 is a schematic flow chart of an airfoil optimization method based on an agent-assisted evolutionary algorithm provided by an embodiment of the application;

FIG. 3 is a schematic diagram of a sub-process of an airfoil optimization method based on an agent-assisted evolutionary algorithm provided by an embodiment of the application;

4 is a schematic diagram of another sub-process of the airfoil optimization method based on the agent-assisted evolution algorithm provided by an embodiment of the application;

5 is a schematic diagram of another sub-process of the airfoil optimization method based on the agent-assisted evolution algorithm provided by an embodiment of the application;

Fig. 6 is a schematic block diagram of an airfoil optimization device based on an agent-assisted evolutionary algorithm provided by an embodiment of the application;

FIG. 7 is a schematic block diagram of a computer device provided by an embodiment of the application.

Detailed ways

The technical solutions in the embodiments of the present application will be described clearly and completely in conjunction with the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are part of the embodiments of the present application, rather than all of them. Based on the embodiments in this application, all other embodiments obtained by those of ordinary skill in the art without creative work shall fall within the protection scope of this application.

Please refer to FIGS. 1 and 2. FIG. 1 is a schematic diagram of an application scenario of an airfoil optimization method based on an agent-assisted evolution algorithm provided by an embodiment of the application; FIG. 2 is an airfoil optimization based on an agent-assisted evolution algorithm provided by an embodiment of the application A schematic flow diagram of the method. The airfoil optimization method based on the agent-assisted evolutionary algorithm is applied to the server, and the method is executed by the application software installed in the server.

As shown in Figure 2, the method includes steps S110 to S180.

S110: Determine whether a data point addition request sent by the client is received.

In order to understand the technical solution of the present application more clearly, the terminals involved are introduced below. This application describes the technical solution from the perspective of the server.

The first is the client. The client can be understood as a user terminal. The user terminal can be a smart phone, tablet computer, notebook computer, desktop computer, personal digital assistant, wearable device and other electronic devices with communication functions. The user terminal sends data points. Add a request to the server. The second is the server. The server receives the data point adding request sent by the client. Based on the initial data point set stored in the server, the initial data point set is selectively added to the sample through the combination of the proxy model and the genetic evolution algorithm. Click to get a collection of data points. The set of data points obtained from the server is sent to the client.

In this embodiment, the server detects whether the data point addition request sent by the client is received, and when the server receives the data point addition request sent by the client, the subsequent step S120 is executed. When the server does not receive the data point addition request sent by the client When the path planning request is made, step S110 is executed again after waiting for the preset delay time.

S120. If a data point addition request sent by the client is received, the initial data point set in the sample library is obtained, and the maximum true evaluation is obtained according to the total number of data points in the initial data point set and the total number of optimized points set in advance Times; where each data point includes a decision variable corresponding to the wing geometric shape control point and an evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector.

In this embodiment, the initial total number in the initial data point set in the sample library is known (for example, 70 initial data points), and each data point in the initial data point set includes the corresponding wing geometry control point For example, 14 control points composed of a B-spline curve are used as wing geometry control points. The positions of these 14 control points constitute the decision variable corresponding to a data point in the initial data point set . Moreover, the evaluation value of each decision variable can be calculated by the following formula (1):

Among them, in formula (1), f _{Airfoil_j} represents the evaluation value corresponding to the decision variable of the jth data point in the initial data point set, and D _1j represents the decision variable of the jth data point under the preset fluid dynamics design condition 1 The resistance obtained by CFD simulation, D _2j represents the resistance obtained by CFD simulation of the decision variable of the jth data point under the preset fluid dynamics design condition 2, L _1j represents the decision variable of the jth data point The lift coefficient obtained by CFD simulation under the set fluid dynamics design condition 1, L _2j represents the lift coefficient obtained by CFD simulation under the preset fluid dynamic design condition 2 for the decision variable representing the jth data point;

The decision variable representing the j-th data point is subjected to CFD simulation under the preset fluid dynamics design condition 1 to obtain the resistance of the baseline design,

The decision variable representing the j-th data point is subjected to CFD simulation under the preset fluid dynamics design condition 2 to obtain the resistance of the baseline design,

The decision variable representing the j-th data point is subjected to CFD simulation under the preset fluid dynamics design condition 1 to obtain the lift coefficient of the baseline design,

The decision variable representing the jth data point is subjected to CFD simulation under the preset fluid dynamics design condition 2 to obtain the lift coefficient of the baseline design.

And because the total number of optimization points is preset (for example, the total number of optimization points is 84), the maximum is obtained from the sum of the total number of data points in the initial data point set and the preset total number of optimization points The number of true evaluations (for example, the maximum number of true evaluations=70+84=154), where the maximum number of true evaluations is obtained to determine the number of iterations of subsequent steps S130-S170, thereby increasing the initial data point set in the sample library by multiple Target data points to improve the defect of insufficient data points in the costly optimization problem of wing airfoil design.

S130: Determine whether the total number of data points in the initial data point set is less than the maximum number of true evaluation times.

In this embodiment, since the initial total number of data points in the initial data point set is less than the maximum number of real evaluations, the subsequent steps S130-S170 need to be iterated multiple times until the initial data point set The total number of middle data points is greater than or equal to the maximum number of real evaluation times, and the initial data point set is sent to the client. Through an iterative process, data points with better performance are continuously added to the initial data point set, so as to realize the addition of data points for expensive optimization problems.

S140. If the total number of data points in the initial data point set is less than the maximum number of real evaluation times, use each data point in the initial data point set as a training sample of the first agent model to be trained to obtain the corresponding first data point set. A current agent model, according to the first current agent model and the preset first individual screening conditions, the first type of final population generated according to the first type of initial population genetic evolution is searched, and the first target individual and the first target individual are obtained. Two target individuals.

In this embodiment, when the total number of data points in the initial data point set is less than the maximum number of true evaluations, then it is necessary to use the data points in the initial data point set as the basis for gradual genetic evolution to generate multiple target data The points continue to increase to the initial set of data points.

In the costly optimization problem of the wing shape of the wing, there must be more superior solution individuals in several sub-regions of the entire feasible region. Therefore, exploring the promising search region in the feasible region is the main purpose of the global search stage (that is, the main purpose of step S140). Purpose). In the global search phase, a first current agent model is used as the agent model in the global search phase. After the surrogate model is established, the evolutionary algorithm begins to search in the entire feasible domain. The surrogate model is used to obtain the predicted value of the target individual, and the uncertainty value corresponding to each target individual is also obtained.

In an embodiment, the first agent model to be trained includes a first Chriskin model to be trained, a first radial basis function model to be trained, and a first polynomial response surface model to be trained; the first current agent model Including the first Chriskin model, the first radial basis function model, and the first polynomial response surface model.

The first Kriging model is the first Kriging model, the first radial basis function model to be trained is the first RBF model, and the first polynomial response surface model is the first PR model. The proxy model is an existing proxy model, and its model expressions will not be repeated here.

Since the first current proxy model includes the above three proxy models, in order to improve the accuracy of the proxy model, the final predicted value of a certain data point input to the first current proxy model is the predicted value corresponding to the three models above. Find the weighted sum.

In an embodiment, in step S140, using each data point in the initial data point set as a training sample of the first agent model to be trained to obtain the corresponding first current agent model includes:

Using the decision variable of each data point in the initial data point set as the input of the first Chriskin model to be trained, and the evaluation value corresponding to each decision variable as the output of the first Chriskin model to be trained, Training the first Chris King model to be trained to obtain the first Chris King model;

The decision variable of each data point in the initial data point set is used as the input of the first radial basis function model to be trained, and the evaluation value corresponding to each decision variable is used as the first radial basis function model to be trained. Output, training the first radial basis function model to be trained to obtain a first radial basis function model;

Using the decision variable of each data point in the initial data point set as the input of the first polynomial response surface model to be trained, and the evaluation value corresponding to each decision variable as the output of the first polynomial response surface model to be trained, The first polynomial response surface model to be trained is trained to obtain the first polynomial response surface model.

In this embodiment, all data points included in the initial data point set are used as the first to-be-trained Chriskin model, the first to-be-trained radial basis function model, and the first to-be-trained The training samples of the polynomial response surface model are correspondingly obtained through training to obtain the first Chriskin model, the first radial basis function model, and the first polynomial response surface model. Through the above process, the acquisition of the first current agent model is achieved.

In one embodiment, as shown in FIG. 3, in step S140, according to the first current agent model and the preset first individual screening conditions, the final population of the first type generated according to the genetic evolution of the first type of initial population is performed. Search, the first target individual and the second target individual obtained include:

S1401, according to the vector feature dimensions of the decision variables in the initial data point set, randomly generate Ng variable solutions in a Latin hypercube design to form the first type of initial population; wherein, each variable solution is the first type of initial population For an individual in, the characteristic dimension of the solution of each variable is the same as the characteristic dimension of the decision variable;

S1402. Obtain the current iteration algebra of the first type, and determine whether the current iteration algebra of the first type reaches a preset maximum iteration algebra; wherein, the initial value of the current iteration algebra of the first type is 1.

S1403. If the current iteration algebra of the first type does not reach the maximum iteration algebra, perform simulated binary crossover and polynomial mutation on the initial population of the first type to obtain the total number of individuals that are the same as the initial population of the first type The first type of subpopulation;

S1404. Combine the first-type initial population with the first-type sub-population to obtain the first-type mixed population;

S1405. Input each individual in the first type of mixed population into the first current proxy model to obtain a predicted value corresponding to each individual in the first type of mixed population, and according to the first type The predicted value corresponding to each individual in the quasi-mixed population obtains the uncertainty value corresponding to each individual;

S1406: Sort each individual in the first-type mixed population in ascending order according to the corresponding uncertainty value to obtain the first-type sorted mixed population;

S1407. According to a preset grouping quantity Q, the first-type sorted mixed populations are equally divided to obtain Q groups of first-type sub-mixed populations; where Q=Ng;

S1408. Obtain the individuals with the smallest uncertainty value in each group of sub-mixed populations of the first type of Q group to form the current population of the first type, and use the current population of the first type as the first type Initial population

S1409. Add one to the current iterative algebra of the first type as the current iterative algebra of the first type, and return to step S1402;

S1410. If the current iteration algebra of the first type reaches the maximum iteration algebra, the initial population of the first type is taken as the final population of the first type, and the individual with the smallest predicted value in the final population of the first type is obtained as As the first target individual, and obtain the individual with the largest uncertainty value in the first type of final population as the second target individual.

In this embodiment, since the vector feature dimension of the decision variable in the initial data point set is known, in order to generate the feature dimension of multiple individuals the same as the vector feature dimension of the decision variable, you can refer to the decision The vector eigendimensions of the variables, Ng variable solutions are randomly generated through the Latin hypercube design (where the value of Ng is a positive integer).

Among them, the Latin hypercube design is Latin hypercube sampling (that is, m samples are sampled in the n-dimensional vector space). In this way, multiple variable solutions can be randomly generated to form the first type of initial population. Among them, each variable solution is an individual in the first type of initial population, and the characteristic dimension of each variable solution is the same as the characteristic dimension of the decision variable. The purpose of randomly generating multiple variable solutions through initialization is to generate better target individuals for evolution. After the initial population of the first type is obtained, steps S1402-S1409 may be performed multiple iterations until the current iteration algebra of the first type reaches the maximum iteration algebra, and the initial population of the first type after multiple iterations is taken as the The first type of final population, the individual with the smallest predicted value in the first type of final population is obtained as the first target individual, and the individual with the largest uncertainty value in the first type of final population is obtained as the second target individual .

In the above population evolution process, simulated binary crossover and polynomial mutation are used. The first type of subpopulation is generated by randomly selecting two individuals from the first type of current population to simulate binary crossover until the crossover obtains Ng According to the mutation probability and polynomial mutation, Ng new individuals of the first type are mutated to obtain the first type of new individuals after Ng polynomial mutations. After these Ng polynomials are mutated, the first type of new individuals form the first type. Subpopulation. Here, the process of randomly selecting two individuals in the current population of the first type for binary crossover is also similar to an iterative process. Until the number of new individuals reaches the population size Ng corresponding to the initial population of the first type, the above multiple times are stopped. The process of binary interleaving. In addition, binary crossover and polynomial mutation are both conventional processing procedures, and will not be repeated here.

After the first type of subpopulation is obtained according to the first type of initial population, and the two are mixed to obtain the first type of mixed population (the total number of individuals in the first type of mixed population is 2Ng), then the first type of subpopulation Each individual in the first type of mixed population is input to the first current proxy model to obtain a predicted value corresponding to each individual in the first type of mixed population, according to each of the first type of mixed population The predicted value corresponding to the individual obtains the uncertainty value corresponding to each individual. Obtaining the predicted value corresponding to each individual in the first type of mixed population and the uncertainty value corresponding to each individual is also convenient to use the above two values as reference parameters for selecting target data points (ie target individuals) value. After that, each individual in the first-type mixed population is sorted in ascending order according to the corresponding uncertainty value to obtain the first-type sorted mixed population. At this time, the first-type sorted mixed population is based on the preset number of groups Q is divided. Since the total number of individuals in the mixed population after sorting of the first type is 2Ng, the number of individuals included in each group of the first type sub-mixed population in the divided Q group is 2Ng/Q. Preferably, if the value of Q is set to Ng, the number of individuals included in each group of the first-type sub-mixed population is 2. At this time, from the 2 individuals included in each group of the first-type sub-mixed population The individuals with the smallest uncertainty value are selected in both, so as to reconstitute the current population of the first type including Ng individuals. At this time, after one iteration of the initial population of the first type, if the current iteration algebra of the first type does not reach the maximum iteration algebra after adding one, steps S1403-S1409 are repeated for multiple times until the current iteration of the first type When the number of generations reaches the maximum number of iterations, the final initial population of the first type is used as the final population of the first type. At this time, the individual with the smallest predicted value in the first type of final population is acquired as the first target individual, and the individual with the largest uncertainty value in the first type of final population is acquired as the second target individual.

In an embodiment, as shown in FIG. 4, in step S1405, each individual in the first type of mixed population is input to the first current agent model to obtain the same value as that in the first type of mixed population. The predicted value corresponding to each individual, including:

S14051. Input each individual in the first type of mixed population to the first Chris King model to obtain a first subtype prediction value corresponding to each individual in the first type of mixed population;

S14052. Input each individual in the first type of mixed population into the first radial basis function model to obtain a second subtype prediction value corresponding to each individual in the first type of mixed population ；

S14053. Input each individual in the first type of mixed population into the first polynomial response surface model to obtain a third subtype prediction corresponding to each individual in the first type of mixed population value;

S14054. Obtain a first weight value corresponding to the first Chriskin model, obtain a second weight value corresponding to the first radial basis function model, and obtain a first weight value corresponding to the first polynomial response surface model. Three-weight value;

S14055. Invoke a pre-stored predicted value weight summation model to obtain a predicted value corresponding to each individual in the mixed population; the predicted value weight summation model is:

in,

Represents the predicted value corresponding to the individual x _{i in the mixed population;}

e ₁ is the root mean square error of the first Chriskin model, e ₂ is the root mean square error of the first radial basis function model, and e ₃ is the mean square error of the first polynomial response surface model. Root square error,

Indicates that the individual x _i corresponds to the predicted value of the first subtype,

Indicates that the individual x _i corresponds to the predicted value of the second subtype,

Indicates that the individual x _i corresponds to the predicted value of the third subtype.

In this embodiment, when calculating the predicted value corresponding to each individual in the first type of mixed population, the individual is first calculated and inputted to the first Chris King model and the first radial basis. After the function model and the polynomial response surface model, respectively correspond to the predicted value of the first subtype, the predicted value of the second subtype, and the predicted value of the third subtype corresponding to the individual. At this time, the predicted value of the first subtype, the predicted value of the second subtype, and the predicted value of the third subtype corresponding to the individual are weighted and summed to obtain the predicted value corresponding to the individual. For the specific calculation process, refer to step S14055. In addition, the weighted summation method can make the predicted value of the individual input to the first current agent model more accurate, which facilitates the subsequent screening of individuals with better performance based on the predicted value as the target individual.

In an embodiment, obtaining the uncertainty value corresponding to each individual according to the predicted value corresponding to each individual in the first-type mixed population in step S1405 includes:

Repeated execution to obtain the predicted value of the first subtype corresponding _{to the individual x i} in the first type of mixed population

Predicted value of the second subtype

And the third sub-type predicted value

The maximum difference between the two is used as the uncertainty value U _ens (x _{i ) corresponding to the individual x i in} the first type of mixed population, until each of the first type of mixed populations is obtained. The uncertainty value corresponding to an individual; among them, the value range of i is [1,2Ng].

In this embodiment, the uncertainty corresponding to each individual in the first type of mixed population is obtained

Certainty value (uncertainty is defined as the maximum difference between two predicted values), the process of obtaining the uncertainty value by other individuals in the first type of mixed population refers to the uncertainty value of the individual x ₁ in the above example The acquisition process. Obtaining the uncertainty value corresponding to each individual in the first type of mixed population is convenient for subsequent selection of individuals with better performance as target individuals based on the uncertainty value.

S150. The data points corresponding to the first target individual and the data points corresponding to the second target individual are both added to the initial data point set to obtain a current data point set; wherein, the data corresponding to the first target individual The points are composed of the first target individual and the first real function value obtained by inputting the first target individual to the pre-stored target function for calculation; the data point corresponding to the second target individual is composed of the second target individual, and the second target individual. The target individual inputs to the target function and is composed of the second real function value obtained correspondingly.

In this embodiment, since each data point in the initial data point set includes a decision variable and an evaluation value corresponding to the decision variable, after acquiring the first target individual and the second target individual, Calculate the first true function value corresponding to the input of the first target individual to the pre-stored target function for operation, and the second true function value corresponding to the input of the second target individual to the target function, and the first target individual and the second target individual A true function value constitutes the data point corresponding to the first target individual, and the second target individual and the second true function value constitute the data point corresponding to the second target individual.

When the data points corresponding to the first target individual and the data points corresponding to the second target individual are added to the initial data point set, if the total number of data points is still less than the maximum number of real evaluations, it is still Step S160 and subsequent steps need to be performed until the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, and the initial data point set is sent to the client. Through steps S140-S150, the first target individual and the second target individual are quickly screened in the first type of final population. The data points corresponding to these two individuals can be added as the two selected data points after this round of iteration. To the initial data point set, the current data point set is obtained.

S160. Obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set, and each data point in the target data point set is taken as the first 2. The training samples of the agent model to be trained are obtained, and the corresponding second current agent model is obtained. According to the second current agent model and the preset second individual screening conditions, the final second type generated according to the genetic evolution of the second type initial population The population searches to get the third target individual.

In this embodiment, after the first round of adding points to the initial data point set is completed through a global search, at this time, the second round of data points to the current data point set is completed through a local search. Click the process of adding points to get the final set of data points after the end of this round of iterative process.

In an embodiment, as shown in FIG. 5, step S160 includes:

S1601, call a pre-stored second radial basis function model to be trained as the second current agent model;

S1602, using the decision variable corresponding to each data point in the target data point set as the input of the second radial basis function model to be trained, and using the evaluation value corresponding to each decision variable as the second radial basis function to be trained An output of a function model, training the second radial basis function model to be trained to obtain a second radial basis function model, so as to use the second radial basis function model as the second current proxy model;

S1603. Obtain a decision variable corresponding to each data point in the target data point set to form a second type of initial population; wherein, the decision variable corresponding to each data point in the target data point set is the same as the second type of initial population. Corresponding to an individual in the population;

S1604. Obtain the current iteration algebra of the second type, and determine whether the current iteration algebra of the second type reaches a preset maximum iteration algebra; wherein the initial value of the current iteration algebra of the second type is 1.

S1605. If the current iteration algebra of the second type does not reach the maximum iteration algebra, perform simulated binary crossover and polynomial mutation on the initial population of the second type to obtain the total number of individuals that are the same as the initial population of the second type The second type of subpopulation;

S1606. Combine the initial population of the second type with the sub-population of the second type to obtain a mixed population of the second type.

S1607: Input each individual in the second type of mixed population into the second radial basis function model to obtain a predicted value corresponding to each individual in the second type of mixed population;

S1608. Sort each individual in the second type of mixed population in ascending order according to the corresponding predicted value to obtain the second type of mixed population after sorting;

S1609. Obtain individuals in the second-type mixed population that are sorted before the ranking threshold to form a second-type current population, and use the second-type current population as the second-type initial population;

S1610: Add one to the current iteration algebra of the second type as the current iteration algebra of the second type, and return to step S1604;

S1611. If the current iteration algebra of the second type reaches the maximum iteration algebra, use the initial population of the second type as the final population of the second type, and obtain the individual with the smallest predicted value in the final population of the second type. As the third target individual.

In this embodiment, a third target individual that satisfies the second individual screening condition is searched for in the current data point set by means of a local search. In the specific process of obtaining the third target individual, first obtain the data points in the current data point set in ascending order according to the true function value and sort the data points before the preset ranking threshold to form the target data point set, and then The decision variables corresponding to each data point in the target data point set are used to form the second type of initial population. After the initial population of the second type is obtained, steps S1604-S1611 may be performed multiple iterations until the current iteration algebra of the second type reaches the maximum iteration algebra, and the initial population of the second type after multiple iterations is taken as the first iteration. The final population of the second type, and the individual with the smallest predicted value in the final population of the second type is obtained as the third target individual.

In an embodiment, step S1605 includes:

Randomly select two individuals in the second type of initial population to perform binary crossover in sequence until M new individuals of the second type after crossover processing are generated, and after M crossover processing, the second type of new individuals are subjected to polynomial mutation, and the polynomial The mutated second-type new individuals form the second-type subpopulation; where M=the ranking threshold -1.

Among them, in the above-mentioned population evolution process, simulated binary crossover and polynomial mutation are used. The second type of subpopulation is generated by randomly selecting two individuals from the second type of current population to simulate binary crossover until the crossover obtains M After crossover processing, the second type of new individuals (where the value of M is a positive integer), and then according to the mutation probability and polynomial mutation, the M second type of new individuals after crossover processing are mutated, and the second type after M polynomial mutation is obtained. Class new individuals, after these M polynomials mutate, the second class new individuals form the second class subpopulation. Here, the process of randomly selecting two individuals in the current population of the second type for binary crossover is also similar to an iterative process, until the number of new individuals reaches the second population size M corresponding to the initial population of the second type, the above process is stopped. The process of multiple binary crossovers.

After the second type of subpopulation is obtained from the second type of initial population, and the two are mixed to obtain the second type of mixed population (the total number of individuals in the second type of mixed population is 2M), then the first Each individual in the second-type mixed population is input to the second current proxy model, and a predicted value corresponding to each individual in the second-type mixed population is obtained. Obtaining the predicted value corresponding to each individual in the second type of mixed population is also convenient to use the predicted value as a reference parameter value for selecting a target data point (that is, a target individual). After that, each individual in the second-type mixed population is sorted in ascending order according to the corresponding predicted value to obtain the second-type sorted mixed population. At this time, the ranking in the second-type sorted mixed population is obtained at the ranking threshold. The previous individuals form the current population of the second type, thereby recomposing the current population of the second type including M individuals. At this time, after one iteration of the initial population of the second type, if the current iteration algebra of the second type plus one does not reach the maximum iteration algebra, steps S1604-S1611 are repeated for multiple times until the current iteration of the second type When the number of generations reaches the maximum number of iterations, the final initial population of the second type is used as the final population of the second type. At this time, the individual with the smallest predicted value in the second type of final population is acquired as the third target individual.

S170. Add the data point corresponding to the third target individual to the current data point set to obtain a final data point set, use the final data point set as the initial data point set, and return to step S130; wherein, the third target The data point corresponding to the individual is composed of the third target individual and the third true function value obtained by inputting the third target individual to the target function.

In this embodiment, when the data point corresponding to the third target individual is added to the current data point set, if the total number of data points is still less than the maximum number of true evaluations, then it is necessary to perform the step to return to execution. Step S120 and subsequent steps, until the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, the initial data point set is sent to the client. Through steps S160-S170, the third target individual can be quickly screened in the second type of final population. The data point corresponding to this third target individual can be used as a selected data point after this round of iteration and added to the current Data point collection, the final data point collection is obtained, and the final data point collection is updated as a new initial data point collection after the end of this round of iteration, and step S120 is returned to.

S180: If the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, send the initial data point set to the client.

In this embodiment, after obtaining the initial data point set in the server, it can be sent to the client. The client can further optimize the wing shape based on a larger number of data points in the initial data point set of the final state after multiple iterations.

This method realizes the combination of agent-assisted and evolutionary algorithms and considers the predicted value and uncertainty of the agent model at the same time. It quickly increases data points in a limited data sample, and the increased sample points have a significant impact on the accuracy of the agent model. improve.

The embodiment of the present application also provides an airfoil optimization device based on the agent-assisted evolution algorithm. The airfoil optimization device based on the agent-assisted evolution algorithm is used to execute any embodiment of the aforementioned airfoil optimization method based on the agent-assisted evolution algorithm. Specifically, please refer to FIG. 6, which is a schematic block diagram of an airfoil optimization device based on a proxy-assisted evolution algorithm provided by an embodiment of the present application. The airfoil optimization device 100 based on the agent-assisted evolutionary algorithm can be configured in a server.

As shown in FIG. 6, the airfoil optimization device 100 based on the agent-assisted evolutionary algorithm includes a point addition request detection unit 110, an initial data point set acquisition unit 120, a total number of data point judgment unit 130, a global search unit 140, and the first round of point addition The unit 150, the local search unit 160, the second round adding unit 170, and the collective sending unit 180 after adding the points.

Wherein, the point addition request detection unit 110 is used to determine whether a data point addition request sent by the client is received.

The initial data point set obtaining unit 120 is configured to obtain the initial data point set in the sample library if the data point adding request sent by the client is received, and according to the total number of data points in the initial data point set and the preset optimization The total number of points obtains the maximum true evaluation times; among them, each data point includes the decision variable corresponding to the wing geometric shape control point and the evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector .

The total number of data points judging unit 130 is configured to judge whether the total number of data points in the initial data point set is less than the maximum number of real evaluation times.

The global search unit 140 is configured to use each data point in the initial data point set as a training sample of the first proxy model to be trained if the total number of data points in the initial data point set is less than the maximum number of real evaluations , Obtain the corresponding first current agent model, search according to the first current agent model and preset first individual screening conditions in the first type final population generated according to the first type of initial population genetic evolution, and obtain the first type of final population A target individual and a second target individual.

In an embodiment, the global search unit 140 includes:

The first agent model training unit is configured to use the decision variable of each data point in the initial data point set as the input of the first to-be-trained Chriskin model, and use the evaluation value corresponding to each decision variable as the first Output of the Chris King model to be trained, training the first Chris King model to be trained to obtain the first Chris King model;

The second agent model training unit is configured to use the decision variable of each data point in the initial data point set as the input of the first radial basis function model to be trained, and use the evaluation value corresponding to each decision variable as the first An output of the radial basis function model to be trained, training the first radial basis function model to be trained to obtain the first radial basis function model;

The third agent model training unit is configured to use the decision variable of each data point in the initial data point set as the input of the first polynomial response surface model to be trained, and use the evaluation value corresponding to each decision variable as the first The output of the polynomial response surface model to be trained is trained on the first polynomial response surface model to be trained to obtain the first polynomial response surface model.

In an embodiment, the global search unit 140 further includes:

The first type of initial population generating unit is used to randomly generate Ng variable solutions in a Latin hypercube design according to the vector feature dimensions of the decision variables in the initial data point set to form the first type of initial population; wherein, each The variable solution is an individual in the first type of initial population, and the characteristic dimension of each variable solution is the same as the characteristic dimension of the decision variable;

The first type of current iteration algebra judging unit, used to obtain the first type of current iteration algebra, and determine whether the first type of current iteration algebra reaches the preset maximum iteration algebra; wherein the initial value of the first type of current iteration algebra Is 1;

The first-type population cross-mutation unit is used to simulate binary crossover and polynomial mutation of the first-type initial population if the current iterative algebra of the first-type does not reach the maximum iterative algebra, to obtain the same as the first-type The initial population has the first type of subpopulation with the same total number of individuals;

The first-type mixed population obtaining unit is configured to merge the first-type initial population with the first-type sub-population to obtain the first-type mixed population;

The first-type parameter acquisition unit is configured to input each individual in the first-type mixed population into the first current proxy model to obtain a prediction corresponding to each individual in the first-type mixed population Value, obtaining the uncertainty value corresponding to each individual according to the predicted value corresponding to each individual in the first type of mixed population;

The first-type sorting unit is configured to sort each individual in the first-type mixed population in ascending order according to the corresponding uncertainty value to obtain the first-type sorted mixed population;

The first-type population dividing unit is configured to divide the first-type sorted mixed populations evenly according to the preset grouping quantity Q to obtain the first-type sub-mixed populations of the Q group; where Q=Ng;

The first-type population screening unit is used to obtain the individuals with the smallest uncertainty value in each group of the first-type sub-mixed populations of the Q group to form the first-type current population. The current population of the class is the initial population of the first class;

The first-type current iterative algebra secondary judgment unit, configured to add one to the first-type current iterative algebra as the first-type current iterative algebra, return to execute to obtain the first-type current iterative algebra, and determine the first-type current iterative algebra Whether the current iteration algebra reaches the preset maximum iteration algebra step;

The first-type target individual acquiring unit is configured to, if the current iteration algebra of the first type reaches the maximum iteration algebra, use the first-type initial population as the first-type final population, and obtain the first-type final population The individual with the smallest predicted value in the population is taken as the first target individual, and the individual with the largest uncertainty value in the first type of final population is obtained as the second target individual.

The first round of adding points unit 150 is configured to add the data points corresponding to the first target individual and the data points corresponding to the second target individual to the initial data point set to obtain the current data point set; wherein, the The data point corresponding to the first target individual is composed of the first target individual and the first real function value obtained by inputting the first target individual to the pre-stored target function for operation; the data point corresponding to the second target individual is composed of the first Two target individuals, and the second target individual input to the objective function corresponding to the obtained second real function value composition.

The local search unit 160 is configured to obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set. Each data point is used as the training sample of the second agent model to be trained, and the corresponding second current agent model is obtained. According to the second current agent model and the preset second individual screening conditions, it is generated according to the genetic evolution of the second type of initial population The second type of final population is searched, and the third target individual is obtained.

In an embodiment, the local search unit 160 includes:

A fourth proxy model acquiring unit, configured to call a pre-stored second radial basis function model to be trained as the second current proxy model;

The fourth agent model training unit is configured to use the decision variable corresponding to each data point in the target data point set as the input of the second radial basis function model to be trained, and use the evaluation value corresponding to each decision variable as the The output of the second radial basis function model to be trained, the second radial basis function model to be trained is trained to obtain a second radial basis function model, and the second radial basis function model is used as the The second current agency model;

The second type of initial population generating unit is used to obtain the decision variable corresponding to each data point in the target data point set to form the second type of initial population; wherein, the decision corresponding to each data point in the target data point set The variable corresponds to an individual in the initial population of the second type;

The second type of current iteration algebra judging unit, used to obtain the second type of current iteration algebra, and determine whether the second type of current iteration algebra reaches the preset maximum iteration algebra; wherein, the initial value of the second type of current iteration algebra Is 1;

The second type of population crossover mutation unit is used to simulate binary crossover and polynomial mutation of the second type of initial population if the current iteration algebra of the second type does not reach the maximum iterative algebra to obtain The initial population has the second type of subpopulation with the same total number of individuals;

The second-type mixed population obtaining unit is configured to merge the second-type initial population with the second-type sub-population to obtain the second-type mixed population;

The second-type parameter acquisition unit is used to input each individual in the second-type mixed population into the second radial basis function model to obtain a corresponding to each individual in the second-type mixed population Predicted value;

The second-type sorting unit is configured to sort each individual in the second-type mixed population in ascending order according to the corresponding predicted value to obtain the second-type mixed population after sorting;

The second-type population screening unit is used to obtain individuals in the second-type mixed population after sorting that are sorted before the ranking threshold to form a second-type current population, and the second-type current population is regarded as the second type Initial population

The second type of current iterative algebra secondary judgment unit, used to add one to the second type of current iterative algebra to serve as the second type of current iterative algebra, return to execute the acquisition of the second type of current iterative algebra, and determine the first Steps of whether the current iteration algebra of the second type reaches the preset maximum iteration algebra;

The second type of target individual obtaining unit is configured to, if the current iteration algebra of the second type reaches the maximum iteration algebra, use the initial population of the second type as the final population of the second type, and obtain the final population of the second type. The individual with the smallest predicted value in the population is taken as the third target individual.

The second round of adding points unit 170 is configured to add data points corresponding to the third target individual to the current data point set to obtain a final data point set, use the final data point set as the initial data point set, and return to execute the judgment The step of determining whether the total number of data points in the initial data point set is less than the maximum number of true evaluation times; wherein, the data points corresponding to the third target individual are input to the target by the third target individual and the third target individual The function corresponds to the obtained third real function value composition.

The point-added set sending unit 180 is configured to send the initial data point set to the client if the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times.

The device realizes the combination of agent assistance and evolutionary algorithms and considers the predicted value and uncertainty of the agent model at the same time. It quickly increases data points in a limited data sample, and the increased sample points have a significant impact on the accuracy of the agent model. improve.

The above-mentioned airfoil optimization device based on the agent-assisted evolutionary algorithm may be implemented in the form of a computer program, and the computer program may run on the computer device as shown in FIG.

Please refer to FIG. 7, which is a schematic block diagram of a computer device according to an embodiment of the present application. The computer device 500 is a server, and the server may be an independent server or a server cluster composed of multiple servers.

Referring to FIG. 7, the computer device 500 includes a processor 502, a memory, and a network interface 505 connected through a system bus 501, where the memory may include a non-volatile storage medium 503 and an internal memory 504.

The non-volatile storage medium 503 can store an operating system 5031 and a computer program 5032. When the computer program 5032 is executed, the processor 502 can execute the airfoil optimization method based on the agent-assisted evolution algorithm.

The processor 502 is used to provide calculation and control capabilities, and support the operation of the entire computer device 500.

The internal memory 504 provides an environment for the operation of the computer program 5032 in the non-volatile storage medium 503. When the computer program 5032 is executed by the processor 502, the processor 502 can execute the airfoil optimization method based on the agent-assisted evolution algorithm.

The network interface 505 is used for network communication, such as providing data information transmission. Those skilled in the art can understand that the structure shown in FIG. 7 is only a block diagram of part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device 500 to which the solution of the present application is applied. The specific computer device 500 may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

Wherein, the processor 502 is configured to run a computer program 5032 stored in a memory to implement the airfoil optimization method based on the agent-assisted evolution algorithm disclosed in the embodiment of the present application.

Those skilled in the art can understand that the embodiment of the computer device shown in FIG. 7 does not constitute a limitation on the specific configuration of the computer device. In other embodiments, the computer device may include more or less components than those shown in the figure. Or some parts are combined, or different parts are arranged. For example, in some embodiments, the computer device may only include a memory and a processor. In such an embodiment, the structure and function of the memory and the processor are consistent with the embodiment shown in FIG. 7 and will not be repeated here.

It should be understood that, in this embodiment of the application, the processor 502 may be a central processing unit (Central Processing Unit, CPU), and the processor 502 may also be other general-purpose processors, digital signal processors (Digital Signal Processors, DSPs), Application Specific Integrated Circuit (ASIC), Field-Programmable Gate Array (FPGA) or other programmable logic devices, discrete gates or transistor logic devices, discrete hardware components, etc. Among them, the general-purpose processor may be a microprocessor or the processor may also be any conventional processor.

In another embodiment of the present application, a computer-readable storage medium is provided. The computer-readable storage medium may be a non-volatile computer-readable storage medium. The computer-readable storage medium stores a computer program, where the computer program is executed by a processor to implement the airfoil optimization method based on the agent-assisted evolution algorithm disclosed in the embodiments of the present application.

The computer-readable storage medium is a physical, non-transitory storage medium, such as a U disk, a mobile hard disk, a read-only memory (Read-Only Memory, ROM), a magnetic disk, or an optical disk that can store program codes. Physical storage media.

Those skilled in the art can clearly understand that, for the convenience and conciseness of description, the specific working process of the above-described equipment, device, and unit can refer to the corresponding process in the foregoing method embodiment, which will not be repeated here.

The above are only specific implementations of this application, but the protection scope of this application is not limited to this. Anyone familiar with the technical field can easily think of various equivalents within the technical scope disclosed in this application. Modifications or replacements, these modifications or replacements shall be covered within the protection scope of this application. Therefore, the protection scope of this application shall be subject to the protection scope of the claims.

Claims (10)

1. An airfoil optimization method based on agent-assisted evolutionary algorithm, which includes:

   Determine whether the data point adding request sent by the client is received;

   If a data point addition request sent by the client is received, the initial data point set in the sample library is obtained, and the maximum number of real evaluations is obtained according to the total number of data points in the initial data point set and the total number of optimized points set in advance; Among them, each data point includes a decision variable corresponding to the wing geometric shape control point and an evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector;

   Judging whether the total number of data points in the initial data point set is less than the maximum number of real evaluation times;

   If the total number of data points in the initial data point set is less than the maximum number of real evaluations, each data point in the initial data point set is used as the training sample of the first agent model to be trained to obtain the corresponding first current The agent model, according to the first current agent model and preset first individual screening conditions, searches the first type final population generated according to the first type of initial population genetic evolution to obtain the first target individual and the second target individual;

   The data points corresponding to the first target individual and the data points corresponding to the second target individual are both added to the initial data point set to obtain the current data point set; wherein the data point corresponding to the first target individual is determined by The first target individual and the first target individual are input to the pre-stored target function for calculation and are composed of the corresponding first true function value; the data point corresponding to the second target individual is composed of the second target individual and the second target individual Input to the objective function corresponding to the second real function value composition;

   Obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set, and each data point in the target data point set is used as the second target data point set. Train the training samples of the agent model to obtain the corresponding second current agent model, and perform the process on the second type final population generated based on the genetic evolution of the second type initial population according to the second current agent model and preset second individual screening conditions Search to get the third target individual;

   The data point corresponding to the third target individual is added to the current data point set to obtain the final data point set, the final data point set is used as the initial data point set, and the execution is returned to determine the total number of data points in the initial data point set. The step of whether the number is less than the maximum number of real evaluation times; wherein the data point corresponding to the third target individual is input by the third target individual and the third target individual to the third true function value corresponding to the target function Composition; and

   If the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, the initial data point set is sent to the client.
2. The airfoil optimization method based on the agent-assisted evolutionary algorithm according to claim 1, wherein the first agent model to be trained includes a first Chriskin model to be trained, a first radial basis function model to be trained, and a first A polynomial response surface model to be trained; the first current agent model includes a first Chriskin model, a first radial basis function model, and a first polynomial response surface model;

   The step of using each data point in the initial data point set as a training sample of the first agent model to be trained to obtain the corresponding first current agent model includes:

   Using the decision variable of each data point in the initial data point set as the input of the first Chriskin model to be trained, and the evaluation value corresponding to each decision variable as the output of the first Chriskin model to be trained, Training the first Chris King model to be trained to obtain the first Chris King model;

   The decision variable of each data point in the initial data point set is used as the input of the first radial basis function model to be trained, and the evaluation value corresponding to each decision variable is used as the first radial basis function model to be trained. Output, training the first radial basis function model to be trained to obtain a first radial basis function model;

   Using the decision variable of each data point in the initial data point set as the input of the first polynomial response surface model to be trained, and the evaluation value corresponding to each decision variable as the output of the first polynomial response surface model to be trained, The first polynomial response surface model to be trained is trained to obtain the first polynomial response surface model.
3. The airfoil optimization method based on the agent-assisted evolution algorithm according to claim 2, wherein the first current agent model and the preset first individual screening condition are generated according to the first type of initial population genetic evolution. The first type of final population is searched, and the first target individual and the second target individual obtained include:

   According to the vector feature dimensions of the decision variables in the initial data point set, Ng variable solutions are randomly generated in a Latin hypercube design to form the first type of initial population; wherein, each variable solution is the first type of initial population For an individual, the characteristic dimension of the solution of each variable is the same as the characteristic dimension of the decision variable;

   Acquire the current iterative algebra of the first type, and determine whether the current iterative algebra of the first type reaches the preset maximum iterative algebra; wherein, the initial value of the current iterative algebra of the first type is 1;

   If the current iteration algebra of the first type does not reach the maximum iteration algebra, perform simulated binary crossover and polynomial mutation on the initial population of the first type to obtain the first population that has the same total number of individuals as the initial population of the first type. One type of subpopulation;

   Combining the first-type initial population with the first-type sub-population to obtain the first-type mixed population;

   Each individual in the first type of mixed population is input into the first current proxy model to obtain a predicted value corresponding to each individual in the first type of mixed population, and the prediction value corresponding to each individual in the first type of mixed population is obtained according to the first type of mixed population. The predicted value corresponding to each individual in the population obtains the uncertainty value corresponding to each individual;

   Sort each individual in the first-type mixed population in ascending order according to the corresponding uncertainty value to obtain the first-type sorted mixed population;

   According to the preset grouping quantity Q, divide the mixed populations of the first type sorted evenly to obtain the first type sub-mixed populations of group Q; where Q=Ng;

   Obtain the individuals with the smallest uncertainty value in each group of sub-mixed populations of the Q group of the first-type sub-mixed populations to form the first-type current population, and use the first-type current population as the first-type initial population ；

   Adding one to the current iterative algebra of the first type as the current iterative algebra of the first type, and return to execute the step of judging whether the current iterative algebra of the first type reaches the preset maximum iterative algebra;

   If the current iteration algebra of the first type reaches the maximum iteration algebra, the initial population of the first type is taken as the final population of the first type, and the individual with the smallest predicted value in the final population of the first type is obtained as the first A target individual, and obtain the individual with the largest uncertainty value in the first type of final population as the second target individual.
4. The airfoil optimization method based on the agent-assisted evolutionary algorithm according to claim 3, wherein each individual in the first type of mixed population is input into the first current agent model to obtain The predicted value corresponding to each individual in the first type of mixed population includes:

   Inputting each individual in the first type of mixed population into the first Chris King model to obtain a first subtype prediction value corresponding to each individual in the first type of mixed population;

   Inputting each individual in the first type of mixed population into the first radial basis function model to obtain a second subtype prediction value corresponding to each individual in the first type of mixed population;

   Inputting each individual in the first type of mixed population into the first polynomial response surface model to obtain a third subtype prediction value corresponding to each individual in the first type of mixed population;

   Obtain the first weight value corresponding to the first Chriskin model, obtain the second weight value corresponding to the first radial basis function model, and obtain the third weight value corresponding to the first polynomial response surface model value;

   The pre-stored prediction value weight summation model is called to obtain the prediction value corresponding to each individual in the mixed population; the prediction value weight summation model is:

   

   in,

   

   Represents the predicted value corresponding to the individual x _{i in the mixed population;}

   

   e ₁ is the root mean square error of the first Chriskin model, e ₂ is the root mean square error of the first radial basis function model, and e ₃ is the mean square error of the first polynomial response surface model. Root square error,

   

   Indicates that the individual x _i corresponds to the predicted value of the first subtype,

   

   Indicates that the individual x _i corresponds to the predicted value of the second subtype,

   

   Indicates that the individual x _i corresponds to the predicted value of the third subtype.
5. The airfoil optimization method based on the agent-assisted evolutionary algorithm according to claim 4, wherein said obtaining the uncertainty value corresponding to each individual according to the predicted value corresponding to each individual in the first type of mixed population, include:

   Repeated execution to obtain the predicted value of the first subtype corresponding _{to the individual x i} in the first type of mixed population

   

   Predicted value of the second subtype

   

   And the third sub-type predicted value

   

   The maximum difference between the two is used as the uncertainty value U _ens (x _{i ) corresponding to the individual x i in} the first type of mixed population, until each of the first type of mixed populations is obtained. The uncertainty value corresponding to an individual; among them, the value range of i is [1, 2Ng].
6. The airfoil optimization method based on the agent-assisted evolutionary algorithm according to claim 3, wherein each data point in the target data point set is used as a training sample of the second agent model to be trained to obtain the corresponding second current agent model , According to the second current agent model and the preset second individual screening conditions, searching for the second type final population generated according to the genetic evolution of the second type initial population to obtain the third target individual, including:

   Calling a pre-stored second radial basis function model to be trained as the second current agent model;

   The decision variable corresponding to each data point in the target data point set is used as the input of the second radial basis function model to be trained, and the evaluation value corresponding to each decision variable is used as the second radial basis function model to be trained To train the second radial basis function model to be trained to obtain a second radial basis function model, so as to use the second radial basis function model as the second current agent model;

   The decision variable corresponding to each data point in the target data point set is obtained to form a second type of initial population; wherein, the decision variable corresponding to each data point in the target data point set is the same as that in the second type of initial population Corresponds to an individual of;

   Acquire the current iteration algebra of the second type, and determine whether the current iteration algebra of the second type reaches the preset maximum iteration algebra; wherein, the initial value of the current iteration algebra of the second type is 1;

   If the current iteration algebra of the second type does not reach the maximum iteration algebra, perform simulated binary crossover and polynomial mutation on the initial population of the second type to obtain the first population that has the same total number of individuals as the initial population of the second type. The second type of subpopulation;

   Combining the initial population of the second type with the sub-population of the second type to obtain a mixed population of the second type;

   Inputting each individual in the second type of mixed population into the second radial basis function model to obtain a predicted value corresponding to each individual in the second type of mixed population;

   Sort each individual in the second-type mixed population in ascending order according to the corresponding predicted value to obtain the second-type mixed population after sorting;

   Acquiring individuals in the second-type mixed population after the ranking that are ranked before the ranking threshold to form a second-type current population, and the second-type current population is used as the second-type initial population;

   Adding one to the current iterative algebra of the second type as the current iterative algebra of the second type, and return to execute the step of judging whether the current iterative algebra of the second type reaches the preset maximum iterative algebra;

   If the current iteration algebra of the second type reaches the maximum iteration algebra, the initial population of the second type is taken as the final population of the second type, and the individual with the smallest predicted value in the final population of the second type is obtained as the first Three target individuals.
7. The airfoil optimization method based on the agent-assisted evolution algorithm according to claim 6, wherein the simulated binary crossover and polynomial mutation are performed on the second type of initial population to obtain the same individuals as the second type of initial population The total number of subpopulations of the second category includes:

   Randomly select two individuals in the second type of initial population to perform binary crossover in sequence until M new individuals of the second type after crossover processing are generated, and after M crossover processing, the second type of new individuals are subjected to polynomial mutation, and the polynomial The mutated second-type new individuals form the second-type subpopulation; where M=the ranking threshold -1.
8. An airfoil optimization device based on agent-assisted evolutionary algorithm, which includes:

   The point addition request detection unit is used to determine whether the data point addition request sent by the client is received;

   The initial data point set acquisition unit is used to obtain the initial data point set in the sample library if the data point addition request sent by the client is received, and the total number of data points in the initial data point set and the preset optimization point The total number obtains the maximum number of true evaluations; among them, each data point includes the decision variable corresponding to the wing geometric shape control point and the evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector;

   A total number of data points judging unit, configured to judge whether the total number of data points in the initial data point set is less than the maximum number of true evaluation times;

   The global search unit is configured to use each data point in the initial data point set as a training sample of the first proxy model to be trained if the total number of data points in the initial data point set is less than the maximum number of real evaluation times, Obtain the corresponding first current agent model, search according to the first current agent model and preset first individual screening conditions in the first type final population generated according to the genetic evolution of the first type initial population, and obtain the first Target individual and second target individual;

   The first round of adding points unit is used to add the data points corresponding to the first target individual and the data points corresponding to the second target individual to the initial data point set to obtain the current data point set; wherein The data point corresponding to a target individual is composed of the first target individual and the first real function value obtained by inputting the first target individual to the pre-stored target function and performing calculation; the data point corresponding to the second target individual is composed of the second The target individual and the second target individual input to the target function corresponding to the second real function value composition;

   The local search unit is used to obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set, and each data point in the target data point set is The data points are used as the training samples of the second to-be-trained agent model to obtain the corresponding second current agent model. According to the second current agent model and the preset second individual screening conditions, the data points are generated according to the genetic evolution of the second type of initial population The second type of final population is searched and the third target individual is obtained;

   The second round of adding points unit is used to add the data points corresponding to the third target individual to the current data point set to obtain the final data point set, use the final data point set as the initial data point set, and return to execute the judgment of the initial data point set. The step of determining whether the total number of data points in the data point set is less than the maximum number of true evaluation times; wherein, the data point corresponding to the third target individual is input to the target function by the third target individual and the third target individual Correspondingly obtained third real function value composition; and

   The point-added collection sending unit is configured to send the initial data point collection to the client if the total number of data points in the initial data point collection is greater than or equal to the maximum number of real evaluation times.
9. A computer device includes a memory, a processor, and a computer program stored on the memory and capable of running on the processor, wherein the processor implements the following steps when the processor executes the computer program:

   Determine whether the data point adding request sent by the client is received;

   If a data point addition request sent by the client is received, the initial data point set in the sample library is obtained, and the maximum number of real evaluations is obtained according to the total number of data points in the initial data point set and the total number of optimized points set in advance; Among them, each data point includes a decision variable corresponding to the wing geometric shape control point and an evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector;

   Judging whether the total number of data points in the initial data point set is less than the maximum number of real evaluation times;

   If the total number of data points in the initial data point set is less than the maximum number of real evaluations, each data point in the initial data point set is used as the training sample of the first agent model to be trained to obtain the corresponding first current The agent model, according to the first current agent model and preset first individual screening conditions, searches the first type final population generated according to the first type of initial population genetic evolution to obtain the first target individual and the second target individual;

   The data points corresponding to the first target individual and the data points corresponding to the second target individual are both added to the initial data point set to obtain the current data point set; wherein the data point corresponding to the first target individual is determined by The first target individual and the first target individual are input to the pre-stored target function for calculation and are composed of the corresponding first true function value; the data point corresponding to the second target individual is composed of the second target individual and the second target individual Input to the objective function corresponding to the second real function value composition;

   Obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set, and each data point in the target data point set is used as the second target data point set. Train the training samples of the agent model to obtain the corresponding second current agent model, and perform the process on the second type final population generated based on the genetic evolution of the second type initial population according to the second current agent model and preset second individual screening conditions Search to get the third target individual;

   The data point corresponding to the third target individual is added to the current data point set to obtain the final data point set, the final data point set is used as the initial data point set, and the execution is returned to determine the total number of data points in the initial data point set. The step of whether the number is less than the maximum number of real evaluation times; wherein the data point corresponding to the third target individual is input by the third target individual and the third target individual to the third true function value corresponding to the target function Composition; and

   If the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, the initial data point set is sent to the client.
10. A computer-readable storage medium, wherein the computer-readable storage medium stores a computer program that, when executed by a processor, causes the processor to perform the following operations:

    Determine whether the data point adding request sent by the client is received;

    If a data point addition request sent by the client is received, the initial data point set in the sample library is obtained, and the maximum number of real evaluations is obtained according to the total number of data points in the initial data point set and the total number of optimized points set in advance; Among them, each data point includes a decision variable corresponding to the wing geometric shape control point and an evaluation value corresponding to the decision variable, and each decision variable is an n-dimensional row vector or an n-dimensional column vector;

    Judging whether the total number of data points in the initial data point set is less than the maximum number of real evaluation times;

    If the total number of data points in the initial data point set is less than the maximum number of real evaluations, each data point in the initial data point set is used as the training sample of the first agent model to be trained to obtain the corresponding first current The agent model, according to the first current agent model and preset first individual screening conditions, searches the first type final population generated according to the first type of initial population genetic evolution to obtain the first target individual and the second target individual;

    The data points corresponding to the first target individual and the data points corresponding to the second target individual are both added to the initial data point set to obtain the current data point set; wherein the data point corresponding to the first target individual is determined by The first target individual and the first target individual are input to the pre-stored target function for calculation and are composed of the corresponding first true function value; the data point corresponding to the second target individual is composed of the second target individual and the second target individual Input into the second real function value composition obtained corresponding to the objective function;

    Obtain the data points in the current data point set that are sorted in ascending order according to the true function value and sorted before the preset ranking threshold to form a target data point set, and each data point in the target data point set is used as the second target data point set. Train the training samples of the agent model to obtain the corresponding second current agent model, and perform the process on the second type final population generated based on the genetic evolution of the second type initial population according to the second current agent model and preset second individual screening conditions Search to get the third target individual;

    The data point corresponding to the third target individual is added to the current data point set to obtain the final data point set, the final data point set is used as the initial data point set, and the execution is returned to determine the total number of data points in the initial data point set. The step of whether the number is less than the maximum number of real evaluation times; wherein the data point corresponding to the third target individual is input by the third target individual and the third target individual to the third true function value corresponding to the target function Composition; and

    If the total number of data points in the initial data point set is greater than or equal to the maximum number of real evaluation times, the initial data point set is sent to the client.

Images