An airfoil generation and intelligent optimization design method and device based on natural language interaction, equipment and medium

Abstract

The application discloses a natural language interaction-based airfoil generation and intelligent optimization design method and device, equipment and medium, and relates to the technical field of computers, which comprises the following steps: determining structured parameters based on a BERT model and input language; performing multi-scale analysis on initial airfoil parameters determined by an encoder and a decoder in a generator based on a VAE-GAN model to obtain a new generator and a new discriminator, iteratively obtaining a target VAE-GAN model, and determining whether a semantic analysis confidence meets a preset high confidence condition; if the condition is met and the instruction type is an airfoil generation and aerodynamic analysis type, generating an airfoil cluster by using the target VAE-GAN model; and if the condition is met and the instruction type is an airfoil intelligent optimization design type, optimizing a candidate airfoil to obtain an optimized airfoil. The airfoil can be generated without accurately inputting airfoil parameters, and the airfoil can be optimized.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of computer technology, and in particular to a wing profile generation and intelligent optimization design method and device based on natural language interaction, equipment and medium. BACKGROUND

[0002] Wing profile design is one of the core and key technologies in the field of aircraft design, and its importance is reflected in its direct impact on the aerodynamic performance, structural strength and economy of the aircraft. Excellent wing profile design can achieve the best balance between lift and drag in different flight states, thereby significantly improving the cruising efficiency and fuel economy of the aircraft; at the same time, reasonable wing profile shape can also effectively optimize the aerodynamic characteristics, reduce noise pollution and reduce the influence of aerodynamic load on the aircraft structure. In addition, wing profile design also relates to the airworthiness and safety of the aircraft, and needs to meet the performance requirements such as stall resistance and flutter resistance in complex flight environments. Under the background of the continuous development of modern aviation technology, wing profile design is not only an important guarantee for efficient and environmentally friendly flight, but also an important driving force for the innovative development of future aircraft.

[0003] The existing wing profile design optimization methods are generally divided into three categories: gradient-based optimization design method, surrogate optimization design method and machine learning-based optimization design method. The first two methods are more professional, and need to call CFD (Computational Fluid Dynamics) calculation multiple times in the optimization design process, which is relatively high in cost. In the machine learning optimization method, the traditional machine learning method uses supervised learning to establish an end-to-end mapping between the input (geometric shape) and the output (performance parameter), but only through data-driven modeling, lacking the fusion of aerodynamic physical laws, resulting in weak result interpretability; while the generative method learns the latent distribution and cross-domain association by modeling the generation process of wing profile data, and implicitly embeds physical laws while generating controllable parameters, significantly improving the interpretability and cross-scene adaptability, providing a new path for intelligent optimization.

[0004] Furthermore, most mainstream airfoil design methods currently require users to have a deep knowledge base. Existing tools' human-computer interfaces (such as XFOIL GUI and ANSYS Fluent scripting systems) require users to input precise parameters (such as Reynolds number, Mach number, lift-to-drag ratio, etc.) through complex graphical interfaces or scripts. This objectively creates a barrier to verifying cross-disciplinary design requirements. Even if concept designers or product managers have clear innovative needs, they cannot bypass the parameterized input process to directly convey their design intent. The fundamental contradiction lies in the fact that traditional methods compress the design process into a parameter iteration loop, while the requirements themselves are not effectively decoupled. Although vague requirements such as "high lift-to-drag ratio" or "helicopter rotor blades" can be transformed into explicit optimization objectives, existing toolchains lack an automatic mapping mechanism from natural language to parameter space, forcing users to perform multiple rounds of debugging in a closed loop of "requirements → parameters → performance simulation → parameter correction." This parameter-driven rather than requirement-driven design process significantly increases labor costs and development cycles.

[0005] As can be seen from the above, how to generate and optimize airfoils without precisely inputting airfoil parameters is a problem that urgently needs to be solved. Summary of the Invention

[0006] In view of this, the purpose of this invention is to provide a method, apparatus, device, and medium for airfoil generation and intelligent optimization design based on natural language interaction, which can generate and optimize airfoils without requiring precise input of airfoil parameters. The specific solution is as follows:

[0007] Firstly, this application provides a method for airfoil generation and intelligent optimization design based on natural language interaction, including:

[0008] The user's input language is converted into a semantic vector based on the BERT model. The structured parameters are determined based on the weights corresponding to the semantic vectors and the professional terms in the airfoil field. The structured parameters include command type, airfoil type, operating condition parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence.

[0009] The different types of airfoil shapes are converted into digital parameters by using the CST parameterization method, a training set is determined based on the digital parameters, a latent vector is determined based on the training set and by using an encoder in a generator of a VAE-GAN model, initial airfoil parameters are determined based on the latent vector and airfoil domain labels corresponding to different types of airfoils and by using a decoder in the generator, the initial airfoil parameters are subjected to multi-scale analysis by using a discriminator of the VAE-GAN model, and the generator and the discriminator are optimized based on the multi-scale analysis results, respectively, to obtain a new generator and a new discriminator, then jumping to the step of determining the latent vector based on the training set and by using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet a preset accuracy condition, to obtain a target VAE-GAN model, and it is judged whether the semantic analysis confidence meets a preset high confidence condition;

[0010] If yes, and the instruction type is an airfoil generation and aerodynamic analysis type, then an airfoil cluster is generated based on target airfoil domain labels corresponding to the airfoil type and by using the target VAE-GAN model, geometric characteristic analysis and aerodynamic characteristic analysis are respectively performed on each candidate airfoil in the airfoil cluster, and first analysis results and second analysis results are displayed to a preset visualization interface;

[0011] If yes, and the instruction type is an airfoil intelligent optimization design type, then a candidate airfoil is determined by using the target VAE-GAN model, and the candidate airfoil is optimized by using the optimization target and the constraint condition, to obtain an optimized airfoil.

[0012] Optionally, the input language of the user end is converted into a semantic vector based on the BERT model, and a structured parameter is determined based on the semantic vector and weights of airfoil domain professional terms; the structured parameter includes an instruction type, an airfoil type, a working condition parameter, an optimization target and a constraint condition, a design result quantity and a semantic analysis confidence, including:

[0013] The input language of the user end is converted into a semantic vector based on the BERT model, and a semantic matching degree between the semantic vector and airfoil domain professional terms is determined by using cosine similarity;

[0014] Different airfoil domain professional terms are subjected to weight allocation by using a preset airfoil domain term library, to obtain domain term weights;

[0015] A target matching degree between the input language and the airfoil domain professional terms is determined based on the semantic matching degree and the domain term weights, and a structured parameter is determined based on the target matching degree and the input language.

[0016] Optionally, the step of converting different types of airfoil shapes into digital parameters using the CST parameterization method, determining a training set based on the digital parameters, determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, determining initial airfoil parameters based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils and using the decoder in the generator, performing multi-scale analysis on the initial airfoil parameters using the discriminator of the VAE-GAN model, and optimizing the generator and the discriminator based on the multi-scale analysis results to obtain new generators and new discriminators, and then jumping to the step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet the preset accuracy conditions to obtain the target VAE-GAN model, includes:

[0017] The CST parameterization method is used to convert different types of airfoil shapes into digital parameters, and the digital parameters are normalized to obtain the processed parameters.

[0018] The processed parameters are divided using stratified sampling to obtain corresponding training, validation, and test sets.

[0019] Based on the training set, the encoder in the generator of the VAE-GAN model is used to map the processed parameters and the airfoil domain labels corresponding to different types of airfoils to the latent space distribution, and the latent vector is determined from the latent space distribution.

[0020] The decoder in the generator is used to reconstruct the latent vector and the airfoil domain label to obtain the initial airfoil parameters;

[0021] The discriminator of the VAE-GAN model is used to perform multi-scale analysis of the global, local and detailed features of the initial airfoil parameters to obtain the corresponding multi-scale analysis results;

[0022] The generator is optimized based on the multi-scale analysis results and the first preset loss function to obtain a new generator; the first preset loss function includes a reconstruction loss function, a KL divergence loss function, and an adversarial loss function.

[0023] The discriminator is optimized using the multi-scale analysis results and the second preset loss function to obtain a new discriminator. Then, the process jumps to the step of determining the latent vectors using the encoder in the generator of the VAE-GAN model based on the training set, until the initial airfoil parameters meet the preset accuracy condition to obtain the trained VAE-GAN model. The second preset loss function includes the WGAN loss function, the gradient penalty loss function, and the cross-entropy loss function.

[0024] The trained VAE-GAN model is validated and tested using the validation set and the test set, respectively, to obtain the target VAE-GAN model.

[0025] Optionally, if the condition is met and the instruction type is airfoil generation and aerodynamic analysis, then based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model, an airfoil cluster is generated. Geometric and aerodynamic characteristic analyses are performed on each candidate airfoil in the airfoil cluster, and the first and second analysis results are displayed on a preset visualization interface, including:

[0026] If the semantic parsing confidence meets the preset high confidence condition, and the instruction type is airfoil generation and aerodynamic analysis, then the target potential vector is determined based on the standard normal distribution. Based on the target airfoil domain label corresponding to the airfoil type and the target potential vector, an airfoil cluster is generated using the target VAE-GAN model. The geometric characteristics of the shape data of each candidate airfoil in the airfoil cluster are analyzed to obtain the first analysis result.

[0027] The aerodynamic characteristics of each airfoil in the airfoil cluster are analyzed using the NeuralFoil tool to obtain a second analysis result. The first analysis result and the second analysis result are then displayed in a scatter plot on a preset visualization interface.

[0028] Optionally, if the condition is met and the instruction type is an airfoil intelligent optimization design type, then the candidate airfoil is determined using the target VAE-GAN model, and the candidate airfoil is optimized using the optimization objective and constraints to obtain the optimized airfoil, including:

[0029] If the semantic parsing confidence level meets the preset high confidence level condition, and the instruction type is the airfoil intelligent optimization design type, then the optimization design objective is determined based on the optimization objective and constraint conditions, and the target potential vector is determined based on the standard normal distribution.

[0030] The target latent vector is decoded using the target VAE-GAN model to obtain the target digital parameters, the target digital parameters are converted into airfoil coordinates, and the candidate airfoil is determined using the airfoil coordinates;

[0031] The candidate airfoil is optimized based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil.

[0032] Optionally, optimizing the candidate airfoil based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil includes:

[0033] If the optimization type corresponding to the optimization design objective is single-objective optimization, then the candidate airfoil is optimized based on the first analysis result and the second analysis result corresponding to the candidate airfoil to obtain the optimized airfoil.

[0034] Optionally, optimizing the candidate airfoil based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil includes:

[0035] If the optimization type corresponding to the optimization design objective is multi-objective optimization, then based on the optimization objective and constraints and the candidate airfoil, the Pareto optimal solution set is determined using a multi-objective optimization algorithm, and the optimized airfoil is determined using the Pareto optimal solution set.

[0036] Secondly, this application provides an airfoil generation and intelligent optimization design device based on natural language interaction, comprising:

[0037] The parameter determination module is used to convert the user's input language into semantic vectors based on the BERT model, and determine structured parameters based on the weights corresponding to the semantic vectors and airfoil domain terminology. The structured parameters include command type, airfoil type, operating parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence.

[0038] The target model determination module is used to convert different types of airfoil shapes into digital parameters using the CST parameterization method, determine a training set based on the digital parameters, determine latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, determine initial airfoil parameters based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils and using the decoder in the generator, perform multi-scale analysis on the initial airfoil parameters using the discriminator of the VAE-GAN model, and optimize the generator and the discriminator based on the multi-scale analysis results to obtain a new generator and a new discriminator. Then, it jumps to the step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet the preset accuracy condition to obtain the target VAE-GAN model, and determines whether the semantic parsing confidence meets the preset high confidence condition.

[0039] The analysis results display module is used to generate an airfoil cluster based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model if the condition is met and the instruction type is airfoil generation and aerodynamic analysis, perform geometric characteristic analysis and aerodynamic characteristic analysis on each candidate airfoil in the airfoil cluster respectively, and display the first analysis result and the second analysis result on the preset visualization interface.

[0040] The airfoil optimization module is used to determine candidate airfoils using the target VAE-GAN model if the conditions are met and the instruction type is airfoil intelligent optimization design type, and to optimize the candidate airfoils using the optimization objective and constraints to obtain the optimized airfoil.

[0041] Thirdly, this application provides an electronic device, comprising:

[0042] Memory, used to store computer programs;

[0043] A processor is used to execute the computer program to implement the aforementioned airfoil generation and intelligent optimization design method based on natural language interaction.

[0044] Fourthly, this application provides a computer-readable storage medium for storing a computer program, wherein the computer program, when executed by a processor, implements the aforementioned airfoil generation and intelligent optimization design method based on natural language interaction.

[0045] This application uses the BERT model to convert user-end input language into semantic vectors. Structured parameters are determined based on the weights corresponding to the semantic vectors and airfoil-specific terminology. These structured parameters include command type, airfoil type, operating parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence. The CST parameterization method is used to convert different types of airfoil shapes into digital parameters, and a training set is determined based on these digital parameters. Based on the training set, the encoder in the generator of the VAE-GAN model is used to determine latent vectors. Initial airfoil parameters are determined based on the latent vectors, airfoil domain labels corresponding to different airfoil types, and the decoder in the generator. The discriminator of the VAE-GAN model is used to perform multi-scale analysis on the initial airfoil parameters. Based on the multi-scale analysis results, the generator and discriminator are optimized to obtain new generators and new discriminators. Then, the process jumps to the step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet the preset accuracy condition to obtain the target VAE-GAN model, and determines whether the semantic parsing confidence meets the preset high confidence condition; if it does, and the instruction type is airfoil generation and aerodynamic analysis type, then based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model, an airfoil cluster is generated, and geometric and aerodynamic characteristic analyses are performed on each candidate airfoil in the airfoil cluster, and the first and second analysis results are displayed on the preset visualization interface; if it does, and the instruction type is airfoil intelligent optimization design type, then the target VAE-GAN model is used to determine the candidate airfoil, and the optimization objective and constraint conditions are used to optimize the candidate airfoil to obtain the optimized airfoil.

[0046] As can be seen from the above, this application utilizes the BERT model to convert the natural language input from the user end into semantic vectors, and combines the weights of airfoil domain terms to output structured parameters, converting multiple types of airfoil shapes into numerical parameters to determine the training set. The generator of the VAE-GAN model is used to determine the initial airfoil parameters, and the discriminator of the VAE-GAN model is used to perform multi-scale analysis on the initial airfoil parameters to optimize the generator and discriminator, thereby obtaining the target VAE-GAN model. If the user end command type is airfoil generation and aerodynamic analysis, an airfoil cluster including candidate airfoils can be generated using the target VAE-GAN model and structured parameters. Then, geometric and aerodynamic characteristic analyses are performed on the candidate airfoils, and the analysis results are transmitted to the user end's visualization interface for interface interaction. If the command type is airfoil intelligent optimization design, the target VAE-GAN model is used to determine candidate airfoils, and the optimization objectives and constraints are used to optimize the candidate airfoils to obtain the optimized airfoil. In this way, users do not need to input professional and complex statements. The target VAE-GAN model can generate candidate airfoils that correspond to user needs. It can also be further optimized according to user needs to obtain high-precision optimized airfoils. This not only lowers the professional threshold for airfoil design, but also greatly improves the efficiency and quality of airfoil design. Attached Figure Description

[0047] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on the provided drawings without creative effort.

[0048] Figure 1 This is a flowchart of an airfoil generation and intelligent optimization design method based on natural language interaction disclosed in this application;

[0049] Figure 2 This is a schematic diagram of the initial airfoil parameters disclosed in this application;

[0050] Figure 3 This is a schematic diagram of a rotor airfoil cluster disclosed in this application;

[0051] Figure 4 This is a scatter plot of the maximum thickness and maximum camber of a rotor airfoil cluster disclosed in this application;

[0052] Figure 5 This is a scatter plot of the lift coefficient and drag coefficient of a rotor airfoil cluster disclosed in this application;

[0053] Figure 6 This is a schematic diagram of an optimized airfoil according to a single-objective optimization design disclosed in this application.

[0054] Figure 7 This is a schematic diagram of the objective function variation curve of a single-objective optimization design disclosed in this application;

[0055] Figure 8 This is a schematic diagram of a Pareto optimal solution set disclosed in this application;

[0056] Figure 9 This is a schematic diagram of two optimal airfoils in a Pareto optimal solution set disclosed in this application; Figure 9 (a) is a schematic diagram of an optimal airfoil in the Pareto optimal solution set; Figure 9 (b) is a schematic diagram of another optimal airfoil in the Pareto optimal solution set;

[0057] Figure 10 This is a flowchart of a specific airfoil generation and intelligent optimization design method based on natural language interaction disclosed in this application;

[0058] Figure 11 This is a schematic diagram of generating a rotor airfoil cluster using a target VAE-GAN model disclosed in this application;

[0059] Figure 12 This is a schematic diagram of a single-objective optimization design disclosed in this application;

[0060] Figure 13 This is a schematic diagram of a multi-objective optimization design disclosed in this application;

[0061] Figure 14 This is a schematic diagram of the airfoil generation and intelligent optimization design device based on natural language interaction disclosed in this application;

[0062] Figure 15 This is a structural diagram of an electronic device disclosed in this application. Detailed Implementation

[0063] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0064] Currently, most mainstream airfoil design methods require users to have a deep knowledge base. Existing tools' human-computer interfaces further necessitate complex graphical interfaces or scripts for precise parameter input, objectively creating a barrier to verifying cross-domain design requirements. Even if concept designers or product managers have clear innovative needs, they cannot bypass the parameterized input process to directly convey their design intent. The fundamental contradiction lies in the fact that traditional methods compress the design process into a parameter iteration loop, while the requirements themselves are not effectively decoupled. Existing toolchains lack an automatic mapping mechanism from natural language to parameter space, requiring users to perform multiple rounds of debugging. This parameter-driven rather than requirement-driven design process significantly increases labor costs and development cycles. To address this, this application provides an airfoil generation and intelligent optimization design method based on natural language interaction. User input does not require complex technical statements; the target VAE-GAN model can generate candidate airfoils corresponding to user requirements and can further optimize them based on user needs, resulting in a high-precision optimized airfoil. This lowers the professional threshold for airfoil design while significantly improving its efficiency and quality.

[0065] See Figure 1 As shown in the figure, this invention discloses an airfoil generation and intelligent optimization design method based on natural language interaction, including:

[0066] Step S11: Based on the BERT model, the input language from the user end is converted into a semantic vector. Based on the semantic vector and the weights corresponding to the airfoil domain terminology, the structured parameters are determined. The structured parameters include instruction type, airfoil type, operating parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence.

[0067] In this embodiment, a pre-trained BERT model is used to convert the user's input language and airfoil domain terminology into high-dimensional first and second semantic vectors, respectively. Cosine similarity is used to determine the semantic matching degree between the first and second semantic vectors. A pre-set airfoil domain lexicon with weighted gradients is constructed. The pre-set airfoil domain lexicon is used to assign weights to different airfoil domain terminology based on their importance, thus obtaining domain term weights. For example, "analysis" and "evaluation" have a weight of 5 for airfoil generation, while "optimization" and "customization" have a weight of 5 for intelligent design; low speed has a weight of 5 for the NACA airfoil cluster, rotor has a weight of 5 for the OA209 airfoil cluster, while slow flight has a weight of 3 for the NACA airfoil cluster and 2 for the OA209 airfoil cluster. Based on the product of the semantic matching degree and the airfoil domain term weights, a target matching score between the input language and the airfoil domain terminology is determined. Then, the input language is parsed, and structured parameters are determined based on the parsed information and the target matching score.

[0068] Specifically, the BERT-based model converts the user's input language into semantic vectors, and determines structured parameters based on the weights corresponding to the semantic vectors and airfoil domain terminology. These structured parameters include instruction type, airfoil type, operating parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence. The process includes: converting the user's input language into semantic vectors using the BERT model; determining the semantic matching degree between the semantic vectors and airfoil domain terminology using cosine similarity; assigning weights to different airfoil domain terminology using a pre-defined airfoil domain lexicon to obtain domain terminology weights; determining the target matching degree between the input language and the airfoil domain terminology based on the semantic matching degree and the domain terminology weights; and determining the structured parameters based on the target matching degree and the input language.

[0069] Step S12: Convert different types of airfoil shapes into digital parameters using the CST parameterization method, determine a training set based on the digital parameters, determine latent vectors based on the training set and the encoder in the generator of the VAE-GAN model, determine initial airfoil parameters based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils and the decoder in the generator, perform multi-scale analysis on the initial airfoil parameters using the discriminator of the VAE-GAN model, and optimize the generator and the discriminator based on the multi-scale analysis results to obtain a new generator and a new discriminator. Then, jump to the step of determining latent vectors based on the training set and the encoder in the generator of the VAE-GAN model until the initial airfoil parameters meet the preset accuracy condition to obtain the target VAE-GAN model, and determine whether the semantic parsing confidence meets the preset high confidence condition.

[0070] In this embodiment, the CST (Class function / shape function transformation) parameterization method is used to convert different types of airfoil shapes into computer-processable digital parameters. These digital parameters are then standardized to obtain the processed parameters. In one specific implementation, the airfoil is divided into a training set, a validation set, and a test set in a 7:2:1 ratio; this ratio can be adjusted according to actual needs. Based on the training set, the encoder in the generator of the VAE-GAN model maps the processed parameters and the airfoil domain labels corresponding to different airfoil types to a latent space distribution, outputting the mean and variance. A latent vector is determined from the latent space distribution, and then the decoder in the generator reconstructs the latent vector and the airfoil domain labels to obtain the initial airfoil parameters. Figure 2This embodiment provides a schematic diagram of initial airfoil parameters. Based on the user's input language, the parameters are analyzed and converted into initial airfoil parameters. The VAE-GAN model includes an encoder, decoder, discriminator, and output layer in the generator. The output layer includes a sigmoid activation function. Then, the discriminator of the VAE-GAN model is used to perform multi-scale analysis of global, local, and detailed features of the initial airfoil parameters to analyze the overall shape, intermediate details, and minute features of the airfoil corresponding to the initial airfoil parameters at multiple scales, so as to obtain the corresponding multi-scale analysis results.

[0071] Understandably, after obtaining the multi-scale analysis results, the generator is optimized based on whether the airfoil corresponding to the initial airfoil parameters is realistic and to which domain it belongs. The generator and discriminator are trained alternately based on the multi-scale analysis results and using a target loss function to obtain new generators and discriminators. Then, the process jumps to the step of determining latent vectors using the encoder in the generator of the VAE-GAN model based on the training set, until the initial airfoil parameters meet a preset accuracy condition, resulting in a trained VAE-GAN model. The trained VAE-GAN model is then validated and tested using the validation set and the test set to obtain the target VAE-GAN model. It is worth noting that the preset accuracy condition can be specifically limited according to actual circumstances.

[0072] Specifically, the CST parameterization method is used to convert different types of airfoil shapes into digital parameters, and a training set is determined based on these digital parameters. Based on the training set, the encoder in the generator of the VAE-GAN model is used to determine latent vectors. Based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils, the decoder in the generator is used to determine initial airfoil parameters. The discriminator of the VAE-GAN model is used to perform multi-scale analysis on the initial airfoil parameters, and the generator and discriminator are optimized based on the multi-scale analysis results to obtain new generators and new discriminators. Then, the process jumps to... The step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet a preset accuracy condition to obtain the target VAE-GAN model, includes: converting different types of airfoil shapes into digital parameters using the CST parameterization method; normalizing the digital parameters to obtain processed parameters; dividing the processed parameters using stratified sampling to obtain corresponding training, validation, and test sets; and, based on the training set and using the encoder in the generator of the VAE-GAN model, matching the processed parameters with different types of airfoils... The corresponding airfoil domain labels are mapped to a latent space distribution, and latent vectors are determined from the latent space distribution. The decoder in the generator is used to reconstruct the latent vectors and the airfoil domain labels to obtain initial airfoil parameters. The discriminator of the VAE-GAN model is used to perform multi-scale analysis of global, local, and detailed features on the initial airfoil parameters to obtain corresponding multi-scale analysis results. Based on the multi-scale analysis results and a first preset loss function, the generator is optimized to obtain a new generator. The first preset loss function includes a reconstruction loss function, a KL divergence loss function, and an adversarial loss function. The discriminator is optimized using the multi-scale analysis results and the second preset loss function to obtain a new discriminator. Then, the process jumps to the step of determining the latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet the preset accuracy conditions to obtain the trained VAE-GAN model. The second preset loss function includes the WGAN loss function, the gradient penalty loss function, and the cross-entropy loss function. The trained VAE-GAN model is validated and tested using the validation set and the test set respectively to obtain the target VAE-GAN model.

[0073] Step S13: If the condition is met and the instruction type is airfoil generation and aerodynamic analysis, then based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model, an airfoil cluster is generated. Geometric and aerodynamic characteristics are analyzed for each candidate airfoil in the airfoil cluster, and the first and second analysis results are displayed on the preset visualization interface.

[0074] In this embodiment, it is determined whether the semantic parsing confidence level meets a preset high confidence condition. If the preset high confidence condition is met, and the instruction type is airfoil generation and aerodynamic analysis, then the target latent vector is determined based on the standard normal distribution. Based on the target airfoil domain label corresponding to the airfoil type and the target latent vector, an airfoil cluster is generated using the target VAE-GAN model. In one specific implementation, Figure 3 This embodiment provides a schematic diagram of a rotor airfoil cluster, where x is the horizontal coordinate of a point on the airfoil, c is the chord length of the airfoil, and x / c is the ratio of the horizontal coordinate to the chord length; y is the vertical coordinate of a point on the airfoil, and y / c is the ratio of the vertical coordinate to the chord length, thus being dimensionless. Geometric characteristic analysis is performed on the shape data of each candidate airfoil in the airfoil cluster to obtain a first analysis result. Then, the NeuralFoil tool is used to perform aerodynamic characteristic analysis on each airfoil in the airfoil cluster to obtain a second analysis result. The first and second analysis results are then displayed in the form of a scatter plot on a preset visualization interface. In one specific embodiment, Figure 4 This is a scatter plot of the maximum thickness and maximum camber of a rotor airfoil cluster provided in this embodiment. Figure 5 This embodiment provides a scatter plot of the lift and drag coefficients of a rotor airfoil cluster, where both the horizontal and vertical axes are dimensionless. Further, if the preset high-confidence condition is not met, a corresponding correction input prompt is generated so that the user terminal can determine a new input language based on the correction input prompt, and then proceed to the step of converting the user terminal's input language into a semantic vector based on the BERT model. It is worth noting that the preset high-confidence condition can be limited according to actual circumstances.

[0075] Specifically, if the conditions are met and the instruction type is airfoil generation and aerodynamic analysis, then an airfoil cluster is generated based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model. Geometric and aerodynamic characteristic analyses are performed on each candidate airfoil in the airfoil cluster, and the first and second analysis results are displayed on a preset visualization interface. This includes: if the semantic parsing confidence level meets a preset high confidence condition and the instruction type is airfoil generation and aerodynamic analysis, then a target latent vector is determined based on a standard normal distribution; an airfoil cluster is generated based on the target airfoil domain label corresponding to the airfoil type and the target latent vector, using the target VAE-GAN model; geometric characteristic analysis is performed on the shape data of each candidate airfoil in the airfoil cluster to obtain a first analysis result; aerodynamic characteristic analysis is performed on each airfoil in the airfoil cluster using the NeuralFoil tool to obtain a second analysis result; and the first and second analysis results are displayed on a preset visualization interface in the form of a scatter plot.

[0076] Step S14: If the conditions are met and the instruction type is airfoil intelligent optimization design type, then the candidate airfoil is determined using the target VAE-GAN model, and the candidate airfoil is optimized using the optimization objective and constraints to obtain the optimized airfoil.

[0077] In this embodiment, if the semantic parsing confidence level meets a preset high confidence condition, and the instruction type is an airfoil intelligent optimization design type, then an optimization design objective is determined based on the optimization objective and constraints, and the candidate airfoil is optimized based on the optimization type corresponding to the optimization design objective to obtain an optimized airfoil. Specifically, if the condition is met and the instruction type is an airfoil intelligent optimization design type, then the candidate airfoil is determined using the target VAE-GAN model, and the candidate airfoil is optimized using the optimization objective and constraints to obtain an optimized airfoil, including: if the semantic parsing confidence level meets a preset high confidence condition, and the instruction type is an airfoil intelligent optimization design type, then an optimization design objective is determined based on the optimization objective and constraints; a target latent vector is determined based on a standard normal distribution; the target latent vector is decoded using the target VAE-GAN model to obtain target digital parameters; the target digital parameters are converted into airfoil coordinates, and the candidate airfoil is determined using the airfoil coordinates; the candidate airfoil is optimized based on the optimization type corresponding to the optimization design objective to obtain an optimized airfoil.

[0078] It is understood that the corresponding optimization scheme is determined based on the optimization type corresponding to the optimization design objective. In one specific implementation, Figure 6This embodiment provides a schematic diagram of an optimized airfoil based on a single-objective optimization design. If the optimization type corresponding to the optimization design objective is single-objective optimization, then the candidate airfoil is parameter-optimized based on the first and second analysis results corresponding to the candidate airfoil, and aerodynamic constraints (such as lift coefficient, drag coefficient, lift-to-drag ratio, etc.) and geometric constraints (such as thickness range, camber range) are added. A penalty mechanism is used to ensure that the optimization results meet the constraint conditions to obtain the optimized airfoil. If the optimization design objective is to optimize the lift-to-drag ratio, Figure 7 This embodiment provides a schematic diagram of the objective function variation curve for a single-objective optimization design. The objective function refers to the index to be optimized, i.e., the objective optimization function determined based on the optimization design objective. For example, if the user input is: minimize the airfoil drag, the objective function is the drag coefficient, and the optimization direction is minimization. Specifically, optimizing the candidate airfoil based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil includes: if the optimization type corresponding to the optimization design objective is single-objective optimization, then optimizing the parameters of the candidate airfoil based on the first analysis result and the second analysis result corresponding to the candidate airfoil to obtain the optimized airfoil.

[0079] In another specific implementation, if the optimization type corresponding to the optimization design objective is multi-objective optimization, then based on the optimization objective and constraints and the candidate airfoil, a Pareto optimal solution set is determined using a multi-objective optimization algorithm to determine the optimized airfoil. If the optimization design objective is to optimize the lift-to-drag ratio and maximum thickness, the corresponding Pareto optimal solution set is as follows: Figure 8 As shown, Figure 9 These are two optimal airfoils in the Pareto optimal solution set. Figure 9 (a) is an optimal airfoil in the Pareto optimal solution set. Figure 9 (b) is another optimal airfoil in the Pareto optimal solution set. Specifically, optimizing the candidate airfoil based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil includes: if the optimization type corresponding to the optimization design objective is multi-objective optimization, then determining the Pareto optimal solution set based on the optimization objective and constraints and the candidate airfoil using a multi-objective optimization algorithm, and determining the optimized airfoil using the Pareto optimal solution set.

[0080] As can be seen from the above, this application utilizes the BERT model to convert the natural language input from the user end into semantic vectors, and combines the weights of airfoil domain terms to output structured parameters, converting multiple types of airfoil shapes into numerical parameters to determine the training set. The generator of the VAE-GAN model is used to determine the initial airfoil parameters, and the discriminator of the VAE-GAN model is used to perform multi-scale analysis on the initial airfoil parameters to optimize the generator and discriminator, thereby obtaining the target VAE-GAN model. If the user end command type is airfoil generation and aerodynamic analysis, an airfoil cluster including candidate airfoils can be generated using the target VAE-GAN model and structured parameters. Then, geometric and aerodynamic characteristic analyses are performed on the candidate airfoils, and the analysis results are transmitted to the user end's visualization interface for interface interaction. If the command type is airfoil intelligent optimization design, the target VAE-GAN model is used to determine candidate airfoils, and the optimization objectives and constraints are used to optimize the candidate airfoils to obtain the optimized airfoil. In this way, users do not need to input professional and complex statements. The target VAE-GAN model can generate candidate airfoils that correspond to user needs. It can also be further optimized according to user needs to obtain high-precision optimized airfoils. This not only lowers the professional threshold for airfoil design, but also greatly improves the efficiency and quality of airfoil design.

[0081] As can be seen from the above embodiments, this application generates candidate airfoils corresponding to user needs based on the target VAE-GAN model, and optimizes them to obtain high-quality optimized airfoils. Therefore, the process of generating candidate airfoils corresponding to user needs based on the target VAE-GAN model and optimizing them to obtain high-quality optimized airfoils is described.

[0082] See Figure 10 As shown in the figure, this invention discloses a specific airfoil generation and intelligent optimization design method based on natural language interaction, including:

[0083] In this embodiment, the user's input language is first obtained. Based on the BERT model, the input language is converted into a semantic vector. Cosine similarity is used to determine the semantic matching degree between the semantic vector and airfoil domain terminology. A pre-defined airfoil domain lexicon is used to assign weights to different airfoil domain terminology to obtain domain term weights. The target matching degree between the input language and the airfoil domain terminology is determined based on the product of the semantic matching degree and the domain term weights. Structured parameters are then determined based on the target matching degree and the input language. The target VAE-GAN model used to generate the initial airfoil parameters includes a generator, a discriminator, and an output layer.

[0084] It is understandable that the semantic parsing confidence level is determined to meet the preset high confidence level condition. If the preset high confidence level condition is met and the instruction type is airfoil generation and aerodynamic analysis type, then the target potential vector is determined based on the standard normal distribution, and the airfoil cluster is generated based on the target airfoil domain label corresponding to the airfoil type and the target potential vector, and the target VAE-GAN model is used. Figure 11 This embodiment provides a schematic diagram of generating a rotor airfoil cluster using a target VAE-GAN model. Based on the user's input language, the corresponding airfoil cluster is generated using the target VAE-GAN model. The geometric characteristics of the shape data of each candidate airfoil in the airfoil cluster are analyzed to obtain a first analysis result. The aerodynamic characteristics of each airfoil in the airfoil cluster are analyzed using the NeuralFoil tool to obtain a second analysis result. The first analysis result and the second analysis result are displayed in the form of a scatter plot on a preset visualization interface.

[0085] Furthermore, if the semantic parsing confidence level meets a preset high confidence condition, and the instruction type is an airfoil intelligent optimization design type, then an optimization design objective is determined based on the optimization objective and constraints, and the candidate airfoil is optimized based on the optimization type corresponding to the optimization design objective to obtain an optimized airfoil. If the optimization type corresponding to the optimization design objective is a single-objective optimization, then the candidate airfoil is parameter optimized based on the first analysis result and the second analysis result corresponding to the candidate airfoil to obtain an optimized airfoil. Figure 12 This embodiment provides a schematic diagram of a single-objective optimization design, which determines the objective function corresponding to the optimized airfoil and the indicators to be optimized based on the input language and the instruction type.

[0086] In this embodiment, if the optimization type corresponding to the optimization design objective is multi-objective optimization, then the Pareto optimal solution set is determined based on the optimization objective and constraints and the candidate airfoil using a multi-objective optimization algorithm, and the optimized airfoil is determined using the Pareto optimal solution set. Figure 13 This embodiment provides a schematic diagram of a multi-objective optimization design. Based on the input language and the instruction type, a Pareto optimal solution set is determined. The Pareto optimal solution set includes two optimal airfoils, i.e., two optimized airfoils, from which the user terminal can determine the optimized airfoil.

[0087] As can be seen from the above, this application utilizes the generator of the VAE-GAN model and the input language to generate initial airfoil parameters. If the user-side command type is airfoil generation and aerodynamic analysis, an airfoil cluster including candidate airfoils can be generated using the target VAE-GAN model and structured parameters. Geometric and aerodynamic characteristic analyses are then performed on the candidate airfoils, and the analysis results are transmitted to the user-side visualization interface. If the command type is airfoil intelligent optimization design, the target VAE-GAN model is used to determine candidate airfoils, and the optimization objectives and constraints are used to optimize the candidate airfoils to obtain the optimized airfoil. In this way, candidate airfoils corresponding to user requirements can be generated directly using the target VAE-GAN model, and airfoil optimization can be performed according to the user-side optimization objectives to obtain a high-precision optimized airfoil, greatly improving the efficiency of airfoil design.

[0088] Accordingly, see Figure 14 As shown, this application also provides an airfoil generation and intelligent optimization design device based on natural language interaction, including:

[0089] The parameter determination module 11 is used to convert the user's input language into a semantic vector based on the BERT model, and determine the structured parameters based on the semantic vector and the weights corresponding to the airfoil domain terminology; the structured parameters include command type, airfoil type, operating condition parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence.

[0090] The target model determination module 12 is used to convert different types of airfoil shapes into digital parameters using the CST parameterization method, determine a training set based on the digital parameters, determine latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, determine initial airfoil parameters based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils and using the decoder in the generator, perform multi-scale analysis on the initial airfoil parameters using the discriminator of the VAE-GAN model, and optimize the generator and the discriminator based on the multi-scale analysis results to obtain a new generator and a new discriminator. Then, it jumps to the step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model until the initial airfoil parameters meet the preset accuracy condition to obtain the target VAE-GAN model, and determines whether the semantic parsing confidence meets the preset high confidence condition.

[0091] The analysis result display module 13 is used to generate an airfoil cluster based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model if the condition is met and the instruction type is airfoil generation and aerodynamic analysis, perform geometric characteristic analysis and aerodynamic characteristic analysis on each candidate airfoil in the airfoil cluster respectively, and display the first analysis result and the second analysis result on the preset visualization interface.

[0092] The airfoil optimization module 14 is used to determine candidate airfoils using the target VAE-GAN model if the conditions are met and the instruction type is airfoil intelligent optimization design type, and to optimize the candidate airfoils using the optimization objective and constraints to obtain the optimized airfoil.

[0093] In some specific embodiments, the parameter determination module 11 may specifically include:

[0094] The matching degree determination unit is used to convert the user's input language into a semantic vector based on the BERT model, and use cosine similarity to determine the semantic matching degree between the semantic vector and the airfoil domain professional terminology;

[0095] The weight allocation unit is used to assign weights to different airfoil domain terms using a pre-defined airfoil domain terminology library in order to obtain domain term weights.

[0096] The parameter determination unit is used to determine the target matching degree between the input language and the airfoil domain terminology based on the semantic matching degree and the domain term weight, and to determine the structured parameters based on the target matching degree and the input language.

[0097] In some specific embodiments, the target model determination module 12 may specifically include:

[0098] The parameter processing unit is used to convert different types of airfoil shapes into digital parameters using the CST parameterization method, and to normalize the digital parameters to obtain the processed parameters.

[0099] The parameter partitioning unit is used to partition the processed parameters using stratified sampling to obtain corresponding training sets, validation sets, and test sets.

[0100] The vector determination unit is used to map the processed parameters and airfoil domain labels corresponding to different types of airfoils to a latent spatial distribution based on the training set and using the encoder in the generator of the VAE-GAN model, and determine the latent vector from the latent spatial distribution.

[0101] An initial parameter determination unit is used to reconstruct the latent vector and the airfoil domain label using the decoder in the generator to obtain initial airfoil parameters;

[0102] The feature analysis unit is used to perform multi-scale analysis of global, local and detailed features of the initial airfoil parameters using the discriminator of the VAE-GAN model, so as to obtain the corresponding multi-scale analysis results.

[0103] The performance optimization unit is used to optimize the generator's performance based on the multi-scale analysis results and a first preset loss function to obtain a new generator; the first preset loss function includes a reconstruction loss function, a KL divergence loss function, and an adversarial loss function.

[0104] The post-training model determination unit is used to optimize the discriminator's performance using the multi-scale analysis results and the second preset loss function to obtain a new discriminator. Then, it jumps to the step of determining the latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model until the initial airfoil parameters meet the preset accuracy conditions to obtain the post-trained VAE-GAN model. The second preset loss function includes the WGAN loss function, the gradient penalty loss function, and the cross-entropy loss function.

[0105] The model testing unit is used to validate and test the trained VAE-GAN model using the validation set and the test set, respectively, to obtain the target VAE-GAN model.

[0106] In some specific embodiments, the analysis result display module 13 may specifically include:

[0107] The geometric analysis unit is used to determine the target latent vector based on the standard normal distribution if the semantic parsing confidence meets the preset high confidence condition and the instruction type is airfoil generation and aerodynamic analysis type; generate an airfoil cluster based on the target airfoil domain label corresponding to the airfoil type and the target latent vector and the target VAE-GAN model; and perform geometric characteristic analysis on the shape data of each candidate airfoil in the airfoil cluster to obtain the first analysis result.

[0108] The aerodynamic analysis unit is used to perform aerodynamic characteristic analysis on each airfoil in the airfoil cluster using the NeuralFoil tool to obtain a second analysis result, and to display the first analysis result and the second analysis result in the form of a scatter plot on a preset visualization interface.

[0109] In some specific embodiments, the airfoil optimization module 14 may specifically include:

[0110] The target vector determination unit is used to determine the optimization design target based on the optimization target and constraints, and to determine the target potential vector based on the standard normal distribution, if the semantic parsing confidence satisfies the preset high confidence condition and the instruction type is the airfoil intelligent optimization design type.

[0111] The parameter conversion unit is used to decode the target latent vector using the target VAE-GAN model to obtain target digital parameters, convert the target digital parameters into airfoil coordinates, and use the airfoil coordinates to determine candidate airfoils.

[0112] An airfoil optimization unit is used to optimize the candidate airfoil based on the optimization type corresponding to the optimization design objective, so as to obtain the optimized airfoil.

[0113] In some specific embodiments, the airfoil optimization module 14 may specifically include:

[0114] The parameter optimization unit is used to optimize the parameters of the candidate airfoil based on the first analysis result and the second analysis result corresponding to the candidate airfoil if the optimization type corresponding to the optimization design objective is single-objective optimization, so as to obtain the optimized airfoil.

[0115] In some specific embodiments, the airfoil optimization module 14 may specifically include:

[0116] The set determination unit is used to determine the Pareto optimal solution set based on the optimization objective and constraints and the candidate airfoil and using a multi-objective optimization algorithm if the optimization type corresponding to the optimization design objective is multi-objective optimization, and to determine the optimized airfoil using the Pareto optimal solution set.

[0117] Furthermore, embodiments of this application also disclose an electronic device, Figure 15 This is a structural diagram of an electronic device 20 according to an exemplary embodiment. The content of the diagram should not be construed as limiting the scope of this application. The electronic device 20 may specifically include: at least one processor 21, at least one memory 22, a power supply 23, a communication interface 24, an input / output interface 25, and a communication bus 26. The memory 22 stores a computer program, which is loaded and executed by the processor 21 to implement the relevant steps in the airfoil generation and intelligent optimization design method based on natural language interaction disclosed in any of the foregoing embodiments. Furthermore, the electronic device 20 in this embodiment may specifically be an electronic computer.

[0118] In this embodiment, the power supply 23 is used to provide operating voltage for each hardware device on the electronic device 20; the communication interface 24 can create a data transmission channel between the electronic device 20 and external devices, and the communication protocol it follows can be any communication protocol applicable to the technical solution of this application, and is not specifically limited here; the input / output interface 25 is used to acquire external input data or output data to the outside world, and its specific interface type can be selected according to specific application needs, and is not specifically limited here.

[0119] In addition, the memory 22, as a carrier for resource storage, can be a read-only memory, random access memory, disk or optical disk, etc. The resources stored thereon can include operating system 221, computer program 222, etc., and the storage method can be temporary storage or permanent storage.

[0120] The operating system 221 is used to manage and control the various hardware devices on the electronic device 20 and the computer program 222, which may be Windows Server, Netware, Unix, Linux, etc. In addition to including a computer program capable of performing the natural language interaction-based airfoil generation and intelligent optimization design method executed by the electronic device 20 as disclosed in any of the foregoing embodiments, the computer program 222 may further include computer programs capable of performing other specific tasks.

[0121] Furthermore, this application also discloses a computer-readable storage medium for storing a computer program; wherein, when the computer program is executed by a processor, it implements the aforementioned airfoil generation and intelligent optimization design method based on natural language interaction. Specific steps of this method can be found in the corresponding content disclosed in the foregoing embodiments, and will not be repeated here.

[0122] The various embodiments in this specification are described in a progressive manner, with each embodiment focusing on its differences from other embodiments. Similar or identical parts between embodiments can be referred to interchangeably. For the apparatus disclosed in the embodiments, since it corresponds to the method disclosed in the embodiments, the description is relatively simple; relevant parts can be referred to in the method section.

[0123] Those skilled in the art will further recognize that the units and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of both. To clearly illustrate the interchangeability of hardware and software, the components and steps of the various examples have been generally described in terms of functionality in the foregoing description. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0124] The steps of the methods or algorithms described in conjunction with the embodiments disclosed herein can be implemented directly by hardware, a software module executed by a processor, or a combination of both. The software module can be located in random access memory (RAM), main memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art.

[0125] Finally, it should be noted that in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0126] The technical solutions provided in this application have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the ideas of this application. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for airfoil generation and intelligent optimization design based on natural language interaction, characterized in that, include: Based on the BERT model, the user's input language is converted into a semantic vector, and the structured parameters are determined based on the weights corresponding to the semantic vectors and the professional terms in the airfoil field. The structured parameters include command type, airfoil type, operating condition parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence level. The CST parameterization method is used to convert different types of airfoil shapes into digital parameters, and a training set is determined based on the digital parameters. Based on the training set, the encoder in the generator of the VAE-GAN model is used to determine latent vectors. Based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils, the decoder in the generator is used to determine the initial airfoil parameters. The discriminator of the VAE-GAN model is used to perform multi-scale analysis on the initial airfoil parameters, and the generator and the discriminator are optimized based on the multi-scale analysis results to obtain new generators and new discriminators. Then, the process jumps to the step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet the preset accuracy condition to obtain the target VAE-GAN model. Finally, it is determined whether the semantic parsing confidence meets the preset high confidence condition. If the conditions are met, and the instruction type is airfoil generation and aerodynamic analysis, then an airfoil cluster is generated based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model. Geometric and aerodynamic characteristics are analyzed for each candidate airfoil in the airfoil cluster, and the first and second analysis results are displayed on the preset visualization interface. If the conditions are met, and the instruction type is airfoil intelligent optimization design type, then the candidate airfoil is determined using the target VAE-GAN model, and the candidate airfoil is optimized using the optimization objective and constraints to obtain the optimized airfoil.

2. The airfoil generation and intelligent optimization design method based on natural language interaction according to claim 1, characterized in that, The BERT-based model converts the user's input language into semantic vectors, and determines the structured parameters based on the weights corresponding to the semantic vectors and the airfoil domain terminology. The structured parameters include command type, airfoil type, operating parameters, optimization objectives and constraints, number of design results, and semantic analysis confidence level, including: Based on the BERT model, the user's input language is converted into a semantic vector, and the cosine similarity is used to determine the semantic matching degree between the semantic vector and the professional terminology in the airfoil field; By using a pre-defined airfoil domain thesaurus, different airfoil domain professional terms are weighted to obtain domain term weights; The target matching degree between the input language and the airfoil domain terminology is determined based on the semantic matching degree and the domain term weight, and the structured parameters are determined based on the target matching degree and the input language.

3. The airfoil generation and intelligent optimization design method based on natural language interaction according to claim 1, characterized in that, The process involves converting different types of airfoil shapes into digital parameters using the CST parameterization method, determining a training set based on these digital parameters, determining latent vectors using the encoder in the generator of the VAE-GAN model based on the training set, determining initial airfoil parameters based on the latent vectors and airfoil domain labels corresponding to different types of airfoils using the decoder in the generator, performing multi-scale analysis on the initial airfoil parameters using the discriminator of the VAE-GAN model, and optimizing the generator and discriminator based on the multi-scale analysis results to obtain new generators and discriminators. Then, the process returns to the step of determining latent vectors based on the training set using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet a preset accuracy condition to obtain the target VAE-GAN model. This process includes: The CST parameterization method is used to convert different types of airfoil shapes into digital parameters, and the digital parameters are normalized to obtain the processed parameters. The processed parameters are divided using stratified sampling to obtain corresponding training, validation, and test sets. Based on the training set, the encoder in the generator of the VAE-GAN model is used to map the processed parameters and the airfoil domain labels corresponding to different types of airfoils to the latent space distribution, and the latent vector is determined from the latent space distribution. The decoder in the generator is used to reconstruct the latent vector and the airfoil domain label to obtain the initial airfoil parameters; The discriminator of the VAE-GAN model is used to perform multi-scale analysis of the global, local and detailed features of the initial airfoil parameters to obtain the corresponding multi-scale analysis results; The generator is optimized based on the multi-scale analysis results and the first preset loss function to obtain a new generator; the first preset loss function includes a reconstruction loss function, a KL divergence loss function, and an adversarial loss function. The discriminator is optimized using the multi-scale analysis results and the second preset loss function to obtain a new discriminator. Then, the process jumps to the step of determining the latent vectors using the encoder in the generator of the VAE-GAN model based on the training set, until the initial airfoil parameters meet the preset accuracy condition to obtain the trained VAE-GAN model. The second preset loss function includes the WGAN loss function, the gradient penalty loss function, and the cross-entropy loss function. The trained VAE-GAN model is validated and tested using the validation set and the test set, respectively, to obtain the target VAE-GAN model.

4. The airfoil generation and intelligent optimization design method based on natural language interaction according to any one of claims 1 to 3, characterized in that, If the above conditions are met, and the instruction type is airfoil generation and aerodynamic analysis, then an airfoil cluster is generated based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model. Geometric and aerodynamic characteristic analyses are performed on each candidate airfoil in the airfoil cluster, and the first and second analysis results are displayed on a preset visualization interface, including: If the semantic parsing confidence meets the preset high confidence condition, and the instruction type is airfoil generation and aerodynamic analysis, then the target potential vector is determined based on the standard normal distribution. Based on the target airfoil domain label corresponding to the airfoil type and the target potential vector, an airfoil cluster is generated using the target VAE-GAN model. The geometric characteristics of the shape data of each candidate airfoil in the airfoil cluster are analyzed to obtain the first analysis result. The aerodynamic characteristics of each airfoil in the airfoil cluster are analyzed using the NeuralFoil tool to obtain a second analysis result. The first analysis result and the second analysis result are then displayed in a scatter plot on a preset visualization interface.

5. The airfoil generation and intelligent optimization design method based on natural language interaction according to claim 4, characterized in that, If the conditions are met, and the instruction type is an airfoil intelligent optimization design type, then the candidate airfoil is determined using the target VAE-GAN model, and the candidate airfoil is optimized using the optimization objective and constraints to obtain the optimized airfoil, including: If the semantic parsing confidence level meets the preset high confidence level condition, and the instruction type is the airfoil intelligent optimization design type, then the optimization design objective is determined based on the optimization objective and constraint conditions, and the target potential vector is determined based on the standard normal distribution. The target latent vector is decoded using the target VAE-GAN model to obtain the target digital parameters, the target digital parameters are converted into airfoil coordinates, and the candidate airfoil is determined using the airfoil coordinates; The candidate airfoil is optimized based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil.

6. The airfoil generation and intelligent optimization design method based on natural language interaction according to claim 5, characterized in that, The optimization of the candidate airfoil based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil includes: If the optimization type corresponding to the optimization design objective is single-objective optimization, then the candidate airfoil is optimized based on the first analysis result and the second analysis result corresponding to the candidate airfoil to obtain the optimized airfoil.

7. The airfoil generation and intelligent optimization design method based on natural language interaction according to claim 5, characterized in that, The optimization of the candidate airfoil based on the optimization type corresponding to the optimization design objective to obtain the optimized airfoil includes: If the optimization type corresponding to the optimization design objective is multi-objective optimization, then based on the optimization objective and constraints and the candidate airfoil, the Pareto optimal solution set is determined using a multi-objective optimization algorithm, and the optimized airfoil is determined using the Pareto optimal solution set.

8. A device for airfoil generation and intelligent optimization design based on natural language interaction, characterized in that, include: The parameter determination module is used to convert the user's input language into semantic vectors based on the BERT model, and determine structured parameters based on the weights corresponding to the semantic vectors and airfoil domain terminology. The structured parameters include command type, airfoil type, operating parameters, optimization objectives and constraints, number of design results, and semantic parsing confidence. The target model determination module is used to convert different types of airfoil shapes into digital parameters using the CST parameterization method, determine a training set based on the digital parameters, determine latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, determine initial airfoil parameters based on the latent vectors and the airfoil domain labels corresponding to different types of airfoils and using the decoder in the generator, perform multi-scale analysis on the initial airfoil parameters using the discriminator of the VAE-GAN model, and optimize the generator and the discriminator based on the multi-scale analysis results to obtain a new generator and a new discriminator. Then, it jumps to the step of determining latent vectors based on the training set and using the encoder in the generator of the VAE-GAN model, until the initial airfoil parameters meet the preset accuracy condition to obtain the target VAE-GAN model, and determines whether the semantic parsing confidence meets the preset high confidence condition. The analysis results display module is used to generate an airfoil cluster based on the target airfoil domain label corresponding to the airfoil type and using the target VAE-GAN model if the condition is met and the instruction type is airfoil generation and aerodynamic analysis, perform geometric characteristic analysis and aerodynamic characteristic analysis on each candidate airfoil in the airfoil cluster respectively, and display the first analysis result and the second analysis result on the preset visualization interface. The airfoil optimization module is used to determine candidate airfoils using the target VAE-GAN model if the conditions are met and the instruction type is airfoil intelligent optimization design type, and to optimize the candidate airfoils using the optimization objective and constraints to obtain the optimized airfoil.

9. An electronic device, characterized in that, include: Memory, used to store computer programs; A processor for executing the computer program to implement the airfoil generation and intelligent optimization design method based on natural language interaction as described in any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that, Used to store a computer program, wherein the computer program, when executed by a processor, implements the airfoil generation and intelligent optimization design method based on natural language interaction as described in any one of claims 1 to 7.

Images